Google's War on AI Slop: The Helpful Content Update Timeline 2022-2026 (and What It Means for SEO)

Google's War on AI Slop: The Helpful Content Update Timeline 2022-2026 (and What It Means for SEO)

On 1 May 2024, the air-purifier review site HouseFresh published a post titled "HouseFresh has virtually disappeared from Google Search results". Its own numbers: since October 2023, visits from Google Search had gone from 4,000 a day to 200, and the March 2024 core update had cost the site 91% of its search traffic. A month earlier, on 3 April 2024, the founder of the retro-gaming site Retro Dodo wrote that Google had cut its organic traffic and revenue by 85% since September 2023. Both losses fall inside the window this article covers: Google's helpful content updates, and the core updates that absorbed them. This is the chronicle of how a narrow August 2022 ranking signal turned into the organizing logic of the modern search index.

The HCU launched in August 2022 as a narrow signal aimed at content written for search engines first. Three months later, ChatGPT shipped. The cost of drafting an article collapsed, the open web filled with machine text, and the HCU stopped being a footnote: it became the central nervous system of how Google ranks everything.

Google's Search Status Dashboard, its own log of ranking updates, lists at least one named core, helpful content, reviews or spam update in every quarter from August 2022 to September 2026. Three carried the helpful content name (August 2022, December 2022, September 2023). From March 2024 the helpful content system became part of the core ranking systems, so every core update since carries it. What site owners documented themselves, HouseFresh and Retro Dodo among them, was a reorganization of the small-publisher and affiliate-review web.

This article is a long, dense reference. It covers what each update did, who got hit, what survived, and what site owners should be doing in 2026 to stay on the right side of the next core update. Dates come from Google's own announcements and its Search Status Dashboard, the traffic figures from the sites that published them about themselves, and every source is linked in section 16. Where the algorithm is opaque we say so, and where a pattern is our reading rather than a measurement we say that too.

TL;DR: Seven Things Every Site Owner Needs to Know in 2026

• The Helpful Content Update has stopped being a separate event. Since the March 2024 Core Update, helpfulness signals are baked into Core Updates. There is no longer a discrete "HCU rollout" you can wait out, every Core Update is now also a helpfulness update.

• Affiliate review sites and "best of X" listicles took the heaviest hits, by the account of the sites themselves. HouseFresh reported a 91% loss of search traffic after the March 2024 core update, and Retro Dodo an 85% drop in organic traffic and revenue since September 2023. Google has published no figures on who lost what.

• Reddit rose while the small review sites fell, in the account of the sites that lost. Retro Dodo's founder called Reddit "the single biggest winner in recent algorithm changes" (3 April 2024), and in HouseFresh's walkthrough of the top ten results for "best air purifier for pets", Amazon and Reddit sat above BuzzFeed's list at position six (19 February 2024). Google has published no figures on who gained.

• AI Overviews changed the click economics. Since the May 2024 launch and 2025 expansion of AI Overviews, informational queries that previously sent clicks to publishers now resolve inside the search results page. In Pew Research Center's browsing data for 900 US adults in March 2025, people clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when none did.

• EEAT is now visual, not just textual. Author bios with photos, schema.org Person markup, demonstrated experience (the author actually used the product, the writer actually visited the city) have shifted from "nice to have" to load-bearing ranking ingredients. Google's own self-assessment for creators asks "Is it self-evident to your visitors who authored your content?" and "Do pages carry a byline, where one might be expected?"

• The slop detection signals appear to combine textual and structural patterns. Banned-phrase frequency ("delve", "tapestry", "in today's fast-paced world"), n-gram repetition, missing internal links to topical clusters, recycled meta descriptions, and absence of original research are the patterns SEOs point to in demoted sites. Google has not published how it classifies unhelpful content, so read this list as an interpretation of the cases, not a measured set of signals. The same fingerprints flag generated text everywhere: see our breakdown of the 40 words that mark copy as AI-generated.

• Bing, Perplexity, You.com, Phind and Komo are surfaces of their own. None of them documents its ranking in detail, so this article does not rate what each one rewards. Check what each one actually sends you before moving effort.

If you only read one section of this piece, read section 4 (what got demoted) and section 12 (the 2026 playbook). The rest is context.

1. What the Helpful Content Update Is

The Helpful Content Update was announced by Google on August 18, 2022 and began rolling out August 25. The official Google Search Central post described it as "a new site-wide signal" and warned that "content created primarily for search engine traffic is strongly correlated with content that searchers find unsatisfying". The phrasing was unusual: most ranking signals operate on a page level. HCU was explicitly sitewide. If a portion of your site looked like it was written for crawlers, the entire domain could be demoted, including pages that would otherwise rank.

The August 2022 launch applied, in Google's words, to "English searches globally to begin with", with other languages planned later.

ChatGPT launched November 30, 2022, and the marginal cost of drafting an article on any topic fell close to nothing. The economics that the HCU was trying to police suddenly inverted. From late 2022 onward, the HCU's importance grew because the supply of content "made for search engines" exploded.

Google's framing evolved. The helpful content system added language about "demonstrating first-hand experience" and "providing original information, reporting, research, or analysis". This is the EEAT framework (Experience, Expertise, Authoritativeness, Trust) extended to apply at the site level, not just the page level.

By March 2024, the Helpful Content System was folded into Google's Core Ranking Systems. There is no longer a standalone HCU. But the signal lives on inside Core Updates. Google's March 2024 post put it this way: "There's no longer one signal or system used to do this".

2. The 2022-2026 Timeline

Master Timeline Table

Every ranking update below is listed on Google's Search Status Dashboard, with the start date it gives (read on 18 October 2026). The two "Ranking is experiencing an ongoing issue" incidents on that page are left out, because they are outages, not updates.

The announcements around the updates

2022. Google announces the helpful content update on 18 August. ChatGPT launches on 30 November. On 15 December, Google adds Experience to E-A-T in its search quality rater guidelines, which becomes E-E-A-T.

2023. On 8 February, Google's guidance on AI-generated content is titled "Rewarding high-quality content, however it is produced". This is the moment Google publicly chose not to ban AI content. The September helpful content update is the one Retro Dodo dates its losses from.

2024. On 22 February, Google announces an expanded partnership with Reddit. The March core update folds helpfulness into the core ranking systems, and the spam update that starts the same day adds a scaled content abuse policy aimed at pages produced in bulk, whatever the mix of automation and human work behind them. On 14 May, AI Overviews start rolling out to everyone in the US.

2025. At I/O in May, Google reports AI Overviews at 1.5 billion monthly users in 200 countries and territories. In July, Pew Research Center publishes the click measurements quoted in section 7.

2026. Core updates in March and May, spam updates in March, June, August and September.

3. What the Two Published Cases Show

Two sites published their own traffic figures. They are the only figures on losses in this article, and the table quotes them as the sites gave them:

Two cases are not a sample. They show what happened to two sites that chose to publish their numbers; they say nothing about how many sites lost traffic, a figure Google has never published.

4. What Got Demoted

We are careful here to cite only patterns that have been publicly reported by site owners or covered by named publications. Specific case studies follow.

HouseFresh

HouseFresh is an air purifier review site. Gisele Navarro, who runs its day-to-day operations, wrote both posts quoted here. The first, "How Google is killing independent sites like ours" (19 February 2024), argues that big media publishers such as Better Homes & Gardens and Real Simple were ranking at the top for "best of" pages that, in HouseFresh's reading, recommend products without first-hand testing. The second (1 May 2024) gives the numbers: Google Search visits down from 4,000 a day to 200 since October 2023, and a 91% loss of search traffic after the March 2024 core update. The site describes its method as buying the products with its own money and spending weeks testing them.

Retro Dodo

Retro Dodo is a retro-gaming review and news site. On 3 April 2024 its founder, Brandon Saltalamacchia, wrote that since September 2023 Google had reduced the site's organic traffic and revenue by 85%, leaving the business "on the edge of going under". He argued that its niche expertise was the kind of content Google says it wants to reward, and named Reddit as "the single biggest winner in recent algorithm changes".

