What is generative engine optimization (GEO) — and which half of it is just SEO
A definition you can quote, the part that is genuinely new, and the check most sites fail before any of the advice applies: whether the AI crawlers can open your pages at all.
awakes.ai editorial ·
What generative engine optimization means
Generative engine optimization (GEO) is the work of getting your business named and linked inside AI-generated answers — in ChatGPT, Perplexity, Claude, Microsoft Copilot and Google's AI Overviews — rather than only ranked in a list of links. The unit of success changes: not a position, but a mention, and the source link that comes with it.
The term is younger than the arguments about it. It comes from a paper called "GEO: Generative Engine Optimization", submitted to arXiv in November 2023 by Pranjal Aggarwal and co-authors and presented at KDD 2024. The authors built a benchmark of user queries and reported that their methods lifted a source's visibility in generated answers by up to 40 per cent — their number, on their benchmark, not a promise for your website.
What survived from the paper into everyday use is narrower: the observation that an answer engine reads pages, picks a handful, and rewrites them into one reply. Everything sold as GEO today is an attempt to be among that handful.
GEO, AEO, AI SEO: three names, one job
GEO, AEO (answer engine optimization) and AI SEO are used interchangeably by most people selling them, and the distinctions are mostly marketing. Google settled its own half of the question in writing. Its guide to generative AI features, last updated on 10 July 2026, says it plainly: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
That sentence does a lot of work. For Google's AI Overviews and AI Mode, there is no separate discipline and no separate lever — the same ranking systems decide. The same guide lists what you do not need to do for it, and that list contradicts a good deal of what is being sold under these three names. We take that list apart in a separate article so this one can stay on the part Google's guide does not cover: every engine that is not Google.
What is genuinely new: there is more than one engine now
For twenty years, being findable meant one thing: Google. AI answers broke that into several independent decisions. Google's AI Overviews run on Google's index. Copilot runs on Bing's. ChatGPT, Perplexity and Claude each send their own crawler and keep their own view of the web. Five engines can hold five different opinions about whether your page exists, and none of them tells the others.
This is the practical difference between GEO and the SEO you already know, and it is not about writing style. It is about access, indexes and measurement — three things that used to have one address each and now have five. The ceiling is worth stating up front: none of this controls what a model says about you. A model picks its sources per question and rewrites them in its own words. What you can control is whether you are in the pool it picks from.
Which robot does what, by the vendors' own documentation
| Engine | Crawler for search and answers | Separate crawler for training | Follows robots.txt |
|---|---|---|---|
| ChatGPT | OAI-SearchBot | GPTBot | Yes for both; ChatGPT-User, which fetches a page when a person asks, is documented as user-initiated and robots.txt rules may not apply |
| Perplexity | PerplexityBot | None declared: PerplexityBot is documented as not used for training | Yes for PerplexityBot; Perplexity-User "generally ignores robots.txt rules" |
| Claude | Claude-SearchBot | ClaudeBot | Yes for all three, including Claude-User |
| Microsoft Copilot | Bingbot — the ordinary Bing crawler | None separate | Yes |
| Google AI Overviews and AI Mode | Googlebot — the ordinary Search crawler | — | Yes |
Checked against the vendors' own documentation on 9 September 2026. OpenAI also runs OAI-AdsBot, which validates the safety of pages submitted as ads on ChatGPT, and publishes IP ranges for each of its four agents separately — a reminder that "block the AI bots" is not one decision but four.
The check that takes a minute, and that most advice skips
Before a single word of content advice matters, one question has to be answered: does the page open for the robot at all? OpenAI's documentation is blunt about the stakes — "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." Not ranked lower. Not shown.
You can test the door yourself. The command below asks your own server for your own home page while identifying as ChatGPT's search crawler, and prints the status code it gets back. A 403 here is not a ranking problem, it is an absence — and it is invisible in every SEO report you have ever received, because a browser and Googlebot both walk in fine.
