Scaled Content Abuse: When AI-Assisted Pages Cross Google’s Line
Using a model to help write a page is not the violation. Many pages whose primary purpose is ranking manipulation, with little or no value to a person, are. Decide before you publish the pile.
You used a model to research a page, or to put a draft in order, and a person still gets something useful from the finished URL. That is assistance. The line you can cross is different: many pages, created so they will rank, that do not help a user, whether a model wrote them, a script stitched them, or someone pasted them by hand.
This is a decision, not a tool review. Impressions in AI Overviews and AI Mode are the previous page, how to measure generative AI visibility in Search Console. Whether the page deserves to exist after you decide it is allowed is the information gain framework. Do not skip the policy and jump to the brief.
The policy names purpose and value, not the model
Sourced from Google. Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users. The practice is typically large amounts of unoriginal content that provides little to no value, no matter how it is created. If you host that content, exclude it from Search. Source: Spam policies for Google web search (Search Central, scaled content abuse section, page last updated 2026-08-28 UTC, checked 28 September 2026).
“No matter how it is created” is the sentence that stops two bad readings. A human clicking publish does not clear a pile of worthless pages. A model in the draft history does not condemn a page that helps someone. The primary purpose and the lack of value are what the policy describes. Assistance is not what it names.
- Generated at scale without value: Using generative AI tools, or similar tools, to generate many pages without adding value for users.
- Scraped and obfuscated: Scraping feeds, search results, or other content to generate many pages, including automated transformations such as synonymizing, translating, or other obfuscation, where little value is provided.
- Stitched: Stitching or combining content from different web pages without adding value.
- Split across sites: Creating multiple sites with the intent of hiding the scaled nature of the content.
- Keyword shells: Creating many pages where the content makes little or no sense to a reader but contains search keywords.
The page says those are examples, not a closed list. If a method is missing from the bullets and the primary purpose is still ranking manipulation, with little or no value to users, you are in the definition, not in a loophole. This article does not attach a percentage, a “strike,” or a recovery timeline. The pages used here do not give those numbers.
What Google says AI assistance can be
Google’s guidance on generative AI content says the tools can be particularly useful when researching a topic, and for adding structure to original content. The next sentence is the limit: using generative AI tools or similar tools to generate many pages without adding value for users may violate the scaled content abuse policy. If you use generative AI content, the work has to meet Search Essentials and the spam policies. Source: Google Search’s guidance on using generative AI content on your website (Search Central, page last updated 2025-12-10 UTC, checked 28 September 2026).
When you create for the web, that page says to focus on accuracy, quality, and relevance, especially if generation is automatic. That includes the title element, the meta description, structured data, and image alt text, because those can appear in Search results. Structured data still has to follow the general guidelines, the rules for the specific feature, and validation if you want eligibility for those features. None of that is a substitute for a page a person needed.
Sharing how a piece of content was created can help give readers context, including how automation was used. That is a reader note. The page does not say a disclosure replaces value, or that labeling a page “AI-assisted” moves it out of the spam policy. For ecommerce, the same page points at Merchant Center policies: AI-generated images must carry IPTC DigitalSourceType metadata of TrainedAlgorithmicMedia, and AI-generated product title and description attributes must be specified separately and labeled as AI-generated. That is a product-data rule. It is not a ranking shield for a city-page generator.
The same guidance mentions Search Quality Rater guidelines, sections 4.6.5 and 4.6.6, for how raters evaluate scaled content abuse and main content made with little effort, little originality, and little added value. It also says those guidelines are not a guide to ranking first. Raters help evaluate search systems. Their ratings do not directly influence ranking. Do not build a checklist whose only goal is to sound like the rater PDF.
Fan-out variants are the temptation this year
Generative Engine Optimization (GEO) is a label for work aimed at AI search. Google’s optimization guide says optimizing for generative AI search, including AI Overviews and AI Mode, is still SEO. It is not a license to stamp out a URL per query variant. The guide says it can be tempting to create separate content for every variation of how people might search, including other questions people asked, or fan-out queries. Doing that primarily to manipulate rankings or generative AI responses violates the scaled content abuse policy. A high quantity of pages does not make a site higher quality or more relevant. Google also says its systems can understand relevance even when the query and the page’s main content are not an exact match, so you do not have to capture every long-tail wording. Source: Optimizing your website for generative AI features on Google Search (Search Central, page last updated 2026-07-10 UTC, checked 28 September 2026).
Query fan-out, on that page, is something Google’s systems do: concurrent related queries to fetch more results. It is not a brief you are required to fill with one URL each. Retrieval-augmented generation (RAG), also called grounding, is also described there as retrieval from the Search index. This page does not turn either term into a build tutorial. The policy point is enough. Manufacturing pages so a fan-out query has an exact URL is the behavior the guide ties to scaled content abuse.
