GEO vs SEO vs AEO: What Is Different and What to Fix First
Written by Usama Afzal, founder and director · Published
SEO, AEO and GEO are three names for overlapping work, and for a small site the order of fixes matters more than the label. Make sure crawlers can reach and index your pages, write answers that stand on their own, make who you are easy to verify, and only then worry about anything AI-specific. No AI product documents a way to guarantee that it will cite you, and anyone selling that is guessing. If you want help with the practical side, see our AI search visibility service; the rest of this article is the reasoning so you can check it yourself.
Three labels, one pile of work
SEO (search engine optimisation) is making pages findable and useful in search results. AEO (answer engine optimisation) is the older of the two newer labels: writing so that a search engine can lift a direct answer, as in a featured snippet or a voice reply. GEO (generative engine optimisation) is the label for being used or cited when a model writes an answer from several sources.
These are industry terms, not technical standards. Google's own guidance, in its post on succeeding in AI search, does not use GEO or AEO as categories. Its advice is the usual Search advice: unique, satisfying content, good page experience, pages that Googlebot can reach, and structured data that matches visible text. Google's documentation for AI Overviews and AI Mode says there are no additional requirements to appear in them, and that you do not need new machine-readable files, AI text files or markup. That is one vendor's statement about its own product, not a statement about every assistant, but it is a useful anchor.
Our position: GEO is mostly SEO done properly, plus two things that are really about trust and access. If someone sells GEO as a separate secret discipline, ask which documented mechanism they are working on.
What the vendors document, and what they do not
We read each vendor's own pages on 9 November 2026. They describe crawlers and controls clearly. They do not describe how an answer decides which sources to cite.
| Vendor and bot | Documented purpose | Control you have | Not documented there |
|---|---|---|---|
| Google: Googlebot and AI features | Pages must be indexed and eligible to show with a snippet to be used as supporting links in AI Overviews and AI Mode | robots.txt for Googlebot; nosnippet, data-nosnippet, max-snippet and noindex for how content is shown | Any guarantee of inclusion; Google says indexing and serving are not guaranteed |
| Google: Google-Extended | A robots.txt token for whether crawled content may be used for training future Gemini models and for grounding in Gemini Apps and Vertex AI | robots.txt token; no separate user agent string | Google says it does not affect inclusion or ranking in Search |
| OpenAI: OAI-SearchBot | Surfaces sites in ChatGPT search features | robots.txt; OpenAI says blocking it means your site will not be shown in ChatGPT search answers, though it may still appear as a link | How sources are chosen or ordered |
| OpenAI: GPTBot | Crawls content that may be used to train its models | robots.txt, independent of OAI-SearchBot | Any link between training access and citation |
| OpenAI: ChatGPT-User | Visits triggered by a person using ChatGPT or a custom GPT | OpenAI says robots.txt rules may not apply to these user-initiated visits | n/a |
| Anthropic: ClaudeBot, Claude-User, Claude-SearchBot | Training collection, user-directed retrieval and search indexing respectively | robots.txt; Anthropic says blocking by IP address is not a reliable opt-out | How Claude picks sources to cite |
| Perplexity: PerplexityBot, Perplexity-User | PerplexityBot surfaces and links sites in Perplexity search; Perplexity-User fetches pages for a person's question | Perplexity recommends allowing PerplexityBot; says Perplexity-User generally ignores robots.txt because a user asked for the fetch | Ranking or citation rules |
Read the last column twice. Everything an agency tells you about what "gets cited" beyond access is an observation, a theory or a sales pitch. Some observations are reasonable. None is a vendor guarantee, and AI answers vary between runs, so one screenshot proves little.