General patterns of demotion

Google's self-assessment for creators warns against "producing lots of content on many different topics", "using extensive automation to produce content on many topics" and "mainly summarizing what others have to say without adding much value". The patterns below are our reading of what that looks like in practice, not a list Google publishes (most of them are also the textual tells that flag a page as machine-written: see the 23 signs of AI-generated code and copy):

• Generic "best of" listicles with no demonstrated testing or first-hand use
• AI-generated reviews with banned phrases like "delve", "tapestry", "in today's fast-paced world"
• SaaS comparison sites that exist primarily as affiliate funnels with minimal original information
• Affiliate product round-ups with thin descriptions and identical structures across articles
• Content farms publishing at volume on unrelated topics
• Programmatic SEO sites generating template-based pages at scale (city + service combinations, etc.)
• Recipe blogs with excessive intro filler and AI-generated headnotes
• News aggregators that summarize without original reporting
• AI-summarized YouTube transcripts repackaged as articles

Demotion vs Penalty

Important distinction the SEO industry has emphasized: HCU is a demotion signal, not a penalty in the manual-action sense. There is no notification in Search Console. There is no removal from the index. The site simply loses ranking. Recovery requires rebuilding helpfulness signals across the entire site, not fixing individual pages.

Table: Demoted patterns vs preserved patterns

5. What Survived (or Rose)

The other side of the demotion story is the rise of certain content categories.

Reddit

Reddit is the site the losing publishers name first. Retro Dodo's founder called it "the biggest change I have ever seen in the search results over the last 5 years": "you will likely now see Reddit results upon searching for something" (3 April 2024). In HouseFresh's walkthrough of the top ten results for "best air purifier for pets", a query it says it had given up on, Amazon and Reddit sat above BuzzFeed's list at position six (19 February 2024). Google has not said why Reddit gained, and it publishes no figures per site. Two documented facts sit next to the timing:

• Google's expanded partnership with Reddit, announced on 22 February 2024, which gives Google access to Reddit's Data API
• Google's self-assessment for creators asks whether content shows "first-hand expertise and a depth of knowledge (for example, expertise that comes from having actually used a product or service, or visiting a place)". A thread of people describing their own use answers that question by its format; that link is our reading, not something Google has said

Forums and community sites

Google's VP of Search, Elizabeth Reid, wrote on 30 May 2024 that "forums are often a great source of authentic, first-hand information, but in some cases can lead to less-than-helpful advice". That is as far as Google has gone in public: it has published no figures on how forums fared in the updates.

Quora

The one dated sighting in our sources is HouseFresh's. On 1 May 2024, its article for "best budget air purifiers", which had ranked second since May 2023, was now "buried deep beneath sponsored posts, Quora advice from 2016, best-of lists from big media sites, and no less than 64 Google Shopping product listings". We found no wider measurement of how Quora ranks.

Personal blogs with author authority

Blogs with named authors who have professional expertise (a doctor writing about medicine, a software engineer writing about engineering) gained ground. The combination of personal voice, named author with real credentials, and topical depth is what Google's self-assessment asks for when it asks who wrote the content and whether they know the topic.

Brand-name commerce

On commercial queries, the space HouseFresh lost went to big media "best of" lists, Amazon and Google's own shopping results: Amazon ranked in the top five for "best air purifier for pets" (19 February 2024), and 64 Google Shopping listings sat above its "best budget air purifiers" article (1 May 2024). Those are two queries seen by one site, not a measurement of the category.

6. The EEAT Framework in 2026

EEAT (Experience, Expertise, Authoritativeness, Trust) was introduced in Google's Search Quality Rater Guidelines in 2014 as E-A-T (without Experience). The "Experience" component was added on 15 December 2022, two weeks after ChatGPT launched. The timing was not subtle.

Experience

Has the author actually used the product, visited the place, or done the thing? This is the EEAT addition that maps most directly to AI content. AI does not have first-hand experience. A review of a vacuum cleaner written by AI does not include the smell of the room after testing. A review by a person who actually used the vacuum can include that.

In 2026, "Experience" is operationalized through:

• First-person language about the actual use ("I tested this for three weeks", not "users have reported")
• Original photography of the product or place
• Specific details that AI does not generate by default (a sticker on the box, a quirk of the user interface, a quote from a barista)
• Author bio specifying relevant lived experience

Expertise

Does the author have credentials in the subject? Expertise is about formal qualifications: a registered nurse writing about medication, a CFA writing about investing. For consumer topics, expertise can be demonstrated through years of hobby experience, professional roles, or published work. Expertise is most important for YMYL (Your Money or Your Life) queries: health, finance, legal, safety.

Authoritativeness

Is the site or author recognized as an authority in the topic? Authoritativeness is about external signals: links from other authoritative sites, mentions in named publications, citations in Wikipedia, presence in subject-matter directories.

Trust

This is the most important of the four, per Google's own framing. Trust covers site security (HTTPS), accurate contact information, transparent ownership, citation of sources, correction policies, and an overall "is this site trustworthy?" assessment.

EEAT scoring per article type

Different article types weight EEAT differently. The following is an interpretation of public Quality Rater Guidelines and SEO industry analysis.

The author bio and schema.org Person markup

By 2026, having a real author bio with photo, credentials, and links to external presence (LinkedIn, scholar profiles, professional sites) has shifted from "best practice" to load-bearing. Schema.org Person markup that links the author to the article via JSON-LD is what helps Google verify the byline.

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "How to Choose an Air Purifier for Allergies",
  "author": {
    "@type": "Person",
    "name": "Gisele Navarro",
    "url": "https://example.com/author/gisele-navarro",
    "image": "https://example.com/authors/gisele-navarro.jpg",
    "sameAs": [
      "https://twitter.com/gisele_example",
      "https://www.linkedin.com/in/gisele-example/"
    ],
    "jobTitle": "Editor-in-Chief",
    "knowsAbout": ["Air purifiers", "Indoor air quality"]
  },
  "publisher": {
    "@type": "Organization",
    "name": "Example Reviews",
    "logo": {
      "@type": "ImageObject",
      "url": "https://example.com/logo.png"
    }
  },
  "datePublished": "2026-04-28",
  "dateModified": "2026-04-28"
}
</script>

The example above uses a hypothetical Gisele Navarro byline. Google does not require this exact structure, but Schema.org Article + Person + Organization is the common pattern in 2026 for sites attempting to maximize EEAT signal.

7. AI Overviews and the Click Decline

AI Overviews began rolling out to everyone in the US on 14 May 2024, after a Search Generative Experience (SGE) preview in Search Labs in 2023. Google said then that it expected to bring them to "over a billion people by the end of the year"; at I/O in May 2025 it reported 1.5 billion monthly users in 200 countries and territories. The product places an AI-generated summary at the top of the search results page for many informational queries, with citation links to sources.

What changed for publishers

The mechanic that drove the click economy of the open web for 25 years was: user types query, scans results, clicks a link, lands on a publisher, publisher monetizes via ads or affiliate. AI Overviews insert a step that often makes the click unnecessary. The user reads the summary and may not click at all.

The exact magnitude is disputed, and it varies by topic and by who measures. The clearest independent measurement so far comes from Pew Research Center, which analysed the browsing of 900 US adults during March 2025 (published 22 July 2025, by Athena Chapekis and Anna Lieb):

On pages with a summary, a click on a link inside the summary itself happened on 1% of visits. Our read: there is a real click decline on queries that show a summary.

Brand impressions vs traffic

A pattern that has emerged: AI Overviews cite sources, so a site that gets cited gains brand impressions even when it loses clicks. Some publishers have shifted strategy to optimize for being cited (the source link in an Overview) rather than being clicked. Whether this is monetizable is an open question. Brand impression without a visit does not generate ad revenue or affiliate clicks, but it may build long-term brand recognition.

Zero-click search

The "zero-click search" concept predates AI Overviews. Featured snippets, knowledge panels, and direct answers were already absorbing clicks before 2024. AI Overviews accelerated the pattern by extending zero-click coverage to a wider range of queries. The percentage of searches that result in no click to an external site has been a topic of public debate (with figures from various sources, ranging from "most informational queries are now zero-click" to more conservative estimates). We do not cite specific decimal figures because the methodology disputes are unresolved.

8. The Slop Site Demotion Patterns

Across publicly shared HCU case studies, a pattern recipe emerges. We list the patterns observable in demoted sites: these are not Google's stated criteria; they are patterns that correlate with demotion.