A decision you can apply to the next URL
The table is a framework in operator language. It is not a matrix Google published. Each row points at a sentence on the pages above. If you cannot answer the value row with something a specific person uses, stop.
| Question | Assistance can stand | Stop |
|---|---|---|
| Why does this URL exist? | A person needs this page. The model helped you research or structure original work. | The primary purpose is to manipulate rankings or generative AI responses. |
| What did the tool add? | Research, an outline, or a draft you corrected for accuracy, quality, and relevance. | Many pages, without value added for users. |
| Could you explain it without the keyword list? | The page makes sense to a reader and holds a point of view or experience that is not a restatement. | The text makes little or no sense and exists to hold search keywords, or it is scraped text run through synonyms or translation. |
| How many siblings are you about to ship? | The pages the offer actually has. One solid page can be relevant without an exact query match. | A page per fan-out variant, city, or synonym, or several sites so the scale is harder to see. Stitched copies with nothing new. |
| It is already live. Now what? | It meets Search Essentials and the spam policies, and it helps a user. | You are hosting scaled content abuse. Exclude it from Search. |
If you are hosting that content, the spam page says to exclude it from Search, and it links the control-what-you-share document for how. Follow that document. Leaving the URLs up because they might earn impressions is still hosting the pages the policy tells you to exclude.
The impression report will not bless the pile. It counts shown links in AI Overviews and AI Mode. It does not score whether a page helped anyone. If you need a method for what a real page contains, use the information gain framework after this decision, and the search intent brief when you are writing one URL on purpose. The beginner SEO guide is the vocabulary. The complete SEO guide is the map. None of them are a waiver.
Commodity pages and the people-first test
The AI optimization guide separates commodity content from a page with a point of view. A generic tips list that anyone could write, or that a model could produce from common knowledge, is the example of commodity. A specific account that goes past common knowledge is the contrast. The guide also says unique, useful content is more likely to matter for generative AI search over time than the other tactics in that document. Write for the person. Organize with headings they can follow. Add images or video when they actually help, under the ordinary image and video docs, not as a GEO trick.
There is a one-question check on that guide: would visitors find this satisfying? If you are unsure, ask that before you ask whether a fan-out query has a URL. The technical side still applies. A page has to be indexed and eligible for a snippet, and the site has to be included in Search generative AI features, before it can show there. Meeting the rules does not guarantee indexing. A page farm that is technically crawlable is still a page farm.
Ignore the side quests the same guide lists. Google Search ignores llms.txt. Publishing it for some other consumer neither harms nor helps Google Search. Chunking a page into tiny pieces is not required. Rewriting in a special AI dialect is not required. Inauthentic mentions are not the plan. Special schema just for AI Overviews is not a thing Google asks for. Structured data still matters for rich results. Skipping those hacks does not, by itself, make a thin page acceptable. The policy is about the pages you did publish.
Filled hypothetical: a page per nearby town
Hypothetical, labeled as such. A service business asks a model for one page per town, per service, and per “near me” wording. The paragraphs swap the place name. A few lines are translated synonyms of the same offer. The founder calls it GEO, because a tool said fan-out queries need coverage. No person in those towns gets a detail that was not already on the main service page. The primary purpose is to hold rankings and AI answers.
That shape matches the policy: many pages, little value, ranking as the point, generative tools as the method. It also matches the fan-out warning. The fix is not a footer that says a model drafted the text. The fix is to stop. Exclude the URLs you already host if they are that pattern. Keep or write the pages where someone in that place gets something the main page does not have: a real constraint, a real process, a real answer. Then measure impressions, if the report exists, without pretending impressions proved the pages were allowed.
What this does not prove
The docs used here do not say that any use of a model is spam, and they do not say that a human editor makes scale safe. They do not publish a percentage of pages you may generate. They do not say a rater score is your rank. They do say that many pages for manipulation, without user value, are abuse, that hosting them means you should exclude them from Search, and that a fan-out URL factory can be that abuse. Original research and structure are described as useful. Useful still has to meet Search Essentials and the spam policies.
If the site is about to grow a hundred URLs that fail the table, do not build them as a website project and hope the policy is a tone problem. Start a conversation with the list of URLs and a one-line reason each one helps a person. If the reason is the query variant, you already have the decision.
Frequently asked questions
Is using generative AI to write a page a spam violation?
Not by itself. Google’s spam policy defines scaled content abuse as many pages generated for the primary purpose of manipulating rankings and not helping users, with little or no value, no matter how they were created. Google’s generative AI guidance says the tools can help you research a topic and add structure to original content, and that generating many pages without adding value may violate that policy. The work still has to meet Search Essentials and the spam policies.
What does Google list as scaled content abuse?
Examples include generative AI or similar tools used to generate many pages without adding value, scraping then synonymizing or translating or otherwise obfuscating, stitching pages together without adding value, multiple sites meant to hide the scale, and many pages that make little sense to a reader but contain search keywords. The list says it is not exhaustive. The test is purpose and value, not the brand of the tool.
If those pages are already on the site, what does Google say to do?
If you are hosting that content, exclude it from Search. That is the instruction on the spam policies page, with a link to the control-what-you-share document. This article does not add a removal percentage or a timeline. Those numbers are not on the pages used here.
Should I publish a page for every fan-out query?
No, if the primary purpose is to manipulate rankings or generative AI responses. Google’s AI optimization guide says creating separate pages for every variation, including fan-out queries, primarily for that purpose violates the scaled content abuse policy. A large number of pages does not make a site higher quality or more relevant. Generative Engine Optimization (GEO) is not a reason to do it. Google treats optimizing for AI Overviews and AI Mode as SEO.
Do the Search Quality Rater guidelines set my ranking?
Google’s generative AI content page points at rater guideline sections on scaled content abuse and on main content with little effort, originality, or added value. The same page says those guidelines are not a guide to ranking first, and that rater ratings do not directly influence ranking. They are used to evaluate how search systems perform. Do not treat the PDF as a score you can hit.