Answer engine optimization: the practitioner's guide to AEO

"Answer engine" covers six surfaces that behave differently. Knowing which one you are aiming at decides how you write the passage.

• Featured snippets. Google lifts a paragraph of roughly forty to sixty words, a numbered or bulleted list, or a table, from a page already ranking on the first page for that query. You cannot mark a page up to request one. You can make the extractable block obvious, and you can only win one if you already rank.
• People Also Ask. Short two- or three-sentence answers to related questions, each with its own source. These are the cheapest entry point, because the questions are published for you in the box itself and the bar is a clean, direct answer rather than a whole ranking page.
• AI Overviews and AI Mode. Google's generative answers, served from the ordinary Search index and the ordinary Googlebot crawl. Google's documentation states there are no additional requirements or special optimisations to appear in them; the only content controls are nosnippet, data-nosnippet and max-snippet.
• Chat assistants with web search. ChatGPT (fetching as OAI-SearchBot and ChatGPT-User) and Perplexity (PerplexityBot and Perplexity-User) retrieve live and cite. They take a passage they can attribute, which is why a named author, a date and a checkable fact matter more here than anywhere else.
• Voice assistants. One answer, read aloud, no list. The winning passage is short, unqualified and self-contained, and a hedge-laden paragraph is unreadable in this format.
• Site-level answer boxes. Your own search, help centre and product Q&A. Nobody thinks of these as AEO, but they are the surface where a badly written answer costs you a support ticket.

The overlap is high enough that one well-built passage serves all six. The differences are in measurement, not in craft: Search Console hints at the first three, and only direct probing shows the fourth.

SEO earns a position for a URL and a click from a person. AEO earns a citation for a passage, and the person may never visit. That difference runs through every decision.

What stays the same. The page still has to be crawlable and indexed, still needs a clear topic, and still benefits from the same trust signals: a named author, a reviewed date, sources, and a site with depth on the subject. Pages that already rank tend to be the ones quoted, because the answer is drawn from results the engine retrieved for the query in the first place.

What changes.

• The target is a question with a knowable answer, not a head term. "cctv installation" is an SEO target; "do I need permission to install CCTV at home" is an AEO target.
• Success is measured in snippet placement and citations, not position alone. Search Console hints at the first; probing the assistants shows the second.
• Structure matters more than length. One question per heading, the answer in the first sentences, then the evidence.
• Original facts beat comprehensive coverage. An engine that already has the generic answer only needs you if you add something it cannot get elsewhere.

Generative engine optimization (GEO) is the newer term and refers specifically to being cited by generative AI: AI Overviews, ChatGPT, Perplexity, Copilot. AEO is older and wider. It grew out of featured snippets and voice search, and it covers any surface that returns an answer, generative or not.

In practice the two overlap almost completely. The passage that wins a featured snippet is the same shape as the passage a language model quotes: a direct statement, a heading phrased as the question, concrete supporting facts, clean markup. Where they differ is measurement and access.

• Measurement. Search Console reports impressions, clicks and average position. It never says a snippet took your text, and it cannot see assistants at all, so snippet placement is inferred from position and click-through rate and assistant citations have to be asked for directly.
• Access. Assistants run their own fetchers, and robots.txt can allow or block each one independently of Googlebot. Google-Extended is not a fetcher: it governs whether pages Googlebot already has are used for Gemini, and blocking it does not remove you from AI Overviews.

Treat them as one discipline with two scoreboards. Our generative engine optimization guide covers the assistant-specific side; this page covers the shared foundation.

Here is a constructed before and after, to show the shape of the change rather than to claim a result. The question is one a UK CCTV retailer's customers ask: "Can I put CCTV on my house if it films the street?" In the before version it sits inside a long, general home-CCTV article, several screens down, under a subheading about privacy.

The after version changes four things and almost nothing else.

• The H2 becomes the question, worded as people ask it.
• The first sentence under it answers: yes, you can, but if the camera captures images beyond your own boundary, UK data protection law applies and you take on the obligations the ICO sets out for domestic CCTV users.
• The next paragraph gives the specifics a machine can verify: what those obligations are (signage, limiting the field of view, handling requests for footage) and the source they come from.
• The question and answer are repeated in FAQPage schema, so the pairing is explicit in the markup, not just implied by layout.

Nothing about the site's authority changed. What changed is that a machine reading the page can find the question, lift a complete answer without stitching paragraphs together, and check it against a named source.