Textual patterns

• Banned-phrase frequency: words like "delve", "tapestry", "leverage", "robust", "seamless", "harness", "revolutionize", "empower", "unleash", and phrases like "navigate the landscape" or "in today's fast-paced world" appear at unusually high rates in AI-generated content.
• Generic intros: many AI articles start with a paragraph about "in the ever-evolving digital landscape" or similar boilerplate that adds no information.
• Symmetric structures: AI-generated articles often have identical section counts, similar section lengths, and parallel construction across articles in a way that hand-written content does not.
• N-gram repetition: phrases repeat across articles on the same site at rates higher than an editorial style guide would explain.
• Recycled meta descriptions: identical or near-identical meta descriptions across pages.
• Identical conclusions: every article ends with "in conclusion" and a paragraph about how "by following these tips" the reader can succeed.

Structural patterns

• No author bio or a generic "Editorial Team" byline.
• No author photo or a stock photo of a generic person.
• Stock images for every article rather than original images.
• No internal links beyond a generic top navigation.
• No external citations to authoritative sources.
• No update dates or update dates that change without content actually changing.
• No comments or community engagement.
• No "About" page or an "About" page that reads like marketing copy.

Distribution patterns

• No newsletter or a perfunctory newsletter signup.
• No social presence for the named authors.
• No external mentions in named publications.
• No backlinks from non-spam sources.

No single item here is conclusive. A real site can have a stock photo or a thin meta description. But stack eight of them on one domain and the site reads as content-farm-shaped to a human rater, and the classifier is trained on what human raters flag.

9. The Detection Signals Google Likely Uses

Google has not published the exact algorithm. The following is informed speculation based on published research, patent filings, and observed behavior.

Likely textual signals

• Perplexity scoring: AI-generated text often has lower perplexity than hand-written text in characteristic ways. "Perplexity" measures how predictable the next word is given the prior context. AI text optimizes toward the likely next word; human text more frequently surprises. Concretely: a human reviewer writes "the filter rattled like a coin in a dryer" because they heard it; a model defaults to "the filter produced an audible noise during operation". The first is high-perplexity and specific; the second is low-perplexity and generic.
• N-gram diversity: human writing uses a wider vocabulary across articles. AI writing on a given topic clusters around the same phrasings, so the same 4- and 5-word strings recur across pages at rates a style guide cannot explain.
• Burstiness: human writing varies sentence length sharply: a three-word sentence next to a forty-word one. AI writing tends toward sentences of similar length.
• Originality scoring: how much of an article is paraphrasable from existing web content versus genuinely new.

Likely structural signals

• Author entity verification: does the byline correspond to a real person with verifiable presence?
• Domain age and history: long-running sites with consistent topical focus tend to outperform fresh sites.
• Topical authority graph: does the site cluster around a coherent topic with internal links and external citations?
• Engagement signals: dwell time, repeat visitors, branded search queries.

Likely cross-signal aggregation

Google's recent ranking work, including the leaked Search API documentation in 2024 and the DOJ trial documents in 2023-2024, suggests the system aggregates many signals into composite scores. NavBoost (engagement-derived ranking) and related systems likely play a role. The HCU signal is one of many, but it weights heavily for sites with the demoted patterns above.

What we do not know

We do not know specific thresholds. We do not know whether perplexity is computed against a current language model or a baseline. We do not know how Google distinguishes "AI-assisted but human-edited" from "raw AI". We do not know the exact site-level aggregation function. SEO industry analysis makes educated guesses; Google has not published the formula.

10. Bing and Copilot

Bing matters beyond its own results page, because other surfaces draw on it. DuckDuckGo says the traditional links in its results come "largely" from Bing (help page, read on 18 October 2026), and Microsoft's Copilot answers from Bing's results. When you optimize for Bing, you are also affecting those surfaces.

Copilot search

Microsoft Copilot integrates with Bing search and answers with a written summary that cites its sources, the same shape as Google's AI Overviews. The signals that put a site in that source list are not documented in detail.

Strategic implication

Check your own numbers before investing: Search Console and Bing Webmaster Tools both report the clicks each engine sends you. Work on Bing comes on top of Google, not instead of it, and it also reaches DuckDuckGo's links and Copilot's answers.

11. Perplexity, You.com, Phind, Komo

These engines share a model: instead of ten blue links, they return a composed answer with cited sources.

Perplexity

Perplexity returns an AI-generated answer with numbered citations to sources. Its crawler documentation (read on 18 October 2026) says PerplexityBot "is designed to surface and link websites in search results on Perplexity" and "is not used to crawl content for AI foundation models", and recommends allowing it "to ensure your site appears in search results". Perplexity does not publish how it picks the sources it cites. For publishers, the incentive is mixed: getting cited builds brand impressions but may or may not generate clicks.

You.com

You.com is an AI search and assistant platform that shows its sources in its answers.

Phind

Phind presents itself as an AI search engine for developers.

Komo

Komo is a conversational search engine that composes answers from cited sources.

What we can say about this cohort

None of these engines documents in detail how it picks the sources it cites, so we do not list what each one rewards or penalizes. Our working assumption, not a measurement: an engine that quotes sources needs pages it can quote, so clear factual writing, a named author and claims that carry their own sources are the safe bet. That is the same hygiene the rest of this article recommends for Google.

12. The 2026 Playbook for Publishers and Marketers

This section is concrete. Based on the patterns across the previous sections, here is what site owners should do in 2026.

12.1 Author bio with photo and credentials on every article

Every article needs:

• A named author (not "Editorial Team" or "Admin")
• A photo of that author
• A bio that includes relevant credentials
• A link to the author's bio page on the site
• The author's bio page should link to external presence (LinkedIn, Twitter/X, professional site, scholar profile)
• Schema.org Person markup tying the author to the article

Cost: low. Impact: high. This is the single most-cited recommendation across HCU recovery analyses.

12.2 Original research, data, or real product testing

Articles need to include something that wasn't already on the web. Options:

• Original survey data: even a small reader survey produces citable content
• Internal data: traffic patterns, customer behavior, support tickets: anything proprietary
• Real product testing: photos of you using the product, weighing it, opening the box
• Site visits: photos of the place you're reviewing
• Interviews with experts: even one quote from a named source raises the originality of an article significantly
• Original analysis: combining public data in a new way

The bar is not "academic research". The bar is "did this article exist on the web before you published?" If the answer is no, you're contributing originality.

12.3 Anti-slop visual design

EEAT is now visual. Sites that look like content farms (generic stock images, no faces, identical layouts) present as low-quality even if the text is strong.

Anti-slop visual design includes:

• Original photography or custom illustration
• Author photos in bylines
• Distinctive layout that's not a generic theme
• Real screenshots from real testing, not stock UI mockups
• Custom diagrams for complex topics
• Visual variety across articles

For more on anti-slop visual design see /blog/de-ai-your-lovable-v0-bolt-site.

12.4 Schema.org Article + Person + Organization

The full schema.org pattern includes:

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Organization",
      "@id": "https://example.com/#organization",
      "name": "Example Publication",
      "url": "https://example.com",
      "logo": {
        "@type": "ImageObject",
        "url": "https://example.com/logo.png"
      },
      "sameAs": [
        "https://twitter.com/example",
        "https://www.linkedin.com/company/example/"
      ]
    },
    {
      "@type": "Person",
      "@id": "https://example.com/author/jane-doe#person",
      "name": "Jane Doe",
      "url": "https://example.com/author/jane-doe",
      "image": "https://example.com/authors/jane-doe.jpg",
      "jobTitle": "Senior Editor",
      "knowsAbout": ["Topic A", "Topic B"],
      "sameAs": [
        "https://twitter.com/janedoe",
        "https://www.linkedin.com/in/janedoe/"
      ]
    },
    {
      "@type": "Article",
      "@id": "https://example.com/article-slug#article",
      "headline": "Article Headline",
      "description": "Article description.",
      "image": "https://example.com/article-image.jpg",
      "datePublished": "2026-04-28T08:00:00-04:00",
      "dateModified": "2026-04-28T08:00:00-04:00",
      "author": {
        "@id": "https://example.com/author/jane-doe#person"
      },
      "publisher": {
        "@id": "https://example.com/#organization"
      }
    }
  ]
}
</script>

12.5 Internal linking around topical clusters

Topical clusters are groups of related articles linked together. The pattern:

• A central pillar article on a broad topic
• Multiple cluster articles on subtopics
• Internal links between cluster articles and the pillar
• Internal links between related cluster articles

What this signals: the site has topical depth. It's not just a single article on a topic; it's a comprehensive coverage of a subject area.