Show the command
curl -sS -o /dev/null -w "%{http_code}\n" \
-A "Mozilla/5.0 (compatible; OAI-SearchBot/1.0; +https://openai.com/searchbot)" \
https://yoursite.com/Or run it here, on your own site
The same request, made from our server instead of your terminal, for six crawlers at once. The ceiling described below applies to this version exactly as it applies to curl.
Check whether AI crawlers can reach your site
We ask your homepage for its response code once per crawler, from our server. Nothing is stored.
The ceiling of that check, and how to do it properly
A 200 is good news but not proof. Anyone can claim to be OAI-SearchBot, so the engines publish the IP ranges their crawlers actually come from, and serious protection checks those addresses rather than the name in the header. A site can wave through anything that says "OAI-SearchBot" and still block the real one, or the other way round. The honest confirmation is your server logs, matched against the published ranges: OpenAI, Perplexity and Anthropic each publish theirs.
Run it against the pages that matter, not only the home page. It is common for a home page to answer and a service page behind a security rule to refuse, which is exactly the page you wanted quoted.
How a site hides itself by accident
The usual cause is not a hostile decision, it is a tidy one. Since July 2025 Cloudflare has offered a one-click switch to block AI bots, and plenty of owners flipped it for a reasonable motive: they did not want their writing feeding somebody's training run. The switch did not separate the crawler that trains a model from the crawler that sends people to you. On one of our sites, with robots.txt fully open, requests identifying as GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot and CCBot came back 403, while Googlebot and Bingbot walked in without a word. The site was invisible in ChatGPT and Perplexity and perfectly healthy in every Google report.
Cloudflare has since split AI traffic into three behaviours you can control separately — Search, which indexes your content to answer questions about it later; Agent, which acts in real time on a person's behalf; and Training, which feeds a model. From 15 September 2026, new domains onboarding to Cloudflare get Training and Agent blocked by default on pages that show ads, while Search stays allowed. That is the right shape for the decision, and it is still a decision — the defaults change over time, they differ by plan, and nobody sends you a notification when your visibility in an answer engine ends.
What actually moves the needle
- 1
Let the search crawlers in, on purpose
Decide separately about search, agents and training, then check the decision from outside with the command above and confirm it in server logs. This is the only step where a mistake makes every other step pointless.
- 2
Be in both indexes, not one
Search Console covers Google. Bing Webmaster Tools covers Bing, and Copilot runs on Bing's index. A page missing from Bing is missing from Copilot, and no amount of content work fixes an absent page.
- 3
Answer the question on the page
Say what you do, who it is for, what it costs and how to start, in plain sentences a person can act on. Do this because customers ask — Google explicitly says you do not need to rewrite content for AI systems, and it is right. A page written for a machine reads badly to a buyer, and the buyer is the point.
- 4
Keep your facts identical everywhere
Company name, services, locations and contact details should agree across your site and your business profiles. An engine assembling an answer about you reads several sources; contradictions between them are the cheapest reason to be dropped from a shortlist.
- 5
Fix a set of questions and re-ask them monthly
Write down the twenty questions a customer would actually type, ask them across the engines, and record who gets named. Done the same way every month, it becomes a trend line rather than an anecdote.
How you know whether it worked
This is where AI search is still awkward, and where most reports quietly change the subject. The one place with a real citation count is Bing Webmaster Tools, whose AI Performance report opened in public preview on 10 February 2026 and shows how often your content is cited across Microsoft Copilot, AI-generated summaries in Bing and selected partners, down to individual URLs, along with the queries the AI used to retrieve the content. Microsoft expanded it in June 2026 with intents, topics, citation share and comparison views.
Google counts its side differently: in Search Console, a click on a link inside an AI Overview counts as a click, and the whole AI Overview occupies a single position that all its links share. Useful, but it is not a citation count. So the honest report has three parts — the citation numbers Bing gives you, the ordinary traffic both search engines give you, and your own monthly set of questions. The last part is a sample, not a measurement: answers vary between runs and between users, and anyone presenting a screenshot of one lucky prompt as proof is showing you weather, not climate.