This is the single most out-of-date claim in AEO advice, so it is worth being blunt about.

In August 2023 Google limited HowTo rich results to desktop and narrowed FAQ rich results to, in its own words, "well-known, authoritative government and health websites". HowTo was retired outright the following month, on 13 September 2023. FAQ went the same way three years later: Google's changelog records that the FAQ rich result "will no longer appear in Google Search starting May 7, 2026", and in June 2026 it removed the FAQPage documentation because the feature "is no longer shown in Google Search results". So no site gets an expanded FAQ listing now, government and health included, and anyone still selling FAQPage markup as a search feature is selling something Google has withdrawn twice.

Add it anyway, for three reasons that survive the change.

• It states the question-answer pairing explicitly, so a parser does not have to infer it from your layout. Retrieval works on passages, and a marked-up pair is an unambiguous passage boundary.
• Writing it forces the answer-first shape into the prose. A question you cannot answer in one self-contained paragraph is a question your page does not really answer.
• It is the same data as the visible block, which is what keeps it honest.

The rule that has not changed: markup must match what the page shows. Google treats structured data describing content a visitor cannot see as a violation, and a parser that catches your schema disagreeing with your copy has a reason to trust both less. Every FAQ block on this site is rendered from the same record that emits the schema, so they cannot drift apart.

What FAQ schema was never going to do is make a weak answer worth quoting. The markup describes the answer; it does not improve it.

• Start from questions with an answer surface. Pull queries from Search Console that begin with how, what, can, does or which, or that already show a snippet or People Also Ask box. Keyword tools give you volume; the result page tells you whether an answer is being extracted at all.
• Write the answer first. Under a heading that is the question, state the answer in forty to sixty words. Then expand. Do not build up to it.
• Make it verifiable. Add the number, date, model, regulation or measurement that lets a fact-checker confirm you. First-party data, your own specifications and your own support tickets, is the strongest version of this, because no competitor holds it.
• Mark it up. FAQPage schema for question pairs, HowTo for procedures, Article with a real author and dateModified.
• Open the door. Check that robots.txt allows the AI fetchers you want citing you; the free AI crawler check reads it for you.
• Measure both scoreboards. In Search Console, track impressions, position and click-through rate for the question itself. For assistants, ask the question and record whether your domain is among the cited sources.

Run a page through the AEO checker to see which of these it already does before you change anything.

These are failure modes we can describe precisely because we ran into them on live sites.

• Trusting a scored check to gate. In August 2026 an article we published to one of our own group's CCTV sites shipped with a markdown link truncated mid-URL, twice. The fact-check caught it and said so verbatim: "the link appears twice as cctvmaster.co. which is not a valid URL". It published anyway, because the check scored 84 against a passing bar of 78. Model judgement raises the average; only deterministic code sets a floor. We now run a pre-publish link guard, and the sweep of the back catalogue it triggered found 46 dead links across 16 of 92 live articles. If a defect must never ship, it needs a rule, not a rating.
• Assuming robots.txt does what it looks like it does. A crawler obeys only the most specific group that matches its user agent and ignores every other group — that is in Google's robots.txt specification, and it is easy to forget. We added crawl rules to the wildcard "User-Agent: *" group on a site we run, and every named group in the file — Googlebot, GPTBot, PerplexityBot, ClaudeBot — carried on ignoring them, because a named group replaces the wildcard group rather than extending it. Verify by fetching /robots.txt and reading the group for the crawler you care about, never the wildcard group. Our free AI crawler check does this per agent.
• Marking up an answer the page does not give. FAQPage schema for a question the copy skirts is a contradiction a parser will resolve against you.
• Answering under a heading nobody searches. "Legal considerations" is a heading; "Can I point a CCTV camera at the street?" is a query. The heading travels with the passage into the chunk that gets scored, so it is part of the answer, not a label above it.
• Measuring once and calling it a baseline. Assistant answers vary between runs. One check tells you almost nothing; the same questions on a fixed schedule tell you a great deal.

Nobody can give you a date, and any provider who does is guessing. What you can do is watch the signals in the order they actually move, so you know whether the work is progressing or stuck.

