How to do AI SEO for Google and ChatGPT (2026)

Last verified · 2026-11-03 · by Moe Ameen

The steps

1. Baseline how each engine sees you today. Write down the real, conversational questions your audience asks in your niche — full sentences with follow-ups, not two-word keywords — then run each one in Google’s AI Mode and in ChatGPT. Record three things per question: whether you appear, whether you are cited by name, and which facts each engine gets right or wrong about you. That gap list is both your competitive map (who owns each answer now) and your scorecard. If you are absent everywhere, that is the whole opportunity; wrong facts tell you what to fix first.
2. Fix access for both engines before anything else. No optimization matters if the engines cannot read you. For Google, confirm the page is indexable and the answer is in the server-rendered HTML, not hidden behind client-side JavaScript. For ChatGPT, confirm OAI-SearchBot is allowed in robots.txt — OpenAI says sites opted out of it will not appear in ChatGPT’s search answers, and it is a separate decision from the GPTBot training crawler and the ChatGPT-User on-demand fetch. Keep a presence in Bing’s index too, since ChatGPT draws on it. Access is the gate everything else sits behind.
3. Reconstruct the real question, not the keyword. People talk to these engines in full questions, and both often fan one query into several sub-questions before composing an answer. So target the question a person would actually type or speak — and its obvious follow-ups — rather than a compressed keyword. For each priority topic, write the primary question and the three or four follow-ups it triggers; that set becomes the outline of the page. A page built around a conversation gives the model clean passages to quote; one built around a keyword gives it nothing to lift.
4. Write the page as a direct answer with first-hand proof. Answer the question in the first paragraph — a tight, self-contained 40-to-80-word answer the engine can lift whole — then earn the rest of the page with proof only a practitioner has: real data, a demonstration, an opinion, specifics. Both engines reward first-hand expertise over commodity rephrasing and discount thin, generic content. Include honest trade-offs; the pages that get cited read like an expert settling the question, not a brand selling into it.
5. Add the structure the engines can read. Make the answer machine-legible: a clear H1 that states the question, descriptive H2s for each sub-question, short scannable passages, and an FAQ block for the follow-ups. Add the structured data that fits — Article, FAQPage, and for products Product schema — so the engines read your facts as data instead of inferring them from prose. Google’s guidance treats schema as helpful but not a special AI requirement, so do it cleanly and move on; structure serves retrieval, it does not replace a good answer.
6. Make yourself one consistent entity across surfaces. Both engines are entity-aware: they surface brands and authors they can identify and verify. So your name, your core claims, your specs, and your point of view must be identical across your site, your social profiles, your video, and your author bios — drift reads as noise and makes the model hedge. Pin the canonical facts once and enforce them everywhere. A smaller but coherent presence out-cites a larger, contradictory one, which is why consistency is a production discipline, not a one-page edit.
7. Corroborate the claims off your own site. Neither engine trusts a single self-interested page, and ChatGPT in particular leans heavily on third-party sources — multiple 2026 analyses find most of its citations point somewhere other than the brand’s own site. So publish native versions of the same accurate claims across the surfaces the engines read: social, video, community discussion, and earned independent coverage. The same fact appearing in agreement across several trusted places is what moves you from ‘a page that asserts it’ to ‘the answer the engine is confident naming.’
8. Measure with the three signals that actually exist, then iterate. First-party AI reporting is thin — Google’s Search Console AI report exists but Google itself has called it inadequate on clicks, position, and queries, and ChatGPT offers none — so triangulate. Run your target questions in both engines on a schedule and log appearance and accuracy; watch server logs for OAI-SearchBot and Google crawler hits on your key pages; and watch brand-search and referral lift, since an AI mention often shows up as people searching your name afterward. Re-baseline monthly and feed the gaps back into the next pages. For the full method, see [run a GEO content audit](/how-to/run-a-geo-content-audit).

Common gotchas

• Treating AI SEO as a separate discipline from SEO for Google. Google’s own guidance says its AI features run on its existing ranking systems — there is no second optimization to run, just a higher bar on originality and access.
• Optimizing ‘for AI’ generically. Google reads the pages it already indexes; ChatGPT crawls with OAI-SearchBot and leans on Bing and third-party sources. One checklist for both leaves the ChatGPT-specific access and corroboration work undone.
• Burying the answer. A 400-word runway before you answer the question gives the engine nothing clean to lift; lead with a tight, self-contained answer.
• Blocking the wrong crawler — or the right one by accident. Opting out of OAI-SearchBot removes you from ChatGPT’s search answers; it is a different decision from blocking GPTBot for training.
• Chasing keywords instead of questions. These engines answer conversational, follow-up-laden queries; a page built around a two-word keyword rarely matches one.
• Letting your facts drift across surfaces. Inconsistent names, claims, or numbers make an entity-aware engine hedge instead of cite you.
• Publishing thin content at volume with no independent corroboration. Both engines discount it, and Google’s spam systems target exactly this pattern.
• Trusting one metric. No single report sees AI visibility cleanly in 2026 — triangulate prompt testing, logs, and brand-search lift.

Where Kompozy fits

Re-read the eight steps and notice where the time actually goes. Steps two through five — access, the real question, the direct answer, the structure — are a desk job you do once per page. The work that decides whether the engines cite you is the last three: being one consistent entity, corroborated, across every surface both engines read, for every question your audience asks — and that work scales linearly with your ambition. Twenty questions worth owning is twenty answer pages plus their native corroboration on eight-plus surfaces, done consistently, forever. Hand-built, that wall is where almost every AI-SEO plan stalls. It is the wall [Kompozy](/) is built to knock down.

Kompozy is a full AI content generation and multi-platform publishing engine, not a repurposing tool, and it runs that production line. For the citable half, it writes [Blog Articles](/glossary/output-buckets) shaped as the direct-answer pages both engines lift from — the question answered first, proof underneath. For the corroboration half, it turns each answer into the native set the engines cross-check: [Clipped Shorts](/glossary/clipped-short) and avatar [Persona Shorts](/glossary/persona-shorts) that demonstrate the claim, brand-exact carousels via [HyperFrames](/glossary/hyperframes), quote graphics, and photo posts — 18 formats, every one held to a single [Persona Brief](/glossary/persona-brief) so your name, claims, and numbers stay identical across all of them, which is the entity consistency step six demands. A per-post review gate keeps a human on an invented number before it ships, so the fact an engine quotes from you is true. Then [Autopilot](/glossary/autopilot) schedules and publishes the matching claims natively across the eight social platforms plus blog and email from one queue — the cross-surface corroboration of step seven, running on a cadence instead of by hand. Kompozy does not configure your robots.txt, write your schema, or track your AI rankings — you own the plumbing and the measurement; it owns the supply. Paid plans start at the Starter tier and scale to Pro and Enterprise as the question list grows.

Frequently asked questions

Related tutorials

• How to optimize product content for AI shopping on Google and ChatGPT (2026)
• How to do generative engine optimization (GEO) and get cited by AI answers (2026)
• How to get recommended by ChatGPT and AI Overviews (2026)
• How to create content for Google AI Mode and AI search (2026)
• How to build an AEO content workflow (2026)

← All how-to guides · Get Started