What to avoid: random internal links that don't reflect topical relationships. Footer links to every page on the site. "Related articles" widgets that pull from the entire site instead of the cluster.

12.6 Update dates that mean something

The "Last updated" date on an article should reflect actual updates. Changing the date without changing the content is a pattern that has been deprecated by Google's spam policies.

Real updates can include:

• New data added
• Outdated information removed
• Photos refreshed
• New product variants reviewed
• Reader questions answered in an FAQ section
• Citations added for new sources

When you do update an article, log what changed. A short "Updated April 2026: added section on new model" note at the top builds trust.

12.7 Avoid the banned-phrase list

The phrases that AI overuses, and that correlate with HCU demotion, include:

• "delve" / "delving"
• "tapestry"
• "leverage" (especially as a verb)
• "robust"
• "seamless"
• "harness" (as a verb)
• "revolutionize"
• "empower"
• "unleash"
• "navigate the landscape"
• "in today's fast-paced world"
• "in the ever-evolving"
• "by following these tips"
• "in conclusion"
• "it's important to note that"

These aren't forbidden words in absolute terms. They're tells. If your article reads as full of them, it pattern-matches to AI output. For more on this see /blog/anti-slop-prompt-template-2026 and /blog/detect-ai-generated-site-30-seconds-21-signs-2026.

12.8 Comments, community, and forum signals

Engagement signals matter. Options for sites that want to add community surface area:

• Article comments (well-moderated)
• A dedicated community on Discord, Discourse, or Circle
• A subreddit for the publication
• Newsletter with replies
• Reader Q&A sections

The point is not to game an engagement metric. The point is that sites with active communities tend to satisfy the patterns Google's helpfulness signals favor.

Mermaid: SEO 2026 decision flowchart

flowchart TD
    A[Have an article idea] --> B{Has it been written elsewhere?}
    B -->|Yes, many times| C{Can you add original data, testing, or analysis?}
    B -->|No, undercovered| D[Proceed]
    C -->|Yes| D
    C -->|No| E[Skip - won't differentiate]
    D --> F{Does the topic require expertise YMYL?}
    F -->|Yes| G[Confirm credentialed author]
    F -->|No| H[Confirm experienced author]
    G --> I[Write with first-person experience]
    H --> I
    I --> J{Does it have original photos / data / testing?}
    J -->|No| K[Add original content before publishing]
    J -->|Yes| L[Add author bio + schema]
    K --> L
    L --> M[Add internal links to topical cluster]
    M --> N[Publish]
    N --> O{Update needed in 6 months?}
    O -->|Yes| P[Real update + dated note]
    O -->|No| Q[Leave as is]

13. SEO Tools: What to Track

The SEO tool industry has had to adapt to a market where keyword optimization is necessary but no longer sufficient.

We do not rate individual vendors here: their feature lists change faster than this article, and we have not tested them side by side. Whatever tool you use, what is worth tracking now goes beyond keyword positions:

• Topical authority (do you cover this topic comprehensively?)
• Content quality and originality scoring
• AI content detection
• Reddit and forum visibility tracking
• AI Overviews tracking (does your site appear in AI answers?)
• Branded query tracking
• Author entity tracking

This is a structural change. The 2015-2022 "find a keyword, rank for the keyword, monetize the traffic" playbook still works in narrow categories but has been broadly displaced in 2026.

14. Predictions for 2027

These are predictions, clearly labeled. We are not stating them as facts.

Full SGE/AI Overviews rollout in all markets

By 2027, AI Overviews are likely to cover a higher percentage of queries globally. The product will mature. Click economics on informational queries will continue to be the central topic of publisher complaint.

Content provenance metadata (C2PA)

The C2PA (Coalition for Content Provenance and Authenticity) standard for embedding origin metadata in content has been gaining adoption. By 2027, schema.org or related standards may include AI-generation provenance fields. Sites may be expected to disclose what was AI-generated and what was human-written. Whether this is mandatory or recommended is uncertain. Whether Google uses it as a ranking signal is uncertain.

End of the affiliate review site as a category

The trend lines for independent affiliate review sites are downward across 2022-2026. By 2027, the category may be functionally extinct as a Google traffic source. Surviving sites will have shifted to other distribution: newsletter, YouTube, paid memberships, branded audiences. The pure SEO-driven affiliate site that was viable in 2018 is not viable in 2027 in our estimation.

Reddit moat fragmenting

If Reddit's gains of 2024, as the publishers above describe them, hold, they amount to a kind of moat. By 2027, that moat may erode for several reasons:

• AI Overviews summarize Reddit threads, reducing click-through to Reddit
• Reddit's licensing arrangements, like the one with Google, shape what gets surfaced
• Quality problems on Reddit, like the spam replies HouseFresh found at the top of a thread (Appendix D), become more visible at scale
• Other forum platforms gain ranking

Branded query optimization

The shift toward brand-name commerce extends to a shift toward branded query optimization. Sites that build a brand with direct query volume ("[brand name] reviews", "[brand name] for X use case") have moats that algorithm changes don't easily disrupt.

The rise of accountability infrastructure

Author bios, schema.org markup, original photography, real testing: these are all "accountability infrastructure". By 2027 we expect this to extend further: verified author identities (perhaps via decentralized identity standards), publisher reputation scoring, formal originality certificates. The web is becoming more accountable to itself, partly under pressure from AI content.

For more on the trajectory of AI content see /blog/ai-slop-2026-state-of-the-ai-generated-web and /blog/ai-slop-economy-2026.

15. FAQ

Q1: Is AI content automatically demoted by Google?

No. Google's stated position, since its February 2023 guidance on AI-generated content, is that AI content is not automatically demoted. The HCU and related systems target patterns of low-quality content regardless of origin. AI content tends to exhibit more of those patterns by default, so AI content is more likely to be demoted, but it's the pattern not the origin that triggers demotion.

Q2: Can I use AI to write articles and still rank?

Yes, with significant editorial work. AI-assisted writing where a knowledgeable human edits, adds original information, and ensures quality can rank fine. The risk pattern is raw AI output published without editorial intervention. The practical workflow that has worked in case studies: outline by hand, generate a draft with AI, treat the draft as the rough material, rewrite from scratch in your own voice, add data and original observation that the AI cannot have, fact-check against primary sources, and only then publish. Sites that follow this kind of editorial loop have been less affected by HCU than sites that ship raw output.

Q3: What if I disclose that I use AI?

Disclosure does not guarantee ranking, but transparency is consistent with Trust signals. There's no penalty for disclosure. Some publishers add an editorial policy page describing how they use AI. It's a Trust signal, not a guarantee.

Q4: Will my HCU-affected site recover?

Recovery is rare on the timescale Google has historically claimed. The two sites that published their numbers, HouseFresh and Retro Dodo, had not recovered when they wrote in spring 2024, and Google has published no recovery figures. Recovery requires structural improvements: real authors, original content, EEAT investment, often over many months.

Q5: Should I move my SEO efforts to Bing?

Probably not as a replacement. Check what each engine sends you first: Search Console and Bing Webmaster Tools both report clicks. Work on Bing is an addition, and it also reaches DuckDuckGo's links and Copilot's answers.

Q6: Will Reddit's ranking surge last?

We don't know. The publishers quoted in this article describe a sharp rise in 2024; Google has published nothing on how much Reddit gained or for how long. What happens next depends on how Google weighs community content, on the quality of Reddit threads, and on Reddit's relationship with Google (the partnership announced in February 2024). Do not build a plan that depends on either outcome.

Q7: How much does schema.org markup matter?

It matters as table stakes. Sites without proper schema.org markup are at a disadvantage. Sites with comprehensive Article + Person + Organization markup are baseline-competitive. Schema.org alone won't save a site with poor content, but its absence removes a signal.

Q8: Should I block AI crawlers in robots.txt?

It depends on goals. Blocking AI crawlers (GPTBot, ClaudeBot, anthropic-ai, Google-Extended, etc.) prevents your content from being used to train AI models. It does not prevent appearing in AI search results that grounded on your live content. Many publishers have chosen to block training crawlers while allowing real-time citation crawlers. The robots.txt block looks like:

User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: anthropic-ai
Disallow: /

User-agent: Google-Extended
Disallow: /

Google-Extended blocks Google's training; it does not affect Google Search ranking.

