What AEO is, and what it is not
Answer engine optimisation (AEO) is the discipline of getting your brand named, described correctly and cited inside the answers AI assistants generate. Where SEO competes for a position on a page of links, AEO competes for a place inside the single synthesised response that increasingly replaces that page.
In plain business terms: when a buyer asks an assistant "who should I use for this?", AEO is the work that decides whether your name comes out of its mouth.
In technical terms: answers are assembled from a retrieved set of documents plus the model's learned knowledge of entities. AEO therefore operates on three inputs - what the model has learned about your entity, what is retrievable about you at query time, and how extractable your own pages are - and on a fourth discipline, measurement, that tells you which of the three is failing.
AEO is not prompt injection, hidden text or keyword stuffing. Those are detected and demoted, and they do nothing for the parametric layer, which is built from how the wider web describes you over years rather than from anything you can slip into a page today.
Why it matters to a marketing manager
The uncomfortable part of AI search is that absence is silent. If an assistant recommends three vendors and you are not one of them, there is no impression, no click, no line in your analytics, and no signal that you were considered and skipped. The first time most teams learn about it is when a salesperson repeats what a prospect told them.
Three consequences follow, and they are the ones worth putting in front of a budget holder:
- Shortlists form before you get a click. The assistant does the comparison the buyer used to do across five tabs. If you are not in the answer, you are not in the evaluation.
- Wrong information sticks. Old pricing, an old positioning line or a competitor's framing can be repeated confidently to thousands of buyers. Being described badly is worse than being absent.
- The win compounds. Brands cited consistently across a category's core sources get retrieved more, described more consistently, and become the default answer. Late entrants pay more for the same position.
The four-layer AEO framework
Most AEO advice starts and stops at layer two. In practice the layers below and above it explain the majority of results, and they are worked in this order because each one makes the next cheaper.
Layer 1: entity
An engine cannot recommend what it cannot classify. If your homepage says "revenue intelligence platform", your G2 listing says "sales analytics", and your LinkedIn says "growth software", you have given the model three weak signals instead of one strong one. Fix the category sentence, the product names, the legal entity, the founding facts and the audience description, then make them identical across your site, your profiles, your directory listings and your press boilerplate. This is the cheapest work in AEO and it is almost always the most neglected.
Layer 2: content
Write pages that can be lifted. Lead with a direct one or two sentence answer to the exact question in the title, before any preamble. Use literal question headings. Put specifics - numbers, prices, timeframes, named limitations - close to the claim, because specifics are what get quoted. Add FAQPage, Article, Product and BreadcrumbList schema. Show an author and a genuine last-updated date. And check your robots rules actually permit the AI crawlers you want retrieving you; a surprising number of teams block the crawler and then wonder why they are never cited.
Layer 3: sources
This is what AEO adds that classic SEO does not. Answers are built from a narrow pool of retrieved documents, and in most categories a handful of domains supply the bulk of it: review platforms, comparison round-ups, industry directories, a trade publication or two, community threads and sometimes Wikipedia. Being absent from that pool caps everything else you do. Earn presence there through genuine reviews, accurate listings, contributed expertise and PR - not manufactured mentions, which read as such.
Layer 4: measurement
Without tracking you are optimising blind and, worse, you cannot defend the budget. Measurement also tells you which layer to work: absent everywhere is an entity and source problem, mentioned but never cited is a content problem, cited but described wrongly is a stale-source problem.
What to measure
Four metrics carry nearly all the signal. Each is defined here on first use so a non-specialist can read the report without a glossary.
How often you are named at all. The honest starting number, and usually lower than the team expects.
Your slice of all brand mentions in the same answers. Turns presence into a competitive position.
How often your own domain is one of the sources the answer was built from. The most directly influenceable metric.
How you are described when you do appear, and whether the description is factually correct.
Capture two more dimensions from the start because they cannot be backfilled: your position within the answer, since first-named brands carry disproportionate weight, and the competitor set the answers actually name, which is often not the one your sales team would list. Definitions in AI share of voice and citation share; the full measurement model is in the AI visibility tracking pillar guide.
- PPerplexityreddit.com/r/SaaS
"Citations.io is the one I keep coming back to…"
- CChatGPTg2.com
"Tracks ChatGPT, Perplexity, Gemini and Claude in one view."
