Answer Engine Optimization (AEO): The Complete Guide

What is answer engine optimization?

Answer engine optimization is the practice of getting a business cited, referenced and recommended by AI answer engines such as ChatGPT, Perplexity, Google AI Overviews, Gemini and Copilot, rather than only ranked by a traditional search engine. It combines technical work that makes content readable to AI crawlers, structured data that establishes who you are, and writing that a retrieval system can quote cleanly.

The shorter version: SEO competes for a position in a list. AEO competes to be the source inside the answer. Those used to be the same contest and they no longer are.

Why does AEO exist as a separate discipline now?

Because the link between ranking and being found broke, and it broke recently enough that most firms have not adjusted. Ahrefs found in February 2026 that only 38% of the pages cited in Google AI Overviews rank in the top ten organically. The majority of cited sources are pages that traditional SEO would call losers.

At the same time the clicks moved. Comparing December 2023 with December 2025 across 300,000 keywords, Ahrefs measured the top organic result's click-through rate falling from 7.3% to 1.6% on keywords that trigger an AI Overview. Position one held. The click did not.

For professional services this hits harder than average, because professional services buying starts with questions rather than vendor names. Someone asks whether a framework applies to them, what a process involves, what it typically costs. Those are informational queries, and informational queries are exactly what answer engines resolve without sending anyone anywhere.

What is the difference between AEO, GEO and SEO?

SEO earns a position in a ranked list of results. AEO earns a citation inside a generated answer. GEO, generative engine optimization, is used almost interchangeably with AEO in practice, though it tends to emphasise content and formatting where AEO also covers the technical and entity layers.

The distinction between AEO and GEO matters less than the marketing around both suggests. The distinction between either of them and SEO matters a great deal, because they fail in different ways.

The two are complementary rather than alternatives. The technical health, topical depth and demonstrable expertise that earn rankings are largely the same signals a retrieval system uses to decide what to trust. Strong SEO makes AEO easier. It just no longer delivers the outcome on its own.

How do AI engines decide which sources to cite?

Differently from each other, which is the single most expensive thing to get wrong. A study of 34,234 AI responses found only around 11% of domains are cited by both ChatGPT and Perplexity, with brand citation rates of 0.59% on ChatGPT against 13.05% on Perplexity. Optimising for one platform tells you very little about your position on another.

What holds across all of them

• •The content has to be reachable. If the crawler that feeds the product cannot fetch your page, nothing else in this guide matters.
• •It has to be parseable without running JavaScript. Most AI crawlers do not execute it, so content that appears only after hydration is invisible.
• •The source has to be identifiable. A model that cannot resolve who published something has a reason not to name it.
• •The passage has to stand alone. Retrieval pulls passages rather than pages, so a paragraph full of unresolved references cannot be lifted.
• •Recency counts. Pages updated within three months average roughly 6 citations against 3.6 for stale pages, and Perplexity weights content under 30 days old around 3.2 times higher.

Where third-party sources come in

For recommendation queries specifically, the models frequently do not cite anybody's website. Ahrefs found across 750 top-of-funnel prompts that best-of lists were the most prominent page type in ChatGPT's sources, including for agency recommendations. Reddit is the single most cited domain in AI answers overall, at roughly 40% of citations in one 150,000-citation sample.

You cannot outrank a best-of list from your own domain. You get onto it. That is a separate workstream from on-site optimization, and any AEO programme that ignores it will underperform on exactly the queries that produce clients.

What is the technical layer of AEO?

This is the part that either works or does not, with very little in between, and it is where most engagements find the actual problem.

robots.txt, per bot

The platforms document their search crawlers and their training crawlers as independent systems. This is the only technical lever OpenAI and Anthropic officially name for visibility in their products, so it is worth getting exactly right.

The practical point is that you can stay fully visible inside ChatGPT and Claude answers while opting out of model training, because those are different crawlers. Firms that blanket-blocked AI bots in 2024 frequently removed themselves from the products as well, and most have not noticed.

Check that a CDN or security plugin is not blocking these independently of your robots.txt. Cloudflare bot rules are a common culprit. OpenAI notes robots.txt changes take around 24 hours to propagate.

Rendering

Load your page with JavaScript disabled, or fetch it with curl, and read what comes back. If your service descriptions, your FAQ answers or your case study numbers are missing, they are missing for most AI crawlers too. Client-rendered accordions are a frequent offender: a component that only inserts an answer into the DOM once a user clicks has published nothing as far as a crawler is concerned.