Q9: How long should an article be in 2026?

It depends on the topic. The "long content always ranks better" myth is dead. What matters is comprehensiveness. If a topic deserves 800 words, write 800 words. If it needs 8,000, write 8,000. Padding short topics produces the bloat patterns that HCU demotes.

Q10: How often should I update content?

When there's something to update. Recipe updates that don't change the recipe but bump the date are exactly the pattern Google's spam policy named in 2024. Real updates: yes. Fake updates: no.

Q11: Should I worry about my old content?

Audit it. If you have 2018-2022 content that was written quickly and lacks author bios or original information, that content is now a sitewide drag. Options: improve it, archive it (noindex), or remove it. Cleaning up old content has helped sites recover in some public case studies.

Q12: Are AI-generated images a ranking factor?

There is no specific public confirmation that Google demotes AI-generated images. But original photography is a signal of first-hand experience that AI images do not provide. For products and places you describe, take real photos.

Q13: Does Google penalize sites that use AI for translation?

Not specifically. Translation is a different use case from content generation. AI-translated content from a high-quality source can rank well if the translation is accurate. The HCU patterns still apply to the original content.

Q14: What about sites that write only press releases or sponsored content?

Sites that are predominantly thin promotional content correspond to the patterns that HCU demotes. Sponsored content can rank if it is well-written, declared as sponsored, and provides real value. Press release aggregators have lost ranking broadly.

Q15: How do I check if my site has been hit by HCU?

Use Google Search Console. Look at organic traffic by month. Look for sharp drops aligned with HCU and Core Update dates (the timeline above gives the dates). Compare ranking position changes for your top queries. Third-party rank trackers can show the same drops on the keywords they track. The diagnosis is empirical.

16. Glossary and Sources

Glossary

• HCU: Helpful Content Update. Google ranking system targeting unhelpful content. Folded into Core Updates from March 2024.
• EEAT: Experience, Expertise, Authoritativeness, Trust. Quality framework introduced as E-A-T in 2014, expanded to EEAT in December 2022.
• AI Overviews: Google's AI-generated summary at the top of search results, launched US May 2024.
• SGE: Search Generative Experience. The Labs version of AI Overviews, 2023.
• YMYL: Your Money or Your Life. High-stakes content categories (health, finance, legal) where EEAT applies most strictly.
• Core Update: Google's broad ranking system updates, several per year.
• Spam Update: Google updates targeting policy-violating content.
• NavBoost: Engagement-derived ranking signal disclosed in DOJ trial documents.
• C2PA: Coalition for Content Provenance and Authenticity. Standard for embedding content origin metadata.
• Site Reputation Abuse: Google policy against parasite SEO, where third-party content uses an authoritative site's domain for ranking.
• Schema.org: Structured data vocabulary for marking up web content.
• Topical Authority: Concept that sites covering a topic comprehensively outrank sites with single articles on that topic.
• Branded Query: Searches that include a specific brand name.
• Zero-Click Search: Searches that resolve in the SERP without a click to an external site.

Sources

Every date and figure in this article comes from one of these, read on 18 October 2026.

• Google, Search Status Dashboard, ranking updates.
• Google Search Central, "What creators should know about Google's August 2022 helpful content update", 18 August 2022.
• Google Search Central, "Our latest update to the quality rater guidelines: E-A-T gets an extra E for Experience", 15 December 2022.
• Google Search Central, "Google Search's guidance about AI-generated content", 8 February 2023.
• Google, "An expanded partnership with Reddit", 22 February 2024.
• Google Search Central, "What web creators should know about our March 2024 core update and new spam policies", 5 March 2024.
• Google, "New ways we're tackling spammy, low-quality content on Search", 5 March 2024, updated 26 April 2024.
• Google, "Generative AI in Search: Let Google do the searching for you", 14 May 2024.
• Google, "100 things we announced at I/O", 21 May 2025.
• Google Search Central, "Creating helpful, reliable, people-first content", last updated 5 October 2026.
• HouseFresh, Gisele Navarro, "How Google is killing independent sites like ours", 19 February 2024.
• HouseFresh, Gisele Navarro, "HouseFresh has virtually disappeared from Google Search results. Now what?", 1 May 2024.
• Retro Dodo, Brandon Saltalamacchia, "Google Is Killing Retro Dodo & Other Independent Sites", 3 April 2024.
• Google, Elizabeth Reid, "AI Overviews: About last week", 30 May 2024.
• Perplexity, "Perplexity crawlers", documentation, read on 18 October 2026.
• DuckDuckGo, "Results sources", help page, read on 18 October 2026.
• The New York Times, robots.txt, read on 18 October 2026.
• Pew Research Center, Athena Chapekis and Anna Lieb, "Google users are less likely to click on links when an AI summary appears in the results", 22 July 2025.

For Sailop readers continuing this thread, see also:

• /blog/ai-slop-2026-state-of-the-ai-generated-web
• /blog/anti-slop-prompt-template-2026
• /blog/de-ai-your-lovable-v0-bolt-site
• /blog/ai-slop-economy-2026
• /blog/detect-ai-generated-site-30-seconds-21-signs-2026

Appendix A: Deep Dive on the September 2023 HCU

The September 2023 Helpful Content Update is, in retrospect, the moment a generation of small publishers learned what algorithmic risk feels like at scale. Google started the update on September 14, 2023, and its dashboard gives a rollout of just under two weeks. In those two weeks, small publishers refreshed Search Console daily and watched curves fall.

What made September 2023 different from August 2022 (the original HCU)? Three things, in our reading.

First, the model the algorithm targeted had matured. By September 2023, Google's helpful content classifier had a year of training data on what "made for search engines" content looked like in the post-ChatGPT era. The classifier was not just looking for content farms. It was looking for the structural patterns that had become characteristic of optimized affiliate content: the long intro before the answer, the comparison table early, the "what to look for in a [product]" section with five generic bullet points, the "frequently asked questions" section with template questions.

Second, it is the update that small publishers' public post-mortems keep returning to (Retro Dodo dates its losses from it), though Google has never published how many sites any update affected.

Third, the update arrived in a content-saturated market. By September 2023, ChatGPT had been public for ten months. Many sites had migrated their workflows toward AI-assisted production. The HCU encountered a much larger surface area of demoteable content than it would have a year earlier.

Who specifically reported losses

In the months following September 2023, several publishers wrote retrospectives. We do not invent names. Two publicly named cases came with their own figures:

• HouseFresh (Gisele Navarro), in two posts on its own site on 19 February and 1 May 2024, made the case that Google rewarded large-publisher "best of" pages over its own product testing, and reported a 91% loss of search traffic after the March 2024 core update.
• Retro Dodo (Brandon Saltalamacchia), in a post on 3 April 2024, reported an 85% drop in organic traffic and revenue since September 2023.

There were many others who shared analytics screenshots on X (formerly Twitter) but did not write longer post-mortems. The total population of affected sites is unknown: Google has not published numbers, and there is no public registry of HCU-affected sites.

The post-mortem patterns

Across the published post-mortems, several patterns emerged:

The hit was sitewide, not page-level. Sites lost ranking on their best content as well as their worst. This is consistent with HCU's stated sitewide nature.

Recovery was slow. Neither HouseFresh nor Retro Dodo reported a recovery in the posts above, written in spring 2024.

The category mattered, in the sites' own telling. HouseFresh's complaint is about commerce queries, the "best air purifier" kind, where big media publishers had built large "best of" sections.

Big publishers were the ones named as winners. HouseFresh's February 2024 post shows Better Homes & Gardens and Real Simple at the top of the results for air purifier queries, and asks what BuzzFeed, Rolling Stone, Forbes, Popular Science and Better Homes & Gardens have in common.

The structural argument

A structural reading of the September 2023 HCU goes something like this. Google had a brand-safety problem: AI-generated content was flooding the index. Google's options were limited. Demoting AI content explicitly would be hard to do without false positives. Demoting based on content patterns would have collateral damage. Promoting big publishers and Reddit was a way to reduce the prominence of any individual algorithmic decision while making the search results "look" higher quality on average.

Whether this structural reading is correct, we don't know. Google has never said this. But the effect the two publishers describe (their own sites down, big publishers and Reddit up) is consistent with it.

