Generative engine optimization (GEO): what is documented, what was studied and what is guesswork
Generative engine optimization (GEO, also sold as AEO or AI SEO) is the work of getting a site quoted in AI answers. What the engines document is short. Let their search crawlers in, keep pages indexable, put answers in visible text. The rest is a study or guesswork.
- By
- WP Ministry
- Published
In short
- GEO, AEO, "AI SEO" and "AI search optimization" name the same work. Google's guide says it is still SEO.
- Google documents no extra requirement for AI Overviews or AI Mode. A page must be indexed and eligible for a snippet.
- Microsoft documents how to write a page Copilot can quote. Clear headings, a short answer that stands alone, nothing hidden in tabs.
- OpenAI and Anthropic each run a search crawler apart from a training crawler. Blocking one is not blocking the other.
- One study found no relation between llms.txt and citations. Another found no clear gain from adding structured data.
- Nobody can promise a citation or a mention. Google tells buyers to check a seller's advice against its own pages.
Generative engine optimization, or GEO, is the work of getting a site's pages linked or quoted in the answers that AI assistants and search engines write. AEO, "AI SEO" and "AI search optimization" are other names for the same work, and Google's own guide says it is still SEO.
What the engines themselves document is short. Their search crawlers have to be able to reach the page. For Google, the page has to be indexed and eligible to be shown with a snippet. Microsoft adds that the answer should sit in visible text under a clear heading. Much of what is advised beyond that rests on a single study, or on nothing an engine has said. This page puts each claim in one of three standings and gives the source for every one.
What GEO, AEO and AI SEO mean
- GEO is generative engine optimization. The name was introduced in a research paper first posted in November 2023.
- AEO is answer engine optimization.
- AI SEO and AI search optimization are plain words for the same thing.
Google's guide says GEO and AEO are both used for work aimed at being visible in AI search.
Google takes neither as its own term. It writes "generative AI features" for AI Overviews and AI Mode, and its guide says: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
Microsoft uses GEO. Bing's Webmaster Guidelines say it is about a page's eligibility to ground and be referenced in AI answers, and open by saying that the fundamentals of SEO still apply, because Bing and Copilot rest on the same crawling, indexing and ranking.
The three standings
- Documented. An engine or a platform says so on its own page.
- A study. Someone measured it. Who ran it, on what, when, and whether they sell something in this trade all matter. A correlation is not a cause.
- Guesswork. Said widely and documented by nobody.
No claim is moved up a standing here. Each statement carries its source's date where there is one, and everything below is as its source read on October 8, 2026.
What the engines document
| Engine | What it documents | Dated |
|---|---|---|
| No extra requirement for AI Overviews or AI Mode. A page must be indexed and eligible for a snippet, and the site not excluded in Search Console. | December 10, 2025; July 10, 2026 | |
| Microsoft | How to write a page Bing and Copilot can quote, which directives limit them, and a report of citations. | October 8, 2025; February 10, 2026; June 16, 2026 |
| OpenAI | OAI-SearchBot decides whether a site is shown in ChatGPT's search answers. GPTBot is for training. | Undated |
| Anthropic | Claude-SearchBot is for search, ClaudeBot for training, Claude-User for a person's request. | April 7, 2026 |
| Perplexity | PerplexityBot is for its search results. Perplexity-User fetches when a person asks. | Undated |
- Requirements. Google's page on AI features says there are no additional requirements to appear in AI Overviews or AI Mode. To be shown as a supporting link, a page must be indexed and eligible to be shown in Search with a snippet.
- What a site can ignore. Google's guide, published on May 15, 2026, and updated on July 10, 2026, lists what a site does not need for Google Search:
llms.txtand other machine-readable files, content split into small chunks, pages rewritten for AI systems, and special markup. Structured data is not required, and there is no special schema.org markup to add. A note added on June 15, 2026, says anllms.txtfile neither harms nor helps. - What it says matters most. Content that people find unique and useful will likely do more for a site's presence in generative AI search than anything else the guide suggests.
- Snippet controls count. Google's robots tag documentation says
nosnippetprevents a page's content from being used as a direct input for AI Overviews and AI Mode, andmax-snippetlimits how much may be used. - A report. Search Console's Generative AI performance report was announced on June 3, 2026, for a subset of sites and, its help page says, rolled out to all websites as of August 31, 2026. It gives impressions in AI Overviews and AI Mode by page, country, date or device. Queries are not among its groupings, and a site with too few impressions sees no report.
