Generative AI and SEO: What Actually Changed

Two things changed at once. Generative AI made content cheap to produce. Generative answers became a new place where buyers meet brands, often without clicking a blue link.

The first change is a production question. The second is a visibility question. Most confusion about generative AI SEO comes from mixing them up.

What stayed the same matters more than the hype suggests. Google's AI Overviews and AI Mode draw on the same index as classic Search. Google's documentation says a page must be "indexed and eligible to be shown in Google Search with a snippet" to appear as a supporting link, and that "there are no additional technical requirements." There is no separate back door into Google's AI features.

Other engines run their own crawlers and indexes, with their own rules. That is where the real new work sits. If you want the wider frame first, read generative engine optimization, explained honestly.

The next move: treat the two halves separately. Decide how you use AI to write, then decide how you make sure AI engines can reach what you wrote.

Google's guidance on AI content is direct: "Appropriate use of AI or automation is not against our guidelines." The tool is not the problem. Volume without value is.

Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings and not helping users." The first example given is using generative AI tools to generate many pages without adding value for users. The policy applies no matter how the content is created.

What a public test showed

Otterly ran a public experiment in 2026 with roughly 2,000 AI-generated posts across two brand-new domains. Google removed both from its index, one after about 14 days and the other after about 90, according to Otterly's figures. Otterly reports zero manual actions in Search Console on either domain, so the removal arrived without a warning.

The site with a fictional expert author box was the one that went first. That is one data point, not proof that author boxes hurt. It is also no evidence that they help.

Otterly sells AI visibility tracking, and the test had one site per arm and no human-written control. Read it as a strong illustration of Google's written policy, not as a controlled study.

What to do with this

• Use AI to draft, research and edit. Publish what a human has checked and improved.
• Do not publish pages at a scale your team could never review.
• Say you run a dental clinic in Austin. Ten pages answering real patient questions with your own prices and procedures beat a thousand generated city pages.

Every AI answer passes through four stages. Knowing which stage you are dealing with tells you whether you can do anything about it.

1. Parametric. What the model knows with no web access, frozen at a training cutoff. Nothing you publish this quarter reaches it.
2. Access. Whether a crawler is allowed and able to fetch and store your page. This is publisher-controlled.
3. Retrieval. Whether your page enters the model's context for a given query. The engine decides, often after rewriting the user's question into its own searches.
4. Generation. Which retrieved pages the model uses and cites. The model decides.

Retrieval is discretionary on every platform. Google's AI Overviews often do not trigger. ChatGPT searches at the model's discretion. Claude.ai search is a user toggle. So any single answer you see may come from live retrieval or from the model's memory.

This matters because nearly every published GEO technique was tested at stage four, on pages already placed inside the model's context. A 2026 critical survey of GEO research found no technique with a stable, cross-platform causal effect on whether pages get discovered in the first place. The gap between those lab results and real engines is covered in lab GEO studies vs live AI engines.

The next move: put your effort into stage two, where you have real control.

Here is the honest split, as of October 2026.

The controllable rows share one trait: they remove blocks. None of them makes a page more attractive to a model. They make sure the page exists for the model at all.

The last row is the closest thing to an additive route. When engines cite other sites instead of yours, being named on those sites is associated with appearing in the answer. Treat it as an association, not a promised effect.

For the full breakdown of influence, see what you can and can't influence in LLM optimization.

Work from the outside in. Each step only helps if the one before it passes.

1. Set robots.txt per bot

robots.txt rules are per bot, and the bots do different jobs. OpenAI's crawler documentation says GPTBot governs training, OAI-SearchBot governs ChatGPT search, and "each setting is independent of the others." A common error is a blanket AI block that silently removes ChatGPT search eligibility.

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

That example allows ChatGPT search and opts out of training. Check PerplexityBot, Claude-SearchBot and Meta-WebIndexer the same way. There is no robots.txt token for Google's AI Overviews or AI Mode; Googlebot rules cover all Google surfaces.

2. Check the firewall and CDN

A firewall or CDN can block a crawler that robots.txt allows. OpenAI's help centre says eligibility requires allowing OAI-SearchBot and confirming your host or CDN allows its published IP addresses. Since 15 September 2026, Cloudflare blocks Training and Agent crawlers by default on ad-displaying pages for new domains and existing free-tier customers, while Search stays allowed. You can test this with the AI Crawler Checker, which requests a page as each AI crawler and shows which ones get turned away.

3. Server-render the facts that matter

Vercel's crawler study found ChatGPT and Claude crawlers fetch JavaScript files (11.50% and 23.84% of requests) but do not execute them. Content that only appears after JavaScript runs may be invisible to them. Googlebot renders JavaScript, so this gap mostly affects non-Google engines. Put prices, product names and key claims in the initial HTML.