Indexing and eligibility

Google documents three conditions for appearing in AI Overviews and AI Mode: the page must be indexed, it must be snippet-eligible, and the site must be included in Search generative AI features, which is a setting in Search Console. Submit your sitemap to both Google Search Console and Bing Webmaster Tools. Bing indexing is the gateway to Copilot, and enabling IndexNow means Bing and Copilot pick up changes in near real time.

llms.txt, honestly

A markdown file at your site root summarising the business and linking to key pages. It costs an hour and we do ship one. It is not a citation lever, and the industry claiming otherwise is not reading the evidence: Google confirmed on 15 June 2026 that it does not affect Search, adoption sits at about 10% of 300,000 domains studied, and one analysis logged 408 llms.txt fetches across 500 million AI bot events.

How should content be written for AI citation?

Mostly the way it should be written for people who are in a hurry, which is a happier answer than the industry usually gives.

Answer first

Lead each page and each section with a direct answer in the first 40 to 60 words. Google is explicit that no special writing style is required for its AI features, so this is not a rule imposed by a platform. It holds because passage retrieval on Perplexity and ChatGPT rewards it, and because burying the answer under three paragraphs of throat-clearing loses human readers too.

Headings as questions

Phrase headings the way people actually ask, not as topic labels. How long does AEO take to work, not Timelines. Conversational query phrasing is how AI search gets used, and matching it costs nothing.

Self-contained sections

Each section should make sense lifted out of the page. That means resolving pronouns, not referring to what was covered above, and repeating the subject where a human editor might cut it. Google has explicitly mythbusted the idea that content must be chunked for AI, and that is correct as far as it goes. The reason to do it anyway is that retrieval on non-Google surfaces pulls passages, and a passage with unresolved references cannot be quoted.

Tables and specifics

Real HTML tables, never screenshots of tables. Structured comparisons are disproportionately favoured for extraction. The same goes for named statistics with sources and dates, and for concrete technical terms. Specificity is what a model treats as substantive rather than generic, and it happens to be what an analytical buyer treats the same way.

What is the entity layer, and why does it decide citations?

An answer engine will not confidently name a business it cannot confidently identify. The entity layer is the work of making your firm resolvable: a specific organisation with a name, a founder, a location in the web of references, rather than a string of text that appears on a website.

• •Organization schema with name, url, logo, description, founder, contact point, and knowsAbout covering what you actually do.
• •Person schema for the founder or named experts, with sameAs pointing at LinkedIn and any other verifiable profile.
• •Service schema on each service page, so a query about one service can be matched to the right page rather than the homepage.
• •Article schema on posts, with dateModified maintained honestly, since recency is a confirmed factor.
• •Identical business name, description and founder name across your site, LinkedIn, and every directory profile carrying your name.

That last point does more work than the schema does. Inconsistent entity data across the web is the most common reason a firm with a technically sound site still does not get named.

What about FAQ schema?

Keep the questions and answers. Treat the markup as optional. Google retired FAQ rich results on 7 May 2026 and its generative AI guidance states no special schema is required for AI Overviews or AI Mode. The widely quoted claim that FAQ schema makes a page 3.2 times more likely to appear in AI Overviews traces back to a source page that no longer exists and has been publicly flagged as unsupported. The visible content earns the citation. The markup is housekeeping.

How do you measure AI visibility?

Start with the two official reports, which are free and which most firms have never opened.

• •Google Search Console added a generative AI performance report on 3 June 2026. Impressions only at launch, rolling out by region.
• •Bing Webmaster Tools has an AI Performance report covering Copilot citation counts, cited URLs and grounding queries, expanded in June 2026 to include intent labels and citation share. It is the deepest official AI citation reporting any platform currently offers.

On top of that, run prompt testing manually: ask each surface the questions your buyers actually ask, record who gets named, and repeat it on a schedule. Benchmark against three named competitors rather than tracking yourself in isolation, because absolute citation counts mean very little without knowing who is taking the space.

Paid tracking tools exist and some are good. None of them are necessary on day one, and starting with the free official reports means you are measuring what the platforms themselves report rather than a vendor's approximation of it.

What are the most common AEO mistakes?

Blocking the wrong crawlers

Blanket-disallowing AI bots to opt out of training, and removing yourself from ChatGPT and Claude answers at the same time. These are separate crawlers and the platforms document them separately.

Treating AEO as an SEO add-on

One line item inside a retainer, usually described as AI SEO, usually meaning nothing was done differently. If AI visibility is a discipline it needs its own baseline, its own workstream and its own measurement.

Selling llms.txt as the mechanism

It is a cheap hedge, not a lever. Leading a pitch with it in front of a buyer who checks is a fast credibility loss, and the buyers in professional services check.

Promising inclusion

Google states plainly that indexing and serving are never guaranteed. OpenAI promises eligibility rather than inclusion. No architecture can ensure appearance in AI answers, and any agency claiming otherwise is either not reading the documentation or assuming you will not.

Publishing once and stopping

Content older than roughly 90 days enters a retrieval decay window. AEO is maintenance, not a project with a completion date, which is also the honest reason this work suits a retainer rather than a one-off engagement.