- A way out. In Search Console, under Settings, then "Search generative AI", a property is included in AI Overviews, AI Mode and the generative AI features in Discover, or excluded from them. Included is the default, and a property under a parent follows its parent. The help page says the control is rolled out to all websites as of August 31, 2026, and that the choice is not a ranking signal for other parts of Search. Google's guide lists being included among the conditions for appearing.
- Spam policies apply. Since a clarification of May 15, 2026, the spam policies name attempts to manipulate generative AI responses in Search beside attempts to manipulate ranking.
How to appear in Google's AI Overviews goes through Google's side in full.
Microsoft: Bing and Copilot
Microsoft is specific about how to write a page, in a guide of October 8, 2025, and in Bing's Webmaster Guidelines, which carry no date.
- How Copilot reads. The guide says Copilot breaks a page into pieces and builds an answer from pieces of several pages.
- What to write. A title, a description and a top heading that agree. Headings that mark where one idea ends and the next begins. Questions with direct answers of one or two sentences that still make sense when lifted out of the page. Lists and tables for steps and comparisons. The guidelines add: key information near the top, one subject to a page, and facts stated on the page itself.
- What to avoid. Long walls of text, key facts that exist only in an image or a PDF, and answers placed in tabs or expandable menus, because its systems may not render hidden content.
- Directives. The guidelines say
noindexkeeps a page out of Bing search and Copilot,noarchiveprevents its content from being used in Copilot's responses,nocachelimits Copilot to the address, title and snippet, andnosnippetmay limit the quality of a citation. - Structured data. It may support clearer grounding, must match the visible content, and does not guarantee visibility.
- Abuse. The guidelines list cloaking, wording engineered to trigger citations, and content added to manipulate the AI models behind Bing or Copilot.
- A report. AI Performance in Bing Webmaster Tools opened as a public preview on February 10, 2026. It counts citations across Copilot, AI summaries in Bing and some partner products, by page, and shows grounding queries: the key phrases the system used to retrieve what it cited. Its help page says these are not a person's full question. On June 16, 2026, Microsoft added Intents, Topics, Citation Share and Compare, still in preview.
- IndexNow. Microsoft recommends this protocol for telling search engines that a page changed. The protocol's FAQ lists endpoints for Amazon, Bing, Naver, Seznam.cz, Yandex and Yep. Google is not on that list.
OpenAI, Anthropic and Perplexity
Each company publishes the names its crawlers answer to in robots.txt.
| Company | Search crawler | Training crawler | Fetch when a person asks |
|---|---|---|---|
| OpenAI | OAI-SearchBot | GPTBot | ChatGPT-User |
| Anthropic | Claude-SearchBot | ClaudeBot | Claude-User |
| Perplexity | PerplexityBot | None named | Perplexity-User |
- OpenAI. The rules for
OAI-SearchBotandGPTBotare independent. A site that disallowsOAI-SearchBotis not shown in ChatGPT's search answers, though it can still appear as a navigational link. A change to robots.txt takes about 24 hours to reach its search systems. OfChatGPT-User, the page says robots.txt rules may not apply, because a person started the request. - Anthropic. Its page says its bots honor robots.txt, and that it supports the non-standard
Crawl-delayline. BlockingClaude-SearchBotorClaude-User, it says, may reduce a site's visibility in search results and in answers to a person's request. - Perplexity. Its page recommends allowing
PerplexityBotin order to appear in its search results, and says that crawler is not used to collect content for training foundation models.Perplexity-User, it says, generally ignores robots.txt because a person asked for the fetch. A change can take up to 24 hours.
OpenAI and Perplexity both recommend more than a robots.txt rule: requests from their published IP ranges should be allowed too, which is a matter for a firewall, a CDN or a host. None of the three pages says what makes a page chosen for an answer.
Blocking or allowing AI crawlers on WordPress covers the rules and what can stop a crawler before WordPress is asked. WordPress robots.txt covers the file itself.