The fix-first ladder
Work from the top. Each step is only worth doing if the one above it is sound, because a blocked page cannot be cited however well it is written. The last column says how much weight the evidence can bear.
| Step | What to fix | Evidence it relies on | How solid |
|---|---|---|---|
| 1. Access | Open pages return a normal success status. robots.txt and any security layer or CDN rule do not block the search crawlers you want. Decide deliberately about training crawlers separately. | Google's AI features page (eligibility) and the OpenAI, Anthropic and Perplexity bot pages | Documented. The most reliable step |
| 2. Indexing | Important pages are indexed in Google, have a canonical address, appear in the sitemap and are not set to noindex or nosnippet by accident. | Google: a page must be indexed and snippet-eligible to be used in its AI features | Documented for Google |
| 3. A real answer | Each page that targets a question states the answer in its first lines, then supports it with specifics: limits, conditions, who it is not for. | Google asks for unique, satisfying content. For other assistants this is our judgement, not a vendor statement | Good practice, unproven mechanism |
| 4. Clear identity | One consistent set of facts about who you are, where you operate and what you do, on your site and on the profiles you control. No contradictions in years, names or addresses. | Our judgement from how people verify companies. No vendor documents it as a citation factor | Sensible, unproven |
| 5. Structured data that matches the page | Markup only for things visible on the page; keep it accurate. | Google says no special schema is needed for AI features and that markup must reflect visible content | Hygiene, not a lever |
| 6. Independent mentions | Profiles, directories, reviews and press from sources you do not control, where you are genuinely eligible for them. | Third-party observations only. No vendor page we read documents it as a factor | Indicative only; slow |
| 7. Measurement | Track Search Console, analytics referrals and a small fixed set of prompts over time. Expect noise. | Google reports AI feature traffic inside the Web search type in Search Console, not as a separate report | Documented for Google; manual prompts are noisy |
You can self-check steps 1 to 5 with our AI search readiness checklist. It is a checklist, not a score that predicts results.
Where this ladder is wrong
- You already rank well and your pages are open. Then steps 1 and 2 are done, and the remaining work is writing and corroboration. Do not pay for a "technical GEO" project to rediscover that.
- You are a new company nobody has heard of. Step 6 does more than the rest, and it cannot be done from your own website. This is a slow job with no shortcut, and a page rewrite will not substitute for it.
- You deliberately block training crawlers. That is a legitimate decision. Just do not block the search crawlers by accident: per the OpenAI and Anthropic pages, search bots and training bots are separate user agents with separate controls.
- You expect AI answers to replace search traffic. Nothing we read supports a forecast either way. Measure your own referrals before you change your plan.
The llms.txt question
Teams often ask about llms.txt early. Google says you do not need AI text files for its AI features. Anthropic's crawler page does not mention the file at all. OpenAI's bot page and Perplexity's bots page each point to an llms.txt, but only as an index of their own documentation; neither says its crawler gives sites credit for having one. That is silence, not a verdict. If you want one anyway, it costs little; our llms.txt generator builds one, and our article on llms.txt goes through the evidence. Treat it as tidy documentation for machines that choose to read it, never as a ranking tool. Our article on getting cited covers tactics once the basics above are in place.
How to measure without fooling yourself
Pick a small, fixed list of questions your buyers genuinely ask. Run them on a schedule across the assistants you care about, record whether your site is named or linked and whether the description of you is accurate, and keep the raw answers. Answers change with wording, time and settings, so look for a direction over months rather than celebrating one good result. Pair that with what you can see directly: Search Console for Google, and referral traffic in your analytics. Google's post on AI search makes a similar point: judge success by the value of visits, not only clicks.
What we do and do not offer
Website Prism Ltd is a registered England and Wales company (no. 17462217) with delivery from a team in Pakistan. The founder, Usama Afzal, has delivered 800+ projects over 8+ years, mostly via Fiverr. We do the access, indexing, content and structured data work on the ladder, and we will tell you which steps your site has already passed. We do not guarantee citations, rankings or traffic from any search engine or assistant, and we do not know of any vendor that offers such a guarantee. If you want a written scope for this work, ask for a quote and tell us your site and the questions you want to be found for.
How we work: refund policy · terms of service · about Website Prism