answer engine optimization, generative engine optimization and llm seo describe one job: making pages that ai search features and assistants can find, use accurately and attribute to you. most of that job is ordinary seo. google says its ai features use the same indexing and snippet eligibility as normal search. the real differences are in writing passages that stand alone and in measuring four outcomes separately. this guide covers both, with a page audit you can run today.
- no seo technique or submission tool guarantees an organic citation in ai answers. ads are a separate product, not covered here.
- google needs a page to be indexed and snippet-eligible, and the site not excluded by its search generative ai control; no special file or schema is required.
- chatgpt search depends on allowing OAI-SearchBot in robots.txt.
- eligibility, citation, click and conversion are separate outcomes. track each one on its own.
four outcomes, not one
most aeo advice blends four things into “ai visibility”. they are measured with different tools, and success at one does not guarantee the next.
| outcome | question | where you can see it | limit |
|---|---|---|---|
| discovery and eligibility | can the system fetch and use the page? | search console indexing and generative ai setting, robots.txt, server logs | eligible is not the same as shown |
| appearance or citation | did an answer show or reference the page? | google generative ai performance (impressions); bing ai performance (citations); manual checks | counts appearances, not rank or authority |
| referral or click | did someone visit from the answer? | analytics referrers; search console web totals | many visits arrive without a usable referrer |
| conversion | did that visitor do something useful? | your own analytics and sales data | small samples; attribution is often unknown |
google's ai features guidance says pages shown in ai overviews and ai mode are counted in the performance report under the normal web search type, and that indexing and serving are not guaranteed even when a page meets every requirement. search console also has a generative ai performance report that shows impressions in ai overviews and ai mode; the same impressions are also included in the standard web report. google notes the dedicated report may not appear for sites with too few such impressions, and it shows impressions, not rank or credit. bing's ai performance report counts citations across copilot, bing ai summaries and some partner integrations. bing says these numbers do not indicate ranking, authority or a page's role in an answer.
write passages that stand on their own
many ai answers quote or summarise part of a page, and the reader sees that part without the rest. so a paragraph that begins "this is why it matters" loses its meaning when lifted out. a paragraph that begins "a referring domain is one linking site, however many pages on it link to you" still makes sense. google's guide to generative ai search says you do not need to split content into tiny pieces or write in a special style for ai. we keep paragraphs self-contained for a simpler reason: it makes them clearer for every reader.
the practical test: read any paragraph with the rest of the page covered. if it still states a complete claim with its subject named, it is easy to quote accurately. if it depends on the sentence before, fix the opening clause.
- answer in the first paragraph. then expand. the inverted pyramid was built for exactly this.
- put key numbers in text. some systems can read images and charts, but text is the most dependable format and also serves screen readers. bing advises keeping text, images and video consistent about the same facts.
- say what you measured and what you inferred. a claim that carries its source and date is safer for anyone to repeat.
a page audit you can run
use this table on one important page at a time. each row is a check google or bing documents for search in general. none of them is a special ai ranking factor, and passing all of them does not guarantee a citation.
| check | what good looks like | how to test | source |
|---|---|---|---|
| clear answer | the main question is answered in the first paragraph, in plain words | read only the lead; does it answer the title? | bing structure guidance |
| natural headings | headings describe the section content and follow the reader's questions | read the table of contents alone; does it outline the page? | bing structure guidance |
| sources and authorship | claims link to evidence; the byline and any method note are accurate | can a reader see who wrote it and where each number came from? | google helpful content |
| appropriate schema | article and faq markup only where it matches visible text | compare the json-ld with the rendered page | google article markup |
| access | indexable, snippet-eligible, and allowed for the crawlers you want | url inspection; robots.txt; preview controls; read the search generative ai setting | google ai features, openai bots |
the sources behind those rows: google's helpful content guidance asks whether it is clear who created content, with accurate bylines where readers expect them. google's article markup documentation explains how markup helps google understand a page, and its structured data rules require markup to reflect visible content. bing's ai performance announcement recommends clear headings, tables and faq sections, with claims backed by examples, data and sources.
example: auditing one passage
this is a bounded, illustrative example from our own writing style, not a measured test. a draft paragraph read:
“as we saw above, this makes a huge difference, so most people should do it.”
it fails the audit three ways: no subject, no claim and no source. the rewrite names all three:
“google says pages need to be indexed and eligible for a snippet to appear in ai overviews, with no extra technical requirements. so check indexing in search console before buying any aeo tool.”
the second version can be quoted alone, points to a checkable source and suggests one action. we would not claim the rewrite gets cited more often. it is simply harder to quote wrongly.
llms.txt: optional, unproven, cheap
llms.txt is a plain-text file at the root of a site describing what the site is and what its important pages cover. we publish one, at /llms.txt, alongside a longer /llms-full.txt. here is the shape:
# CrawlGraph > one-sentence description of what this site is and who it is for. ## Core pages - [Gap analysis](https://example.com/gap-analysis): what the page does, stated as a capability rather than a slogan. - [API documentation](https://example.com/docs/api): what the free tier covers and where it stops, in the same words the page uses. ## What this tool does not do - rank tracking, keyword research, live re-crawl.
it is not required. google says you do not need new machine-readable files or ai text files to appear, and its generative ai search guide says google search disregards llms.txt. we have not seen another major vendor document that it reads the file. we publish it because it costs an afternoon and the writing exercise is useful: describing each page in one line surfaces the pages you cannot describe, which are often the ones competing with another page on your own site. our guide to finding keyword cannibalization in search console shows how to confirm that overlap with data.