What studies have found
Four of these five come from companies that sell software for tracking AI answers. Each finding is given as its authors state it.
| Question | Who ran it | Sample and date | Finding |
|---|---|---|---|
| Do assistants cite newer pages? | Ahrefs, which sells SEO software and an AI visibility tool | About 17 million cited addresses from ChatGPT, Perplexity, Gemini, Copilot, AI Overviews and Google's ordinary results. July 28, 2025 | Pages cited by the assistants averaged 1,064 days old, against 1,432 for ordinary results: 25.7% newer, and still 2.9 years old. AI Overviews did not follow. Their cited pages were 16 days older than ordinary results. |
| Do AI Overviews cite the top ten? | Ahrefs | 863,000 searches and 4 million AI Overview addresses. March 2, 2026 | 38% of pages cited in an AI Overview also ranked in the top ten for the same search, against about 76% in its July 2025 study. The authors suggest query fan-out as the cause, and say they changed how they collect citations between the two. |
| Does adding structured data bring citations? | Ahrefs | 1,885 pages that added JSON-LD between August 2025 and March 2026, against 4,000 matched pages that did not. May 11, 2026 | No clear effect. AI Overviews down 4.6% relative to the matched pages, which the authors call real, small and unexplained. AI Mode up 2.4% and ChatGPT up 2.2%, which they say could be noise. Every page was already cited heavily. |
Does llms.txt relate to citations? | SE Ranking, which sells SEO software and AI visibility trackers | Nearly 300,000 domains. November 7, 2025 | 10.13% had the file. No correlation between having it and being cited. |
| Can changes to a page raise its visibility? | Academic authors at Princeton University and IIT Delhi, with two independent researchers. The paper offers no product | 10,000 queries, run through an answer engine the authors assembled and through Perplexity. Posted November 2023, accepted to KDD 2024 | Adding citations, quotations and statistics raised a page's visibility in the answer by 30 to 40% on one of the paper's two measures. Adding keywords did not perform well. Results varied by subject. |
The first, second and fourth are correlations. They show which pages were cited at one time, and not why. The third compared matched pages before and after a change, which is stronger, and found nothing clear. The fifth is a laboratory benchmark, and the Perplexity it tested is the one of that time.
What is guesswork
Each of these is sold or advised, and documented by no engine as a way to be cited.
| Claim | Why it has no standing |
|---|---|
An llms.txt file gets a site cited | Google's guide says Search ignores the file. The crawler pages of OpenAI, Anthropic and Perplexity do not say their crawlers read one. The one large study found no relation. See llms.txt on WordPress. |
| A special "AI schema" or "LLM schema" | No engine documents a schema type for AI answers. Google says there is none to add. Bing says structured data does not guarantee visibility. The one controlled study found no clear effect. |
| FAQ markup for AI answers | Google's documentation log says the FAQ rich result stopped appearing in Search from May 7, 2026, and its documentation was removed on June 15, 2026. |
| Splitting pages into chunks, or rewriting them "for AI" | Google's guide lists both as unnecessary and says to write for the audience. |
| Bought mentions, forum links and paid places in lists | Google's guide says seeking inauthentic mentions is less helpful than it seems, and that its AI features depend on the systems that block spam. Bing's guidelines list schemes that simulate popularity as abuse. |
| A page for every variation of a question | Google's guide says doing this mainly to manipulate rankings or AI responses violates its scaled content abuse policy. |
| A separate version of a page for AI crawlers | No engine asks for one. Google says Search does not use Markdown or other machine-readable files. Google's spam policies and Bing's guidelines both name cloaking: showing crawlers something other than what visitors see. |
| Text placed on a page to steer an assistant | Bing's guidelines list content designed to manipulate its AI models as abuse that may lead to removal. Google's spam policies cover attempts to manipulate AI responses. |
| An "AI visibility score" as proof of progress | Google says third-party tools have no access to its internal data. Bing's help page says its own report does not measure rankings, authority or importance. |
| Changing the date, or refreshing on a schedule | No engine documents a schedule. Bing's guidelines say to update a page when the facts change. The freshness study did not hold for AI Overviews. |
| A promised citation or mention | No engine offers one. Bing's guidelines say GEO does not guarantee grounding or citations. |
A checklist for a WordPress site
Step 1: Let the search crawlers in
With no robots.txt file of its own, WordPress answers with one group of rules for every crawler, and that group closes only
/wp-admin/. Check that a plugin, a CDN or the host has not changed that forOAI-SearchBot,Claude-SearchBot,PerplexityBot, Googlebot or Bingbot. The training crawlers are a separate decision. Standing: documented by OpenAI, Anthropic and Perplexity. Google's page names robots.txt, a CDN and hosting as places where crawling has to be allowed.Step 2: Keep pages indexable and quotable
Under Settings, then Reading, "Discourage search engines from indexing this site" must be unticked. What that box does covers it, and the page indexing check reads one address for a
noindexor a robots.txt rule. Look also fornosnippet,noarchiveornocacheon pages you want quoted, and at the "Search generative AI" setting in Search Console. Standing: documented by Google and by Bing.Step 3: Write the answer in visible text under a heading that asks the question
Open with one or two sentences that answer it and still make sense alone. Keep answers out of a Details block, an Accordion block or a tab, each of which can keep its content closed until a visitor opens it. Standing: documented by Microsoft. Google asks for no rewriting, only that important content is available as text.
