SEO but for AI: What the New Discipline Really Is

When someone searches for "SEO but for AI", they usually want to know if there's a new rulebook. There is, partly. The goal hasn't changed: be the source an engine uses when a buyer asks a question. The route to that goal has more gates.

Much of the GEO market sells techniques: rewrite pages in a quotable style, add FAQ schema, publish an llms.txt. The evidence doesn't back them. A July 2026 critical survey of 45 GEO studies (arXiv 2607.14035) concluded that already-retrieved content can change how it gets cited, but no reviewed technique shows a stable, cross-platform effect on being discovered in the first place.

That leaves a smaller but more useful definition. SEO for AI is classic technical SEO plus a new layer of access plumbing, plus plain, specific facts on pages a crawler can actually read. If you've been wondering whether SEO is dead, the short answer is no. It now has more doors to keep unlocked.

An AI answer passes through four stages. Each has a different owner.

1. Parametric. This is what the model knows with no web access, frozen at a training cutoff. Nobody can change it this quarter.
2. Access. This decides whether a crawler may fetch and store your page. You and your CDN control it.
3. Retrieval. This decides whether your page enters the model's context for a given question. The engine controls it.
4. Generation. This decides which retrieved pages the model uses and cites. The model controls it.

Retrieval doesn't fire every time. As of October 2026, ChatGPT decides per question whether to search, Claude.ai search is a user toggle, and Google AI Overviews often don't appear at all. When no search runs, the answer comes from parametric memory. OpenAI's model docs list a February 16, 2026 knowledge cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path. That's why AI often repeats your old news.

What it means: stage 1 is the only stage fully in your hands, and stages 2 and 3 depend on it. Start there.

Four things differ from the SEO most teams learned.

Per-bot permissions

Each AI company runs several crawlers with separate jobs. OpenAI's crawler page says GPTBot is for training, OAI-SearchBot is for ChatGPT search, and "each setting is independent of the others." A site can opt out of training and stay in search:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

Anthropic and Perplexity split their bots the same way. The block training, keep search playbook covers the full set.

Multiple indexes

Google states that a page must be indexed and eligible to show with a snippet to appear as a supporting link in AI Overviews or AI Mode. Microsoft Copilot grounds on Bing. These are separate indexes with separate health. In Otterly's 2026 experiment, Google deindexed two AI-generated test sites while Bing kept them, and Copilot produced 2,718 of the 3,934 citations Otterly tracked for one site. If you're indexed in Google but missing from AI Overviews, check each index on its own.

A network gate

OpenAI's help centre says eligibility for ChatGPT search means allowing OAI-SearchBot and confirming your host or CDN allows OpenAI's published IP addresses. Both halves matter. A firewall can return 403 to a bot that robots.txt allows. Cloudflare also changed its defaults on September 15, 2026, setting Training and Agent crawlers to blocked on ad-displaying pages for new domains and free-tier customers. If your robots.txt allows AI bots but you never show up in ChatGPT, this gate is the first suspect.

No rank data

Google's Search Console report for generative AI features shows impressions only: no clicks, no click-through rate, no position. OpenAI says only that ChatGPT ranks results using "multiple factors." No ground truth exists for any AI position number.

Plenty of classic SEO still applies. Indexability matters, because Google's AI features draw from its normal index with "no additional technical requirements." Snippet controls still work: noindex, nosnippet, max-snippet:0 and data-nosnippet keep content out of AI answers on Google and Bing. Ranking still overlaps with citation. Ahrefs found that 37.9% of URLs cited in AI Overviews also appeared in the top 10 results in its March 2026 study.

So strong SEO helps but doesn't guarantee AI visibility. Some popular advice doesn't carry over at all:

• 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. Don't rip existing markup out, though. Google says it does no harm. It just isn't a lever for AI answers.
• llms.txt. Google says Search ignores it, and no other engine documents reading it.
• E-E-A-T scores and domain authority. Google says E-E-A-T isn't a specific ranking factor. Domain Authority is a Moz metric no engine computes.
• Chunking and "AI-friendly" writing. Google's AI optimization guide rejects both and says there's no ideal page length.

What survives on the content side is narrow: a specific fact (a price, a number, a one-line definition) on a page a crawler can fetch gives the engine a sentence it can quote.

Be clear with your team about where effort pays off.

In your control:

• Whether your key content is in the HTML before JavaScript runs.
• Which bots your robots.txt allows, and whether your firewall agrees.
• Whether your pages are indexed and snippet-eligible in Google and Bing.
• Whether your facts are stated plainly and consistently across your own site.
• Whether you're named on the third-party sites engines already cite. The evidence shows an association here, not a promised effect.

Out of your control:

• What the model learned before its cutoff.
• Whether a given question triggers a search. OpenAI says ChatGPT rewrites the user's question into one or more targeted queries, which you can't see.
• Where your page lands in the model's context, and which retrieved sources it cites. Engines retrieve more pages than they cite, which is why cited and retrieved sources are different measures.

This is the core difference between answer engines and search engines: you can earn eligibility, but you can't earn a slot.

Work from the gates inward. Each step removes a possible block before you spend time on anything else.

1. Check rendering. Fetch your key pages with JavaScript off and compare the text with the rendered page. Pricing, product names and definitions that appear only after hydration are likely invisible to most AI crawlers.
2. Read your robots.txt per bot. Confirm OAI-SearchBot, PerplexityBot, Claude-SearchBot and Googlebot are allowed. Decide on training bots separately.
3. Test the network gate. Request a page as each AI crawler and record the status code. The AI Crawler Checker shows which crawlers your firewall turns away.
4. Check each index. Use Google Search Console and Bing Webmaster Tools separately. In Search Console, confirm the generative AI inclusion control hasn't been turned off. It defaults to included, and child properties inherit the parent's setting.
5. Hunt accidental suppression. Search templates for data-nosnippet, nosnippet and max-snippet:0 applied by mistake.
6. State your facts plainly. Put price, plan limits and a one-line description on fetchable pages. Make them match everywhere on your site.
7. Ask the engines, one at a time. Record what each engine says about you and score each one separately. The free checker on this site gives a one-time snapshot for three buyer questions.
8. Get named where engines look. If engines cite review sites or directories instead of you, being listed there is the route into the answer.

• A blanket AI block. A single rule that blocks every AI user agent also removes OAI-SearchBot, which takes you out of ChatGPT search.
• Allowing GPTBot to get found. GPTBot governs training only. Search uses OAI-SearchBot.
• Trusting robots.txt alone. A CDN or firewall can block a bot that robots.txt allows, and nothing in your dashboards will flag it.
• One blended score. Engines draw on different indexes and behave differently. In Otterly's experiment, Copilot cited the test sites heavily while ChatGPT almost never did.
• Paying for "AI rank". No engine exposes position data, so the number can't be verified.
• Mass publishing. Otterly's two 1,000-post AI sites were both removed from Google, with no manual action reported. Google's spam policies name scaled content abuse directly.
• Bumping dates. Changing dateModified without changing content has no evidence behind it.
• Hidden text or prompt injection. Text written to steer AI crawlers, or cloaked content, is illegitimate. It breaks engine policies and puts your reputation at risk. Don't do it, and watch for third-party pages that try it against your brand.

If engines recommend your competitors but never you, work through the order above before changing any copy.