What is Answer Engine Optimization?

An answer engine is the collective name for interfaces that produce a direct answer to a query instead of a list of links: featured answer boxes in search results, AI summaries inside search, and assistants such as ChatGPT, Perplexity, Claude and Gemini. Their shared behaviour is this: they do not display a page as a whole, they take one passage from it and name the source.

That behaviour changes the unit of content. In classic search optimisation the unit is the page, and the page ranks for a query. In answer engines the unit is the passage; however good the page as a whole may be, the passage carrying the answer will not be quoted if it cannot be understood when read on its own. This is why the definition of AEO sits where content strategy meets technical markup: identifying the question, positioning the answer, sizing the passage and describing the same information to a machine a second time.

AEO is not a label. For a page to count as AEO-ready, three conditions must all hold: the page must be crawlable, the answer must be findable as text (not as an image and not as content produced only on the client), and the passage must still be correct once it has been separated from its context.

The third column is the side of the work that faces generative engines, and it is a separate discipline: GEO (Generative Engine Optimization). In practice all three accumulate on the same page. Without technical foundations AEO does not work, and without AEO there is no passage for GEO to hold on to. We describe the delivery side of that split separately on the enterprise SEO and AEO service pages.

This is why AEO is a writing discipline rather than a trick. The part of the system you can control is these five properties:

• Match. The heading is phrased close to the question the user typed, and the answer sits immediately beneath it.
• Independence. The passage is correct and comprehensible once lifted off the page; it contains no unresolved pronouns and no backward references.
• Position. The answer is not hidden behind introductory sentences or paragraphs of background.
• Form. One idea, plain sentences, measurable length. Use a list where a list is needed and a table where a table is needed.
• Verifiability. If a number appears, its source and date sit beside it; if a term appears, it is defined on the page.

Google's official statement on answer boxes also clarifies the part you cannot control: a page cannot be marked up “to get into” a featured snippet, only marked up to opt out of snippet generation. Source: Google Search Central — Featured snippets and your website ↗(opens in a new tab) · Accessed: 29 July 2026.

Forty to sixty words is not a law of nature; it is our editorial threshold. That range is enough to state a single idea together with its conditions, and short enough to fit inside a quotation. The six rules we apply:

1. The heading becomes a question. Not “Our services”, but “What does this service cover?”.
2. The answer ends in the first sentence. The second sentence is detail, not the answer itself.
3. One passage, one idea. Answering two questions in the same paragraph makes both unquotable.
4. Terms are defined. An abbreviation is expanded where it first appears and explained in one sentence.
5. Numbers are sourced. A number without a source and an access date is not written.
6. Limits are stated. The case in which the answer does not hold is named; this raises the credibility of the passage.

We set out the detail behind the second draft, together with the threshold table, in the Core Web Vitals guide.

The block below is the minimum FAQPage markup for a single question on an FAQ page. The text in the text field must be identical to the answer visible on the page:

Type selection, @id architecture and entity graph construction are covered in the Schema.org and JSON-LD guide; the delivery side is on the structured data service page.

We also write down the limit of this framework: we do not guarantee rankings, and we have no control over the output of generative models. What is measured in that channel is a sample of how often the brand is named as a source across a defined question set; it is recorded with its date and shared in raw form in the report. We apply the same standard to our own site: only measured values are published on the proof page. The reporting rhythm and the acceptance criteria are written down in how we work.

• The selection is not yours. The platform decides which passage gets quoted; markup is a declaration, not a request.
• Output is not repeatable. The same question can produce two different answers in one assistant, which is why measurement rests on sampling rather than a single observation.
• Rich result rules change. Platforms can narrow which markup earns a visual treatment over time; the FAQ decision of August 2023 is the example.
• Zero-click answers are real. On informational queries a user can take the answer without visiting the site. We do not hide that: we measure AEO alongside the share of qualified pages and form completions, not by session count.
• It does not rescue weak content. Marking up an incorrect, incomplete or unsourced answer does not make it correct; it only makes the error easier for a machine to read.

There is no number in this guide reporting a conversion rate, a traffic uplift or a citation percentage. The reason is simple: those numbers have no verifiable public source, and our own measurement does not yet cover a corporate sample. We publish the values we do measure on the proof page, and the reasoning behind this position on why Baki Bilisim.

The version of these seven steps run as a programme at corporate scale is on the AEO service page; the contractual equivalent of the deliverables and acceptance criteria is written down in how we work.

• Dumping the FAQ at the bottom. When questions sit in the last block of the page, the answer is not associated with the page's main topic. The question should be the heading of the relevant section.
• Delaying the answer. A three-paragraph introduction beginning with “In today's world…” pushes the answer far below the fold.
• Marking up invisible content. If the answer written in the structured data is not on the page, the markup is invalid.
• Answering two questions in one passage. A passage cannot be split; a paragraph carrying two ideas is quoted for neither.
• Leaving a term undefined. Not expanding an abbreviation where it first appears makes the passage context-dependent.
• Writing unsourced numbers. A percentage without a source loses you both the reader and the verification process.
• Duplicating templates. Pages built by swapping a few words in the same answer suppress each other.
• Starting without measurement. The effect of an improvement made before a baseline was recorded cannot be shown afterwards.

The free digital asset audit measures the structural side of AEO on your current pages — a single h1, question-form headings and how much of the home page text is present in the raw HTML — and reports the findings item by item. The editorial mistakes above are found by reading the pages; the audit does not assess them.

The technical side of this list — crawling, indexing and speed — is completed by the Core Web Vitals guide and the Schema.org guide.

• AEO — making a page quotable in answer engines.
• GEO — making a brand citable in generative engines.
• SEO — ranking and visibility work in search results.
• SERP — the search engine results page.
• AI Overviews — the AI summary above the search results.
• LLM — large language model.
• Crawler — the software that fetches and processes pages.
• JSON-LD — the recommended syntax for structured data.
• schema.org — the shared structured data vocabulary.
• Rich result — the enriched presentation of a search result.
• Entity — an identified object recognised by search systems.
• E-E-A-T — the experience, expertise, authoritativeness and trust framework.

The rest of the guide series: What is GEO? · Core Web Vitals guide · Schema.org and JSON-LD guide. All of them are listed in reading order in the knowledge centre.

Questions on other topics are collected on the frequently asked questions page.