• Crawled — the page appears in server logs and in Search Console's crawl data. Days, if the site is healthy.
• Indexed — Search Console reports it as indexed. This is where most programmes silently fail. Across four sites we publish for, measured on 25 August 2026, the indexed share of published articles ranged from 25 of 27 down to 12 of 25, on the same pipeline in the same week.
• Impressions — the page starts appearing for the question. Usually the first honest evidence that the passage is relevant.
• Position — it climbs into the range where extraction becomes possible at all. Featured snippets are drawn from first-page results.
• Extraction or citation — the snippet, the People Also Ask entry, or the assistant naming you.

Each stage gates the next, which is why "how long does AEO take" has no single answer: a site stuck at stage two has an indexing problem wearing an AEO costume. Diagnose by finding the first stage that is not happening, and fix that one.

On cadence: we probe Perplexity nightly and ChatGPT weekly for the sites we run, and skip the ChatGPT probe on other days rather than substituting a cheaper measurement, because two different methods in one column produce a trend line that means nothing. Whatever you choose, keep it identical week to week.

There is no single AEO tool, because the job has three parts that different software does well.

• Finding the questions. Google Search Console is the strongest free source, because it shows the queries you already appear for. Paid keyword databases add volume estimates and the People Also Ask tree for questions you do not yet rank for.
• Writing and structuring the answers. Auditing tools tell you what to change; producing the passage, the FAQ block and the schema is still a writing job unless something writes it for you.
• Measuring citations. Search Console cannot tell you whether ChatGPT or Perplexity named you. That needs something which asks the assistants your questions and records the sources they return.

Pick the tool that covers the part you are missing. To see where a page stands before buying anything, the free AEO checker scores a URL against nine fixed on-page checks, and the AI visibility checker puts buyer questions to ChatGPT and Perplexity and reports whether your domain is cited. If you want all three parts done for you, our AEO services page describes how the engine handles each one.

IT Master is an autonomous content engine, and AEO is built into how each article is produced rather than bolted on afterwards.

Discovery starts from real search demand: Search Console queries for the customer's own site, plus keyword, result-page and People Also Ask data from DataForSEO. Research pulls competitor pages, forum threads and, where it exists, the customer's first-party data: product specifications, datasheets, support Q&A. That last source is what gives an answer something an engine cannot find elsewhere.

A Claude model writes the draft. Then a validation gauntlet run by models from a different vendor, Gemini and GPT, fact-checks it, scores its novelty against the competitors, critiques E-E-A-T and screens for AI tells, with up to three revision rounds. Across 186 Standard-tier runs, 21% of drafts cleared every check on the first attempt with no revision, and 51% got there within the three revisions allowed; the rest never published.

Passing articles publish with an FAQ, JSON-LD schema, internal links and a hero image, and IndexNow pings the search engines. A nightly loop refreshes decaying pages and records Search Console data alongside whether GPT and Perplexity cite the site.

It runs today on live UK CCTV trade and retail sites. You pay per published article from a prepaid balance, with no subscription; see pricing and how it works.

Primary sources for the claims on this page, all read on 4 September 2026.

• Google Search Central, documentation changelog — the FAQ rich result "will no longer appear in Google Search starting May 7, 2026", and the FAQPage documentation was removed in June 2026 because the feature "is no longer shown in Google Search results": developers.google.com/search/updates
• Google Search Central, "Changes to HowTo and FAQ rich results", August 2023 — the announcement of both changes: developers.google.com/search/blog/2023/08/howto-faq-changes
• Google Search Central, "AI features and your website" — that AI Overviews and AI Mode need no separate optimisation and use the ordinary Googlebot crawl: developers.google.com/search/docs/appearance/ai-features
• Google Search Central, "Robots.txt specification" — that a crawler obeys only the most specific matching group: developers.google.com/search/docs/crawling-indexing/robots/robots_txt
• OpenAI, "Bots" — OAI-SearchBot, ChatGPT-User and GPTBot: developers.openai.com/api/docs/bots
• Perplexity, "Bots" — PerplexityBot and Perplexity-User: docs.perplexity.ai/guides/bots
• Google Search Central, "General structured data guidelines" — that markup must describe content visible to the reader: developers.google.com/search/docs/appearance/structured-data/sd-policies
• Schema.org, the FAQPage type: schema.org/FAQPage
• Information Commissioner's Office, "Home CCTV systems" — the guidance behind the worked example above: ico.org.uk/for-the-public/home-cctv-systems
• Our own measurements are published in the AI Citation Index and the AI Content Index.