AI-Generated Content and SEO: What Google's Rules Actually Say

Google's position has two halves, and most confusion comes from reading only one.

The first half is permissive. Google's guidance on generative AI content says: "Appropriate use of AI or automation is not against our guidelines." Using a model to draft, outline, summarize or translate is not a violation on its own. Google does not ask you to label or hide it.

The second half is a spam policy. Google defines scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." Its first example is "Using generative AI tools or other similar tools to generate many pages without adding value for users."

Put together, the rule is simple. Google judges the pages, not the tool. A page written with AI that answers a real question well is fine. A thousand pages produced to capture search traffic, with nothing a reader couldn't get elsewhere, is the pattern the policy describes. The policy applies "no matter how it's created," so a human content farm sits under the same rule.

For a fuller reading of the documents, see Google's guidance on AI-generated content, digested. If your real question is whether you'll be punished for using AI at all, does Google penalize AI content covers that directly.

The policy turns on two words in Google's definition: "many" and "primary purpose." Neither has a published number. Google does not say how many pages is too many or how much value is enough. That is deliberate, and it means you can't game a threshold.

What you can do is ask the questions the definition implies:

• Why does this page exist? If the honest answer is "to rank for a keyword variant," that's the pattern Google names.
• What does it add? A page that restates what ten other pages say, in different words, adds little.
• Who checked it? Unreviewed output at volume is how wrong facts, duplicate pages and thin answers ship.
• Would you publish it if search didn't exist? Google's people-first content guidance is built around that question.

Volume alone is not the violation. A large catalogue of genuinely useful pages is normal. Volume without value is the problem, and AI makes that combination cheap. The trade-offs are laid out side by side in scaled AI content vs people-first content.

OtterlyAI ran a public experiment in 2026. It loaded two brand-new domains about capybaras with roughly 1,000 AI-generated posts each, published in one click. One site showed an author box presenting a fictional wildlife biologist. The other had no author.

The results, from Otterly's write-up and its September 2026 BrightonSEO deck:

Three things matter for anyone publishing AI content.

1. Removal was silent. Otterly reports zero manual actions on either domain, so the removal was algorithmic. Nothing in Search Console warned the owners. The difference between the two kinds of removal is explained in manual action vs algorithmic deindexing.
2. The fake expert did not help. The site with the invented biologist went first. This fits Google's statement that E-E-A-T "isn't a specific ranking factor." With one site per arm, it doesn't prove author boxes hurt. It is no evidence that they protect.
3. Bing behaved differently. Bing never deindexed either site, and Otterly found Microsoft Copilot produced 2,718 of the 3,934 AI citations it tracked for the fake-expert site.

Keep the limits in view. It is one pair of sites on a low-stakes topic, with no human-written control, run by a vendor that sells AI visibility tracking. It shows the policy working as written. It doesn't tell you where the threshold is for your site.

Some teams assume AI search is a way around Google's rules. The evidence points the other way for Google's own AI surfaces.

Google states that to appear as a supporting link in AI Overviews or AI Mode, a page "must be indexed and eligible to be shown in Google Search with a snippet." It also says there are no additional requirements or special optimizations for those features. So if a scaled content pattern gets your pages removed, you lose AI Overviews and AI Mode at the same moment. In Otterly's test, Google's AI surfaces almost never cited the sites once they were gone.

Other engines use other indexes. Copilot grounds on Bing, which is why it kept citing the capybara sites. That is not a strategy. Google is still the largest channel, and Copilot citations on a low-competition topic say little about a commercial category. In the same test, ranking in Bing produced effectively zero ChatGPT citations, even with ChatGPT's crawlers visiting.

One more point: Google's AI optimization guide says "You don't need to write in a specific way just for generative AI search." There's no AI-friendly style that earns citations, and the question of whether engines favour machine-written text is covered in do AI engines prefer citing content written by AI.

Split the problem into stages and the picture gets clearer.

In your control

• Whether each page adds something. Original data, a real price, a worked answer to a specific question, first-hand detail. Stating a specific fact on a fetchable page gives an engine a sentence it can quote.
• How many pages you publish and how fast. A thousand pages overnight on a new domain is the exact pattern in Otterly's test.
• Review. Who checks facts, removes duplicates and kills pages that don't earn their place.
• Access. Whether crawlers can fetch the page, whether key content is in the HTML before JavaScript runs, and whether noindex or nosnippet is set by accident.

Not in your control

• Google's classifiers. Google doesn't publish thresholds or say which system removed a site.
• Whether an engine retrieves or cites you. Being indexed buys eligibility, not inclusion. Google says indexing and serving "isn't guaranteed."
• Warnings. An algorithmic removal may arrive with no message at all.

Nobody can promise rankings or citations for AI-assisted content. What you can do is remove the reasons a page would be judged spam or be unreachable.

If you use AI in your content process, work in this order.

1. Audit intent before output. List the pages you plan to publish and the reader question each one answers. Cut any page whose only purpose is a keyword variant. Google's AI optimization guide says you don't need to capture every long-tail variation.
2. Use AI for drafting, not deciding. Let a model draft and restructure. Keep a person responsible for what the page claims and why it exists.
3. Add something only you have. Your prices, your process, your data, your customers' real questions. If a page could appear on any competitor's site unchanged, it adds little.
4. Fact-check specifics. AI drafts invent numbers, features and dates. A wrong price on your own site can be repeated by AI answers about your brand.
5. Publish at a pace you can review. There's no safe published rate. A pace your team can actually check is the honest limit.
6. Check indexing engine by engine. Watch Google Search Console and Bing Webmaster Tools separately. One being healthy tells you nothing about the other. A sharp drop in impressions with no manual action is the signal to investigate.
7. Confirm crawler access. A robots.txt that allows a crawler while your CDN or firewall returns a 403 removes you quietly. Test robots.txt rules for each AI bot, then confirm the firewall side with the AI Crawler Checker, which requests a page as each AI crawler.

Once the pages are indexed and reachable, the free checker on this site shows whether AI engines name you for up to three buyer questions. Treat it as a snapshot.

• Treating volume as the strategy. Publishing hundreds of near-identical pages is the behaviour the spam policy names first. Where AI-generated SEO at scale breaks goes into the failure points.
• Inventing experts. A fictional author misleads readers. In Otterly's test it didn't stop removal, and E-E-A-T isn't a score you can raise with a byline.
• Rewriting for an "AI style." Google says you don't need to write in a specific way for generative AI search. Chunking and style rewrites are not levers.
• Adding schema or llms.txt to compensate. Google says structured data isn't required for generative AI search, and Google Search ignores llms.txt. Neither fixes thin pages.
• Hiding text or instructions for AI crawlers. Hidden text, prompt injection and cloaking are manipulation. They're a risk to your site and your reputation, and they shouldn't be used.
• Reading one engine as all engines. Copilot citing you while Google has dropped you is not a recovery. Measure each one separately.
• Waiting for a warning. Algorithmic removals can be silent. Monitor impressions, not just the manual actions report.