For Sailop readers continuing to think about this kind of structural shift, see /blog/ai-slop-2026-state-of-the-ai-generated-web.

Appendix B: How the March 2024 Core Update Worked

The March 2024 Core Update started on March 5, 2024 and rolled out for 45 days, to April 19, per Google's dashboard. Google's announcement for creators called it "a more complex update than our usual core updates". Its product post the same day said that, with its earlier efforts, the update would "collectively reduce low-quality, unoriginal content in search results by 40%"; on 26 April it reported "45% less low-quality, unoriginal content in search results versus the 40% improvement we expected".

Three things changed.

First, the Helpful Content System was absorbed into Core Ranking Systems. The standalone "HCU" ceased to exist as a discrete update. From March 2024 forward, helpfulness signals are computed continuously and applied through Core Updates.

Second, Google introduced new spam policies. The most consequential was the "scaled content abuse" policy, which targeted sites that publish bulk AI-generated content with little human oversight. The policy language was deliberately broad: scale alone is not the issue, but scale combined with low quality and lack of oversight is.

Third, Google introduced a "site reputation abuse" policy targeting parasite SEO. Parasite SEO is the practice of placing third-party content (often affiliate or sponsored) on a high-authority domain to inherit its ranking power. The policy applied to sites like established news organizations that had begun publishing extensive coupon, deals, or product comparison content under their main domain. Google enforced this policy with manual actions in some cases.

The combined impact

The March 2024 Core Update, taken together with the Spam Update that landed the same week, produced a second wave of impact on top of the September 2023 HCU. Sites that had been wounded but not killed by September 2023 took further losses. Sites that had been operating parasite SEO arrangements with established publishers had those arrangements unwound.

On the gains side, the sites that wrote post-mortems name big media publishers (HouseFresh) and Reddit. Retro Dodo's founder wrote that the Reddit partnership became public "just days before the update that catapulted Reddit's traffic to the moon" (3 April 2024); announced on 22 February 2024, it preceded the start of this update by twelve days. We know of no published case of a specialty publisher that this update rewarded for cleaning up.

Documented commentary

The only figure on the outcome is Google's own 45%, measured on its own evaluations. We know of no independent measurement of the same thing. What the update did to individual sites is documented by the sites themselves, as HouseFresh did.

Appendix C: The AI Overviews Launch in Detail

AI Overviews (originally Search Generative Experience) launched in the US in May 2024 at Google I/O. Within days, screenshots of wrong or absurd Overviews circulated. Google answered on 30 May 2024, in a post by Elizabeth Reid, its VP of Search, that names two of them: "How many rocks should I eat?", where the Overview linked to satirical content "republished on a geological software provider's website", and Overviews that "featured sarcastic or troll-y content from discussion forums", such as "using glue to get cheese to stick to pizza".

The same post lists what Google changed: "better detection mechanisms for nonsensical queries", limits on "satire and humor content" and on "user-generated content in responses that could offer misleading advice", and "triggering restrictions for queries where AI Overviews were not proving to be as helpful".

Mechanism

AI Overviews use a model (a Google-developed LLM, with details that have evolved) to compose an answer to a query, drawing on web sources. The Overview is shown above the traditional ten blue links. Sources are cited inline, with links to the originating sites.

The mechanism creates several effects:

• For the user: the answer is faster, and sometimes the user does not need to click any source.
• For the publisher: the click is sometimes lost; the citation is sometimes gained as a brand impression.
• For Google: the search experience improves on speed; the relationship with the open web complicates because Google is now competing with the publishers it indexes.

Publisher reactions

The publisher reaction has been varied. Some publishers have gone to court over the use of their articles by AI companies, and some have negotiated content licensing deals. Some block AI crawlers while allowing search: the New York Times' robots.txt, read on 18 October 2026, disallows GPTBot, ClaudeBot, CCBot, Google-Extended and PerplexityBot, among others, while leaving Googlebot in the general rules. Others have changed nothing.

The strategic options for publishers in 2026 are:

• Block training crawlers, allow search: defensive on training data, neutral on AI Overviews
• Block all AI access: most defensive, but does not actually prevent AI Overviews if Google indexes you for search
• Allow everything, optimize for citation: aggressive, aiming for brand impression and partnership
• Negotiate licensing: only available to large publishers with leverage

What's likely to evolve

By 2027, we expect AI Overviews coverage to expand. Click decline on informational queries to continue to be a publisher concern. Content licensing deals to multiply. Possibly: a regulatory dimension, especially in the EU, where digital markets and AI transparency rules may shape what Google can do with publisher content.

For more on the broader ecosystem, see /blog/ai-slop-economy-2026.

Appendix D: A Closer Look at the Reddit Surge

Reddit is the site the publishers who lost traffic in 2023 and 2024 name as the winner (sections 3 and 5). Here is what is documented about it, and what is only our reading.

The licensing deal

On 22 February 2024, Google announced an expanded partnership with Reddit: Google gets access to Reddit's Data API, which it says it will use to "display, train on, and otherwise use" Reddit content. The announcement gives no value for the deal. Retro Dodo's founder wrote that the news became public "just days before the update that catapulted Reddit's traffic to the moon".

Whether the ranking changes follow from the partnership is disputed; the timing is what observers point to.

The fit with HCU goals

Our reading, not something Google has said: Google's self-assessment asks for first-hand experience, and a thread is people describing their own use, with follow-up questions and answers underneath. That format matches what the questions ask for in a way a generated article does not, whatever the quality of the individual replies.

The downsides of Reddit dominance

A thread is only as good as its replies. In its 19 February 2024 post, HouseFresh opened the Reddit thread ranking just after BuzzFeed for "best air purifier for pets" and found "lots of spammy replies", the top one posted by an account Reddit had since suspended. Google's own 30 May 2024 post says forums "in some cases can lead to less-than-helpful advice". Promoting forums broadly means promoting that variance.

What 2027 might look like

The Reddit surge may not last in its current form. Possible evolutions, all of them guesses:

• AI Overviews summarize Reddit threads, extracting the answer without driving a click to Reddit
• Google adjusts the Reddit weighting after publisher pushback
• Other UGC platforms (Quora, Stack Exchange properties, smaller forums) gain comparable ranking
• Reddit's own content quality declines under spam pressure

In any of these scenarios, the strategic implication for publishers is: treat the Reddit moat as temporary. Build a brand and an audience that does not depend on Google ranking decisions.

Appendix E: The Originality Problem

A central tension in 2026 SEO is the originality problem. Google says it rewards original content. The web is full of derivative content. AI accelerates derivative content production. How does Google measure originality?

Patent-disclosed approaches

Google has filed patents over the years describing originality detection mechanisms. These include:

• Comparing a passage against the corpus of indexed content to identify near-duplicates
• Identifying content that introduces new information not previously in the index
• Citation network analysis (which content gets linked, quoted, referenced)
• Author network analysis (which authors are cited by other authors)

We don't know which of these mechanisms are in production. Patents indicate research interest, not necessarily deployment.

The originality vs comprehensiveness tension

A long, comprehensive article that synthesizes existing knowledge is helpful to users, but it might score low on originality if every claim is paraphrased from elsewhere. A short, narrow article with one new insight might score high on originality but low on comprehensiveness.

Google's apparent solution: reward articles that combine both. Comprehensive coverage with at least some original component (data, photos, quotes, analysis, or first-hand experience) tends to outperform either pure synthesis or pure original snippets.

The implication for AI-assisted writing

If your AI-assisted article is purely synthesized from existing web content, it is by construction not original. To make AI-assisted writing competitive, you need to inject originality at the editorial stage. Photos. Data. Real experience. Quotes. New analysis. Anything that is not in the AI's training data and not in the existing top-ranking pages.

For a deeper treatment of how to write AI-assisted content that survives, see /blog/anti-slop-prompt-template-2026.

Appendix F: Robots.txt for the AI Era

The robots.txt file has become an active battleground in 2024-2026. Multiple AI crawlers exist. Different sites take different positions. Below is a comprehensive robots.txt template covering the major AI crawlers as of 2026.