4. Confirm index and snippet eligibility

noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers. Templates often inherit them by accident. Check Google Search Console and Bing Webmaster Tools separately, because one healthy index tells you nothing about the other.

5. State specific facts plainly

A price, a number or a one-line definition on a fetchable page gives an engine a sentence it can quote. Make sure the same fact reads the same on every page.

A structured way to run these checks is laid out in the free GEO audit guide.

Most of these come from old blog posts and vendor pages. The primary sources contradict them.

• Adding schema to get cited. Google says structured data isn't required for generative AI search. Ahrefs' matched-control test found AI Overview citations fell 4.6% after pages added schema. FAQ rich results were deprecated in May 2026. Existing markup does no harm, so there is no need to remove it. It just is not a lever for AI answers.
• Publishing an llms.txt. Google says Search ignores it, and no other engine documents reading it.
• Chasing an E-E-A-T score. Google says E-E-A-T isn't a specific ranking factor. There is no score to improve.
• Building domain authority. Domain Authority is a Moz metric. No engine computes it.
• Allowing GPTBot so ChatGPT finds you. GPTBot governs training only. ChatGPT search uses OAI-SearchBot.
• Rewriting in an AI-friendly style or chunking pages. Google rejects both and says there is no ideal page length. C-SEO Bench found most citation-oriented rewriting methods ineffective and often harmful.
• Trusting the 40% uplift claim. It came from a lab engine with sources preloaded into context.
• Mass-publishing AI pages. Google's scaled content abuse policy targets exactly this, and removal can arrive silently.
• Hidden text and prompt injection. Some pages hide instructions aimed at AI systems. This is manipulation, it exposes the brand to spam action and reputational damage, and it is never a legitimate tactic.

For the evidence side by side, compare on-page rewrites vs access fixes. If you are weighing whether the whole field holds up, read whether GEO is legit or hype.

Measure each engine on its own. In Otterly's test, Microsoft Copilot produced 2,718 of the 3,934 citations tracked for one site, while ChatGPT, Gemini and Google's AI features almost never cited it. A blended score would have hidden that entirely.

Do not track an AI rank. Google's generative AI report in Search Console shows impressions only, with no position data, so no position number can be checked against anything.

Classic rankings still matter, but they are not the whole story. Ahrefs found that 37.9% of URLs cited in AI Overviews also appeared in the top 10 results, in its March 2026 study. The rest came from deeper positions.

Look for wrong facts, not just mentions. AI platforms get specific facts wrong about specific brands, such as old prices or retired product names. A wrong answer is often a bigger problem than a missing one.

A reasonable first step is a snapshot: the free checker on this site shows whether AI engines name you for up to three buyer questions. It is a point-in-time check, not continuous tracking. Then fix what blocks you and check again.

Does Google penalize AI-generated content? Not for being AI-generated. Google says appropriate use of AI or automation is not against its guidelines. It acts against scaled content abuse: many pages produced mainly to manipulate rankings without helping users. In Otterly's 2026 experiment, Google removed two new sites carrying about 1,000 AI posts each, with no manual action reported. The lesson is to publish only what adds value, whatever tool drafted it.

Do I need a separate strategy for generative AI search? For Google, mostly no. Google says there are no special optimizations needed for AI Overviews or AI Mode beyond being indexed and snippet-eligible. For ChatGPT, Perplexity, Claude and Meta AI, you need extra access checks: per-bot robots.txt rules, firewall reachability for each crawler, and content in the raw HTML. That is added hygiene, not a new content strategy.

Should I block GPTBot? That depends on whether you want your content used for OpenAI model training. Blocking GPTBot stops training use. It does not remove you from ChatGPT search, which uses OAI-SearchBot. OpenAI states each setting is independent, so you can block GPTBot and still allow OAI-SearchBot. Decide each bot on its own job, not as a single AI on or off switch.

Does ranking in Bing get me into ChatGPT? Not reliably. Bing's index backs Microsoft Copilot, and ChatGPT search sometimes uses other search providers, but nobody has measured that path end to end. In Otterly's test, sites that kept ranking in Bing earned heavy Copilot citations and effectively zero ChatGPT citations, even with ChatGPT's crawlers visiting. Being in Bing's index is worth checking. It is not a shortcut into ChatGPT.

How long do access fixes take to show up in AI answers? Nobody can promise a timeline. Answers built with live retrieval can reflect a fix once the engine recrawls your page. Answers from the model's memory only change with a new model, and OpenAI lists a February 16, 2026 knowledge cutoff for its GPT-5.6 family. Fixing access makes you eligible. Whether an engine retrieves and cites you remains its decision.