Step 4: Publish material of your own
Your own prices, measurements, photos and process, and what you learned doing the work. Standing: documented by Google as what it expects to matter most, and by Bing, which asks for original content.
Step 5: Verify the site in Bing Webmaster Tools
Bing's help page says a site is added to an account only after you prove you own it, and the AI Performance report is inside that account. Standing: documented by Microsoft.
Step 6: Look at both first-party reports
Search Console shows which pages appeared in Google's AI features and how often. Bing Webmaster Tools shows which pages were cited and the phrases behind them. Each reading is a count for its dates, not a score. Standing: documented by Google and by Microsoft.
Step 7: Keep dates honest
Show a date, label it as published or updated, and change it when the page changes. Standing: Google documents how to state a date, and Bing says to update a page when the facts change. That assistants cite newer pages is one study, and it did not hold for AI Overviews. That changing a date brings citations is guesswork.
If you would rather hand this list to someone, WP Ministry does this kind of work as AI search optimization. What is sold is the work, never a citation.
What nobody can promise
No engine promises a citation. Google's page on AI features says a page that meets every requirement may still not be crawled, indexed or served. Bing's guidelines say SEO does not guarantee rankings or traffic and GEO does not guarantee grounding or citations.
Google published guidance on June 5, 2026, for anyone buying this work, and it names "AEO" and "GEO" tools and services directly:
- Tools. "Third-party tools don't have access to our internal ranking data."
- Advice. Good advice either presents a claim as opinion based on data or experience, or backs it with official guidance.
- Sellers. Its page on hiring an SEO gives a buyer questions to ask, among them: "Do they cite official Google documentation as supporting evidence for their recommendations?" It also asks whether advice on AI search agrees with Google's own guide, and says no one can guarantee a first-place ranking.
Common questions
Is GEO different from SEO?
Not according to Google or Bing. Google's guide calls optimizing for generative AI search still SEO. Bing's guidelines say the fundamentals of SEO still apply to AI answers. The practical difference is small: the same indexable, readable page, with more attention to whether its answer can be lifted out and still make sense.
Do I need an llms.txt file?
Not for Google, which says Search ignores it. No crawler page from OpenAI, Anthropic or Perplexity says its crawler reads one, and the one large study found no relation between the file and being cited. llms.txt on WordPress covers what the file is and whether a plugin has already made you one.
Does schema markup help a page get cited?
Nothing documented says it does. Google says structured data is not required for its AI features. Bing says it may support clearer grounding and does not guarantee visibility. The one study that compared matched pages found no clear change after JSON-LD was added. Its documented use in Google is rich results.
If I block GPTBot, do I disappear from ChatGPT?
Not by OpenAI's account. Its page says GPTBot is for training, OAI-SearchBot decides whether a site is shown in ChatGPT's search answers, and the two rules are independent. Anthropic separates ClaudeBot from Claude-SearchBot the same way. For Google, the rules for Googlebot control crawling for Search, its AI features included. Google-Extended covers Gemini training and grounding, and Google says it does not affect a site's inclusion in Search.
- GuideHow to stop AI training on your WordPress content and stay in AI search answers
- Guidellms.txt on WordPress: what it is, who reads it and whether your site needs one
- GuideAI bots slowing down your website: how to confirm it on WordPress and what to do, mildest first
- GuideBing Webmaster Tools for WordPress: setup, IndexNow, and what it means for ChatGPT and Copilot
- GuideHow to appear in Google AI Overviews: what Google requires and what keeps a WordPress page out
- GuideHow to block AI crawlers on WordPress, or let them in: robots.txt, Cloudflare, hosts and plugins

