Answer engine optimization (AEO), defined: how a page earns a citation inside an AI answer

What AEO means, in plain words

Classic search returns a list and lets the buyer choose. An answer engine composes one answer and chooses for them. AEO is the work of becoming the page that answer is built from. The unit of success is not a ranking position; it is a citation: your URL attached to the sentence the buyer reads. Practitioners also call this "LLM SEO"; the work is the same.

It is a page-level discipline. Generative engine optimization (GEO) is the brand-level cousin: it cares about how the brand is described whether or not your page is the source. AI visibility is the measurement both are trying to move.

AEO versus SEO versus GEO

The three stack rather than compete. Google's guide is explicit that AI Overviews and AI Mode retrieve from the core Search index, so a page that is not indexed and snippet-eligible is not a candidate for AEO at all.

What actually earns a citation

1. The eligibility bar: indexed, crawlable, snippet-eligible

For Google's AI features the requirements are the ones you already have: the page must be indexed, crawlable by Googlebot, and eligible to show a snippet. Google's "AI features and your website" page says there are no additional requirements to appear in AI Overviews or AI Mode, and lists nosnippet, data-nosnippet, max-snippet and noindex as the controls that remove you (Google Search Central, last updated 10 December 2025). Since 31 August 2026 there is a fourth, invisible gate: the Search Console "Search generative AI" control, which a property owner can set to Exclude. For the other engines the bar is the retrieval crawler: OAI-SearchBot for ChatGPT search, PerplexityBot for Perplexity, Claude-SearchBot for Claude, each per its vendor's own bot page (see the seven crawler tokens, by role). The snippet half is covered in snippet controls are your AI visibility off switch.

2. Answer-first structure

The sentence an engine quotes is usually the one that answers the question in the first line under the heading. Put the definition, the number or the verdict first, then the reasoning. Headings phrased as the buyer's question, short paragraphs, one idea per paragraph. This is not a trick; it is what makes a passage extractable. Google's guide describes AI Mode's query fan-out, where one question is split into sub-questions that are searched separately, so each section of a page is judged on its own.

3. Named entities and real numbers

Answers are assembled from specifics: named products, named people, dated figures with a source. A page that says "many tools" gives an engine nothing to cite; a page that names the tools, the date and the study does.

4. Third-party presence: the part most AEO advice skips

The largest citation studies we can name all point the same way: the pages engines cite most are lists and comparisons on other people's domains.

• Third-party lists carry the citations. Wix Studio's AI Search Lab analysed 75,000 AI answers and over 1 million citations across ChatGPT, Google AI Mode and Perplexity: listicles took 21.9% of citations, articles 16.7% and product pages 13.7%; for professional-services queries, third-party listicles carried 80.9% of citations against 19.1% for self-promotional lists (vendor study, reported by Search Engine Land, 24 March 2026).
• On your own domain, depth beats lists. SE Ranking's experiment with a fabricated brand (825 prompts, 15,835 AI answers, five AI systems, the first month of a 16-month study) found that deep guides drew about 900 AI answers per page, comparison pages about 145, "best of" listicles 11 and "alternatives" listicles 4 (vendor experiment, reported by Search Engine Land, 29 April 2026).
• Rank inside the list matters. Peec AI's analysis of about 200,000 AI responses across eight engines (September 2025 to March 2026) found that being placed at rank one in a cited third-party listicle lifted B2B SaaS brand mentions by 16.5 percentage points (vendor study).

Read together: on your own domain, write the deep guide and the comparison, not the listicle; off your domain, earn a row in the lists the engines already cite. The second half is outreach, and it is where most of the citation share is.

What does not earn a citation, per Google

• llms.txt. Google's guide to generative AI features says of AI-specific text files, markup and Markdown that "Google Search ignores them" (Google Search Central, last updated 10 July 2026). The llms.txt specification was revised on 10 August 2026; Google's position did not change.
• Chunking pages for AI. No chunking requirement exists in Google's guidance. Write for the reader and let the engine extract.
• AI-specific schema. Structured data is not required for generative AI features. Schema still earns the rich results it was built for; it is not an AI switch.
• Allowing or blocking GPTBot, ClaudeBot or Google-Extended. Those are training and policy tokens; they do not govern whether ChatGPT search, Claude or Google's AI features can read your page. The retrieval tokens do.

Anything sold as an "AI schema plugin" or an "llms.txt generator for Google" should be judged against that one sentence from Google's guide. Our fuller reading of the guide is in Google's AI search guide: what actually matters.

How to measure whether it worked

1. Fix the prompt set first. Buyer questions per topic, across at least two intent types; the IAB's August 2026 guideline treats fewer than 50 queries per programme as exploratory rather than directional.
2. Record the baseline: mention rate, citation share and position per engine, with the number of runs beside each.
3. Ship the change, then re-run the same prompts on the same engines. Compare with the sample size beside both numbers; a move inside the run-to-run band is not a result.
4. Check Google's side in Search Console's Generative AI report: impressions in AI Overviews and AI Mode by page. It shows no clicks and no queries, so it confirms exposure, not the prompt.
5. Keep the verbatim answers. When a client asks why the number dropped, the answer is in the text, not the chart.

ClarAI does steps 2 to 5 on a schedule and prints the sample on every number (how it works); the free grader covers the readiness half of step 1, which is eligibility, not visibility.

AEO for agencies

For an agency the question is not whether AEO is real but whether it fits a retainer. The shape is simple: the prompt set is the scope, the monthly re-run is the deliverable, and the citation earned in a third-party list is the result the client can see for themselves. We wrote the retainer maths up in the economics of an AEO retainer, and the agency page states exactly what white-label reporting ClarAI ships today and what is still roadmap.

Glossary: ten terms, one line each

Answer engine

A search surface that composes one answer instead of returning a list: AI Overviews, AI Mode, Perplexity, ChatGPT search.

Citation

A URL an answer links to as its source. In AEO, the unit of success.

Snippet eligibility

Whether Google may show text from a page; nosnippet, data-nosnippet, max-snippet and noindex remove it.

Query fan-out

Google's term for splitting one question into sub-queries that are searched separately before the answer is composed.

Answer-first

Putting the definition, number or verdict in the first line under a heading, then the reasoning.

Named entity

A specific product, person, place or dated figure an engine can quote and attribute.

Third-party listicle

A best-of or comparison list on someone else's domain; the page type engines cite most.

Retrieval crawler

The vendor token that decides whether an engine can read your page: OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot.

Search generative AI control

Google's Search Console setting (since 31 August 2026) that can exclude a site from AI Overviews and AI Mode with no public trace.

LLM SEO

Practitioner slang for the same work; used as a synonym, never as a discipline of its own.