When GEO is not worth your money yet
Three pages of adjectives cannot be cited by anything, human or machine. The first job is a page that answers a real question; GEO comes after that, not instead of it.
For some local trades, customers still use a map and a phone. Try the questions yourself before buying a programme built on the assumption that they don't.
Engines lean on sources they have seen before. On a domain with no history, the first months of work are ordinary SEO wearing a newer name.
This is a slow channel — crawl, index, citation, click. When the pipeline is empty now, spend on something with a shorter loop and start this in parallel.
Questions people ask about GEO
- What is generative engine optimization (GEO)?
Generative engine optimization is the practice of making a business findable inside AI-generated answers — in ChatGPT, Perplexity, Claude, Microsoft Copilot and Google's AI Overviews — rather than only in a ranked list of links. In practice it covers three things: making sure each engine's crawler can reach your pages, being present in both the Google and Bing indexes, and having pages that answer a customer's question clearly enough to be quoted. The term comes from a 2023 research paper presented at KDD 2024.
- Is GEO replacing SEO?
No. For Google's own AI features it is not even a separate discipline: Google's documentation states that from Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO. What GEO adds is everything outside Google — ChatGPT, Perplexity and Claude each crawl the web themselves and decide separately whether your site is visible to them. That part is new work, and it is mostly about access and measurement rather than writing.
- Is SEO dead now with AI?
No, and the mechanism explains why. AI answers are assembled from web pages that an engine has crawled and indexed, so the things that made a page findable — it loads, it can be crawled, it answers a real question, other sites reference it — are the same things that get it cited. What has changed is that some searches now end without a click, so the value of being the source an answer names has gone up relative to the value of a link position.
- What is the difference between GEO, AEO and AI SEO?
Very little in practice. GEO stands for generative engine optimization, AEO for answer engine optimization, and AI SEO is the catch-all sales term for both. Google's documentation treats them as names for the same work. When a supplier draws a sharp line between them, ask what they would do differently under each name — the answer is usually nothing.
- How do I check whether ChatGPT can see my website?
Request one of your own pages while identifying as ChatGPT's search crawler and read the status code: curl -sS -o /dev/null -w "%{http_code}\n" -A "Mozilla/5.0 (compatible; OAI-SearchBot/1.0; +https://openai.com/searchbot)" https://yoursite.com/. A 403 means that request was refused, and OpenAI's documentation is explicit that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. A 200 is encouraging but not proof: confirm with server logs against OpenAI's published IP ranges, and test your service pages, not just the home page.
- Do I need an llms.txt file for AI search?
Not for Google. Its guide to generative AI features says you do not need to create new machine-readable files, AI text files, markup or Markdown to appear in Google Search, and that doing so will neither harm nor help your visibility there. No other major engine documents a requirement for one either. Adding the file is cheap and harmless; treating it as an AI search strategy is not.
- How do I know if my site is being cited in AI answers?
Bing Webmaster Tools is currently the only place with a direct citation count. Its AI Performance report, in public preview since 10 February 2026, shows how often your content is cited across Microsoft Copilot, AI summaries in Bing and selected partners, including which URLs are cited and which queries retrieved them. Google's Search Console counts a click on a link inside an AI Overview as an ordinary click, with all links in the overview sharing one position, but does not report citations as such. For ChatGPT, Perplexity and Claude, the practical method is your own fixed set of questions, asked on the same schedule each month.
- Can anyone guarantee that ChatGPT will mention my business?
No, and a guarantee is a reason to walk away. Engines choose their own sources per question, and the same question can produce different answers minutes apart. What can be promised is the work underneath: the crawlers get in, both indexes are complete, the pages answer questions worth quoting, and the reporting shows what actually changed. That improves the odds; the rest belongs to the model.
Not sure whether AI search can see your site at all?
Send us your address. In 30 minutes we will go through what the crawlers get, what both indexes hold, and which pages answer a customer's question well enough to be quoted. Free, no strings.
See how we do AI SEO