the most useful section is the one listing what the tool does not do. stating scope plainly, on the page and in the file, makes it less likely that a summary credits you with a feature you do not have.
structured data, and the rule that matters
schema markup is not a requirement for appearing in google's ai features; google says so directly. it can still remove ambiguity, and it has one rule worth enforcing: the markup must say what the visible page says. markup that disagrees with the rendered text breaks google's structured data guidelines and leaves machines with two versions of your claim.
do not expect a rich result from faq markup. google's search updates log records that the faq rich result stopped appearing in google search from 7 may 2026, and its documentation was removed in june 2026. we still emit FAQPage, with no rich-result benefit expected, because every faq on this blog is declared once as data and read twice: the visible questions and the json-ld are generated from the same array, so they cannot drift apart. that keeps the markup honest whatever it is used for.
be fetchable and indexable first
none of the above matters if a system cannot fetch the page. for google, that means the page is indexed and snippet-eligible, and you control what appears with nosnippet, data-nosnippet, max-snippet or noindex. search console also has a search generative ai control under settings. it includes sites by default; excluding a property removes its content from ai overviews, ai mode and generative ai features in discover. check its current value as a read-only step before changing anything. for chatgpt search, openai's crawler documentation separates OAI-SearchBot, which surfaces sites in search answers, from GPTBot, which collects training data. sites that block OAI-SearchBot will not be shown in chatgpt search answers, and you can allow search while blocking training.
no submission route guarantees an organic citation, but discovery tools exist. sitemaps and url inspection help google find pages, and bing recommends indexnow for telling it about changes. these help a system find a page; none guarantees that an answer will use it.
links matter for discovery too. google says it uses links to find pages and as a relevance signal in search. no vendor says links decide which pages an ai answer cites, so treat links as part of being found and trusted in ordinary search. what the link data shows about that is in whether seo is dead, answered from the link graph, the strategic half of this question.
the underrated one: ship an interface, not just pages
assistants can do more than read your marketing pages: some can call your product. we publish an mcp server, crawlgraph-mcp on npm, which lets an assistant run a backlink lookup or a gap analysis directly and get structured results back rather than prose scraped off a page.
this is not a ranking trick and we would not claim it lifts anything in search. it is a distribution channel with different economics: a person asking their assistant to check a domain's backlinks can be served by your product inside the conversation, with no page view involved. if you sell something an assistant could invoke, that interface may deserve more attention than another schema checklist.
what to ignore
- anyone quoting a weighting. no vendor publishes how heavily its systems use any signal. specific percentages in this category are invented.
- guaranteed organic citation. google says serving is not guaranteed, and openai makes no ranking promise for sites that allow its crawler.
- special ai schema or files sold as required. google says none is needed for its ai features.
- one blended visibility score. citation counts, search clicks and referral visits measure different things.
measure visits separately from whether an answer mentions your brand. ai search referral tracking explains how to classify observable referral sessions and compare their outcomes, while retaining an unknown bucket for visits without a usable referrer. those sessions cannot establish how often an answer engine cited you.
vendor statements come from the google, openai and bing pages linked inline, read on 25 october 2026. statements about our own setup (llms.txt, the shared faq data, the mcp server) are checkable on this site and on npm. we have not run a controlled citation experiment, so this guide makes no claim that any step raises citations.
more on measurement and strategy for ai search is collected in the seo for ai search topic.
faq
What is answer engine optimization?
Answer engine optimization is the practice of making a page easy for AI search features and assistants to find, use accurately and attribute to you. Most of it is ordinary SEO and clear writing. Google says its AI features rely on the same pages and practices as normal search, so the useful additions are measurement and writing passages that stay accurate when read on their own.
Is answer engine optimization different from SEO?
Mostly no. Google says a page needs to be indexed and eligible for a snippet to appear in AI Overviews or AI Mode, with no extra technical requirements, and the site's Search generative AI control in Search Console must allow inclusion, whether set on the property or inherited from a parent property. Inclusion is the default. What changes is what you measure: whether a page is eligible, whether an answer cites it, whether anyone clicks through and whether those visitors convert are four separate outcomes, and none of them guarantees the next.
Does llms.txt actually do anything?
Not for Google: Google says Search disregards llms.txt files and that you do not need AI text files to appear in its AI features. We have not seen another major vendor document using it either, so treat it as optional and unproven. It is cheap to publish, and writing a one-line description of each important page is a useful exercise because it exposes pages whose purpose you cannot state.
How do I get cited by ChatGPT or Perplexity?
Nobody can guarantee an organic citation, and no submission form produces one. For ChatGPT search, OpenAI says sites that block OAI-SearchBot will not be shown in search answers, so first make sure robots.txt allows it. Then publish the answer in crawlable text, state it clearly near the top, show sources for your numbers and keep the page indexable in the search engines these products may draw on.
Do I need structured data for AI search?
No. Google says no special schema is required for its AI features. Structured data can still help search engines understand a page, as long as it matches the visible content exactly. Markup that disagrees with the page is a quality risk and gives machines two versions of your claim.
How do I measure answer engine visibility?
Use separate measures for separate outcomes. Search Console has a Generative AI performance report showing impressions in AI Overviews and AI Mode; those impressions are also included in the standard Performance report under the Web search type. The dedicated report may be missing for sites with few such impressions. Bing Webmaster Tools reports citation counts for Copilot and Bing AI summaries, and analytics shows referral visits only when a referrer survives. None of these shows authority or ranking, so do not add them into one score.
writes the queries we run internally. ships one tactical post a week.
plus one when a new common crawl release lands. that is all.