# Search engines (allow)
User-agent: Googlebot
Allow: /

User-agent: Bingbot
Allow: /

User-agent: DuckDuckBot
Allow: /

# AI training crawlers (block, optionally)
User-agent: GPTBot
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: anthropic-ai
Disallow: /

User-agent: Google-Extended
Disallow: /

User-agent: CCBot
Disallow: /

User-agent: Bytespider
Disallow: /

User-agent: Amazonbot
Disallow: /

User-agent: Applebot-Extended
Disallow: /

User-agent: cohere-ai
Disallow: /

User-agent: Meta-ExternalAgent
Disallow: /

# AI answer engine (block only if you do not want to be cited there)
User-agent: PerplexityBot
Disallow: /

# General default
User-agent: *
Allow: /

Sitemap: https://example.com/sitemap.xml

A few notes on this template.

Google-Extended is Google's training-only crawler identifier. Blocking Google-Extended does not affect your appearance in Google Search or AI Overviews: those use the regular Googlebot. Blocking Google-Extended only blocks training data inclusion for Google's standalone AI products.

PerplexityBot is Perplexity's crawler. Perplexity's crawler documentation (read on 18 October 2026) says it "is designed to surface and link websites in search results on Perplexity" and "is not used to crawl content for AI foundation models", which is why the template lists it apart from the training crawlers: blocking it is a choice about being cited, not about training.

Bytespider is associated with ByteDance, TikTok's parent company.

Applebot-Extended is Apple's training-only crawler, separate from Applebot which is for search.

Site reputation considerations: blocking aggressively can be a brand statement or a strategic choice. The New York Times' robots.txt, read on 18 October 2026, disallows every AI agent in this template, PerplexityBot included, except Amazonbot, which it keeps out of its Wirecutter pages only. Allowing aggressively can be a discoverability bet (you want to be the cited source).

Appendix G: Topical Cluster Architecture in Practice

Topical clusters are simple in concept and tricky to execute. Here's a worked example.

The structure

Imagine you run a site about home brewing coffee. A topical cluster around "espresso" might look like:

• Pillar: "The Complete Guide to Home Espresso" (4,000-6,000 words covering everything)
• Cluster pages:

- "How to Choose an Espresso Machine" - "Pump Espresso vs Lever Espresso vs Manual" - "Pressure Profiling Explained" - "How to Calibrate a Grinder for Espresso" - "Tamping Technique" - "Pulling a Shot: Step by Step" - "How to Steam Milk for Latte Art" - "Espresso Machine Maintenance Schedule" - "Decalcifying an Espresso Machine" - "Best Espresso Beans for Home Brewing"

The pillar links to all cluster pages. Each cluster page links to the pillar and to relevant sibling pages. The internal link graph forms a star with cross-links between siblings.

The signal to Google

Google's topical authority signals respond to this structure. The site is recognizable as a coffee-and-espresso site. Searches for any of these subtopics can find the right page. The cluster signals depth without requiring every page to be 8,000 words.

Common mistakes

Sites attempt topical clusters and fail in characteristic ways:

• Random internal links: linking from every page to every other page dilutes the signal. The "Related articles" widget that pulls from the entire site does not build a cluster.
• Pillar that's just a hub: a pillar page that's only a list of links to cluster pages does not satisfy users who land on it. The pillar should be a useful, comprehensive article in its own right.
• Cluster pages that overlap: if two cluster pages cover overlapping topics, you cannibalize. Each cluster page should have a distinct intent.
• Missing cluster coverage: if you have a pillar but only three cluster pages, the cluster is thin. The cluster needs enough pages to demonstrate depth.

Cluster maintenance

Topical clusters need maintenance. As the topic evolves, new cluster pages need to be added. Old cluster pages need updating. The pillar needs revision to reflect updates. Sites that build a cluster and let it stagnate lose the topical authority benefit over time.

Appendix H: Update Cadence and Date Stamping

The "Last updated" date is more contentious than it should be. Some sites stamp the date as the current date on every page load (or via build), which Google has explicitly criticized as a manipulation. Some sites never update content, which makes the published date the only meaningful timestamp.

The right cadence depends on the topic.

Topic-specific cadence

What constitutes a real update

A real update changes the content. A few examples of legitimate updates:

• Adding a new section that addresses reader questions from comments
• Replacing screenshots with current versions when the UI has changed
• Adding new data, citations, or sources
• Updating product variants, prices, or availability
• Removing outdated claims and replacing them with accurate ones
• Adding new internal links to recently published cluster pages

What does not constitute an update

• Changing the date stamp without changing the content
• Reordering paragraphs without adding information
• Replacing words with synonyms
• Adding "Updated April 2026" to the title without changing anything below

Google's spam policies have explicitly named the "stale content with refreshed date" pattern. Don't do it.

Appendix I: Author Bio Patterns That Work

The author bio has become a load-bearing piece of EEAT signaling in 2026. Here are patterns that work.

Minimum viable author bio

• A photo of the author (real, identifiable, professional but not stock)
• The author's name as a clear byline
• One or two sentences of professional context
• Years of experience in the topic
• A link to the author's bio page

Comprehensive author bio (what high-EEAT sites use)

• All of the above
• A longer biography on the author's bio page
• Education and credentials
• Professional affiliations
• Published work (book, articles in named publications, peer-reviewed papers)
• External links: LinkedIn, personal site, scholar profile, Twitter/X
• Schema.org Person markup
• Email contact (corrections, tips)
• The author's specific expertise topics

Anti-patterns to avoid

• "Editorial Team" as the byline
• A generic photo (stock, AI-generated, pixar-style avatar)
• A bio that reads "John Doe is an experienced writer who covers [topic]": empty boilerplate
• An author with no external presence (Google can detect this)
• An author bio shared by multiple authors (the "team bio" pattern)
• Authors with names that don't appear elsewhere on the web
• AI-generated authors (this has happened at multiple sites and has been called out publicly)

The author photo specifically

The author photo is a load-bearing signal. Stock photos of "professional women looking thoughtful" or "diverse stock people in office" are a tell. Real authors look like real people. They may have casual photos. They may have less polished setups. The realness of the photo correlates with the realness of the author.

For sites that want to take this further: photo galleries of the author actually using products, visiting places, or doing the thing they write about. This is the "demonstrated experience" signal embodied in image form.

Appendix J: Original Photography in Practice

Original photography is one of the strongest signals of first-hand experience. It is also one of the hardest things to fake. AI image generators in 2026 can produce convincing photographs, but the photographs they produce do not have the specific contextual cues of real photos.

What original photography looks like in product reviews

• The product unboxed, photographed in your actual space (a real desk, a real kitchen, a real shelf)
• The product in use, with realistic context (the cat is in the photo, the lighting is what it is)
• The packaging, including the sticker that came with the specific unit
• The product alongside other reference items (a coin for scale, a previous version for comparison)
• Multiple angles, including the boring back panel

What original photography looks like in travel content

• Real photos of the place from your own visit
• Photos of you (or the contributor) in the place, with the place identifiable
• Photos that show the actual conditions (it was raining, it was crowded, it was off-season)
• Photos that include details a stock photo would miss (the menu in handwriting, the signage in the local language, the staff)
• Time-stamped or geotagged metadata that matches the claimed visit

What original photography is not

• Stock photos
• Manufacturer photos (these can be in the article, but they should be additional to your own)
• AI-generated photos
• Photos from previous visits a year ago, claimed as current
• The same photo recycled across multiple articles

Practical workflow

A practical workflow that has been documented by surviving review sites:

• Plan the photo coverage before the review
• Take more photos than you think you need
• Photograph in natural light when possible
• Include scale references
• Photograph the unboxing, the use, the maintenance, the disposal
• Edit minimally: over-editing destroys the realness signal
• Use the photos in the article in a way that tells the story of your testing

For sites that want to invest, this becomes a competitive moat. AI cannot generate the specific photos of a specific testing session. Sites that build a photo archive across years of testing accumulate a defensive position that AI farms cannot replicate.

Appendix K: The Competitive Landscape of AI Search Engines

The AI search engine cohort is competitive in 2026. Several startups and incumbents are vying for share.

Google AI Overviews

Shown inside Google Search; Google reported 1.5 billion monthly users of AI Overviews in May 2025 (section 7). Click economics are the central issue.

Microsoft Copilot

Integrated with Bing and bundled with Microsoft 365, so it reaches people through Microsoft's own products.

ChatGPT Search

OpenAI's search product, inside ChatGPT, so it reaches people through ChatGPT's own user base.