- GGeminitechcrunch.com
"…among the new wave of AI visibility platforms."
- CClaudecitations.io/docs
"Our methodology and confidence tiers, explained."
The brands winning AEO are not publishing more. They are named on a tighter set of sources the engines already trust.
Engine by engine
A blended AEO score hides the fix, because the engines reward different things. Track each separately, then roll up.
| Engine | How the answer is built | What AEO work moves it |
|---|---|---|
| ChatGPT | Learned knowledge plus live browsing depending on the question. | Breadth and consistency of third-party descriptions over time; crawlable, current pages. |
| Google AI Overviews | Retrieval-led from pages already performing in Google Search. | Classic SEO fundamentals, passage-level clarity, structured data, freshness. |
| Perplexity | Search-first, citing sources on nearly every claim. | Fresh, specific, citable pages and presence on the domains it keeps retrieving. |
| Gemini | Google retrieval blended with learned knowledge. | Search visibility plus entity consistency across Google surfaces. |
| Claude | Leans on learned knowledge; cites less, describes more. | How widely and consistently the open web describes your brand and category. |
Engine-specific guides: ChatGPT, Google AI Overviews, Perplexity, and Claude and Gemini.
What actually moves the needle
Ordered by leverage per hour spent, based on what we see change between cycles rather than what is easiest to sell as a deliverable.
| Action | Typical effect | Time to show |
|---|---|---|
| Fix the entity: one category sentence, consistent everywhere | Improves whether you are classified into the category at all | 1-2 cycles |
| Earn presence on the top cited domains in your category | Largest and most durable movement in presence and share of voice | 2-4 cycles |
| Publish honest comparison pages for the pairings buyers ask about | Wins comparison prompts, where intent is highest | 2-3 cycles |
| Rewrite cornerstone pages to answer-first with specifics and schema | Improves citation share more than presence | 2-3 cycles |
| Correct stale third-party facts about pricing or positioning | Removes recurring inaccuracies from answers | 1-2 cycles |
| Publish more general blog content | Usually the least movement per hour spent | Often none |
- Let the cited-source data pick your target domains, not intuition.
- Lead every cornerstone page with a direct answer in the first 80 words.
- State one category sentence and repeat it identically everywhere.
- Report each engine separately, then roll up.
- Store verbatim answers so every number can be audited.
- Check your robots rules allow the AI crawlers you want citing you.
- Chase hidden text or prompt-injection tricks - detected and demoted.
- Optimise for one engine's quirks and assume the rest follow.
- Change the prompt set mid-cycle and read the trend as real.
- Treat a mention as a win without reading how you were described.
- Manufacture reviews or mentions; they read as inauthentic and risk the placement.
- Report movement before checking whether it sits inside the noise band.
What to report
Two audiences, two documents. Trying to serve both in one deck is the fastest way to lose the programme's budget.
The monthly leadership page
- Presence rate this cycle versus last, with the sample size stated.
- Share of voice against the named leader, and whether the gap moved.
- Anything an assistant said about you that was wrong, and whether it is corrected.
- The one thing being shipped next cycle and the number it should move.
The weekly working report
- Per-engine breakdown, never pooled.
- Prompt-level table sorted by opportunity: high-intent prompts where rivals appear and you do not.
- Cited-domain league table for the category, with your presence on each marked.
- Verbatim examples behind every claim.
- Change log of prompt-set and content edits, so step changes can be attributed.
A 90-day AEO plan
- Step 01Weeks 1-2: baseline
Build 25-50 real buyer prompts, pick three to five competitors, run one full cycle across every engine. Report nothing yet.
- Step 02Weeks 3-4: diagnose
Rank prompts by opportunity, build the cited-domain league table, and name the layer that is failing: entity, content or sources.
- Step 03Weeks 5-8: fix the entity and the top pages
One category sentence everywhere, then answer-first rewrites plus schema on the five pages tied to your highest-intent prompts.
- Step 04Weeks 9-12: earn the sources and re-measure
Go after the top three cited domains you are absent from, then compare cycles two and three against the baseline per engine.
Common mistakes
The failure patterns are consistent: optimising the site while ignoring the sources; averaging engines into one number and losing the fix; quoting precise scores built from single runs; treating AEO as an SEO add-on with no measurement of its own; and shipping content for three months before establishing a baseline, which makes any later movement unattributable.