Perplexity

Standalone AI search, with its own documented crawler (Appendix F).

You.com

AI search and assistant platform.

Phind

Presents itself as AI search for developers.

Komo

Conversational search.

Anthropic's positioning

Anthropic (the maker of Claude) does not currently operate a public search product as of April 2026. Claude has search capability through tool use in Claude apps, but it is not a competitor to Google Search in the traditional sense.

Strategic implication for publishers

The cohort is fragmented, and none of these products documents in detail how it picks the sources it cites, so we do not rate them against each other. Our working assumption, not a measurement: what makes a page quotable for one (clear factual writing, a named author, claims that carry their own sources) is what makes it quotable for the others. The risk adds up the same way: if all of these surfaces answer without sending a click, the losses accumulate.

Appendix L: What HCU Recovery Actually Looks Like

Recovery from HCU is rare on the timescale Google has historically claimed. But some sites have partially recovered. What has worked, in publicly documented cases?

Pattern 1: Aggressive cleanup

Some sites that suffered HCU in September 2023 spent the following months auditing their content, removing or merging weak articles, rewriting articles that lacked author bios or original information, and adding photos and original data where they could. After the August 2024 Core Update, some of these sites partially recovered.

The pattern: don't try to game the recovery. Actually fix the patterns the HCU was designed to demote.

Pattern 2: Distribution diversification

Some sites that lost ranking diversified their traffic. They invested in newsletter, YouTube, social, and direct audience. They reduced their dependency on Google traffic. The Google traffic did not recover, but the business survived.

The pattern: HCU recovery is often a story of business model recovery, not Google ranking recovery.

Pattern 3: Niche refocus

Some sites that lost ranking on broad consumer queries refocused on narrower niches where Google's big-publisher promotion did not extend. A site that lost on "best air purifier" might find ranking on more specific subtopics, technical depth, or hobbyist-specific angles.

The pattern, as we read it: the broad commercial queries, where HouseFresh found big publishers, Amazon and Reddit on top, are where a specialty site competes worst. Narrower angles are where it can still be the best answer.

Pattern 4: Brand transformation

Some sites have transformed from anonymous content sites into branded properties with a named voice, a podcast, and a community. The content that once was just SEO-optimized articles becomes a richer offering.

The pattern: the sites that recover are often the sites that stop being SEO sites and start being publications.

What has not worked

• Adding a fake author bio (Google detects)
• Stamping recent dates on old content (Google's spam policy names this)
• Buying links (Penguin-era spam, still detected)
• AI-rewriting all content "to fix the AI patterns" (the rewrites are still AI)
• Disavowing links (rarely the issue for HCU)
• Migrating to a new domain (the patterns travel with you)

The HCU is not gameable in a meaningful way. Recovery requires structural improvement.

Appendix M: Glossary Extended

• A/B testing in content: comparing two versions of an article to see which ranks or converts better. Less common in SEO than in product engineering, but used by larger publishers.
• Black hat SEO: tactics that violate Google's guidelines. Examples: link buying, cloaking, hidden text.
• Canonical tag: HTML element specifying the preferred URL when content exists at multiple URLs.
• Content pruning: removing or noindexing low-quality content to improve sitewide signals. A common HCU recovery tactic.
• Crawl budget: the number of pages Google will crawl on a site per cycle. Important for very large sites.
• Domain authority: a third-party (Moz) score predicting ranking strength. Not a Google signal.
• Featured snippet: a special SERP feature showing a direct answer above organic results.
• Knowledge panel: an entity-specific information box on the SERP.
• Long-tail query: a low-volume, specific query. Long-tail traffic has historically been a small publisher's livelihood.
• Page experience: a Google ranking signal covering Core Web Vitals and other UX factors.
• Pillar page: the central article in a topical cluster.
• PBN (Private Blog Network): a network of sites used to manipulate links. Black hat.
• Rich result: a SERP feature with structured data (recipes, reviews, FAQs).
• SERP: Search Engine Results Page.
• Sitelinks: additional links beneath a search result, typically for branded queries.
• Voice search: queries via voice assistants. Different optimization patterns.
• White hat SEO: tactics that comply with Google's guidelines.

Appendix N: A Reading List for 2026 SEO

A short reading list for site owners who want to think about HCU and AI-era ranking carefully. These are publications and authors to follow; the figures in this article come only from the linked sources in section 16.

• Google Search Central blog: official source for update announcements
• Google Search Quality Rater Guidelines: the public document describing how Google's human raters evaluate quality
• Search Engine Land: industry coverage
• Search Engine Journal: industry coverage
• Marie Haynes: newsletter on ranking changes
• Lily Ray: posts on X about ranking volatility
• Glenn Gabe (G-Squared Interactive): posts on X and case studies of ranking drops
• Casey Newton's Platformer: long-form analysis of Google and AI policy
• Stratechery (Ben Thompson): long-form analysis of platform dynamics
• The New York Times tech section: ongoing coverage of AI and search
• Wired: ongoing coverage of AI and search
• The Verge: ongoing coverage of Google policy and AI
• MIT Technology Review: deeper takes on AI generally

For Sailop-specific anti-slop reading, see /blog/ai-slop-2026-state-of-the-ai-generated-web, /blog/anti-slop-prompt-template-2026, /blog/de-ai-your-lovable-v0-bolt-site, /blog/ai-slop-economy-2026, and /blog/detect-ai-generated-site-30-seconds-21-signs-2026.

Closing Note

The Helpful Content Update started as a small ranking adjustment in August 2022 and grew into the central organizing logic of how Google ranks the modern web. The war on AI slop is not a campaign with an end date. It is the new equilibrium. Sites that produce content with named authors, real expertise, original information, and transparent infrastructure have a path forward. Sites that produce content optimized only for crawlers, with hidden authors and recycled material, have been progressively de-ranked across four years of updates.

The 2026 question is no longer "will Google crack down on AI content?" That happened. The 2026 question is: how do you produce work that is recognizably worth reading, by humans and by the algorithms designed to surface what humans would want to read? That question has the same answer it had in 2002, before Google was Google: write things that are true, that you have learned, that you are accountable for, and that you would be willing to put your name on.

The technical infrastructure (schema.org, EEAT, anti-slop visual design, topical clusters) is the frame. The content inside the frame still has to be content worth framing.

Revision note

Updated on 18 October 2026. The opening figure was wrong: HouseFresh did not lose "roughly 90%" of its traffic in the two weeks of the September 2023 update; it reported a 91% loss after the March 2024 core update. That figure and Retro Dodo's 85% now come from the sites' own posts, linked. Removed: a "70-90%" range across three sites (we found no published figures for Travel Lemming, whose case is gone), "thousands" of affected publishers, "roughly fifteen" updates, a "$200-500" to "$0.10" cost comparison, two averaged ASCII curves of traffic and clicks with no dataset behind them, a "30-40%" recovery figure, a table rating content types from "very weak" to "very strong", "50+ articles per week" and a "15-25 words per sentence" range. The timeline now follows Google's Search Status Dashboard, which corrected several entries (there was no September 2025 core update, the 2025 spam update was in August, the June 2025 and May 2026 core updates were missing) and runs to September 2026. Added from primary sources: Google's 40% and 45% figures for the March 2024 update, the reach of AI Overviews as Google reported it, and Pew Research Center's click measurements.

Later the same day, every judgement about a named site or engine was checked for a dated source. Removed for lack of one: Condé Nast "disputing" AI Overviews, Stack Overflow's traffic decline, the New York Times' "positions" on AI Overviews, Quora's swings in quality, Substack and Stratechery, YouTube transcripts, Wikipedia and government sites rising, Amazon, Walmart, Best Buy and Target gaining, Bing being "more permissive" of AI content, the Bing, Google and Perplexity rating table, what Perplexity, Phind and the other answer engines "favor", the notes on Ahrefs, Semrush, Moz, Surfer and Frase, and the Cleveland Clinic example. The glue and rocks examples now come from Google's own post of 30 May 2024, which attributes the glue advice to discussion forums, not to Reddit by name. Added from dated sources: Reddit, Amazon and Quora as HouseFresh saw them in its results, Retro Dodo's description of the Reddit rise, DuckDuckGo's statement that its links come largely from Bing, Perplexity's documentation of its crawler, and the New York Times' robots.txt.