AEO and GEO: getting cited in AI answers

Answer Engine and Generative Engine Optimisation: being the source an AI answer quotes. Entity graph work (Wikidata, Crunchbase, LinkedIn, directories), answer-first page engineering, and a quarterly prompt audit that measures whether the assistants name you or a competitor.

What are AEO and GEO, and how do they differ from SEO?

AEO is Answer Engine Optimisation: being the source an extractive feature pulls from, such as a Google AI Overview or a featured snippet. GEO is Generative Engine Optimisation: being named inside a synthesised answer from ChatGPT Search, Gemini, Perplexity or Claude. SEO wins a position in a list. Both of these win a mention in an answer that may never show the list at all.

The distinction between AEO and GEO matters less in practice than the vendors selling them suggest, and most practitioners now use the terms interchangeably. Treat it as one discipline with two surfaces: the same entity records, the same answer-first structure, and the same evidence feed both.

The distinction from SEO is the one worth taking seriously. AI Overviews now appear on roughly a quarter of Google queries, with commercial-intent coverage rising sharply year on year, and when one appears the click-through rate for the top organic result falls by around half. Being cited inside the overview is what recovers it: cited pages see materially higher click rates than uncited ones at the same position. Meanwhile only about 12% of the pages AI engines cite are in Google's top ten for that query, which is the clearest evidence available that this is a separate game rather than a by-product of ranking well.

Which levers actually move AI citations?

The ones with measured effects rather than the ones with the best pitch. The KDD 2024 study on generative engine optimisation tested content changes against an answer engine and found the largest lifts came from adding quotations from credible sources, adding statistics, and citing sources inline. The most commonly sold tactic, rewriting copy in a more authoritative tone, underperformed.

• Quotations from identifiable sources: the single strongest lever in the study, at roughly a 41% improvement in visibility within generated answers.
• Statistics with a number and a source attached: roughly a 33% improvement. A claim a model can attribute is a claim it is willing to repeat.
• Inline source citation: roughly a 28% improvement, and the cheapest of the three to apply retroactively to pages you already have.
• Topic clusters over standalone pages: a hub with several linked spokes earns several times the citations of an unconnected page on the same subject, because the engine sees a body of work rather than an assertion.
• Format matters more than it should: numbered comparison and list pages account for a disproportionate share of AI citations, which is why a well-built 'X vs Y vs Z' page for your category is usually the highest-return single asset.
• Original research is the compounding one: a page with first-party data, a survey, a methodology or a calculator earns several times the citations per URL of a page that summarises other people's work, and it keeps earning as others cite it.

Why entity work comes before content

Because a model cannot recommend a business it cannot identify. Before any page is written, the work is making your company a resolvable entity: consistent name, address, phone and description across every record, a Wikidata item, a Crunchbase profile, a LinkedIn page whose opening description matches the Wikidata one word for word, and the founder linked to the company as a person rather than a logo.

The citation data explains the priority. Wikipedia is by a distance the most cited source in ChatGPT answers, and Reddit occupies a similar position for Perplexity. Neither is somewhere you can simply publish your way into, but both sample the same underlying entity graph that Wikidata, Crunchbase and consistent directory records feed. Getting that graph right is slow, unglamorous and largely a one-time cost.

Inconsistent business details across directories remain the most common reason a Cambodian company has no knowledge panel and no assistant recognition: a phone number that differs by a space, three variations of the company name, an address written four ways. We fix that first, then keep it fixed with a quarterly audit, because it is the foundation everything else is measured against.

Wikipedia itself is a year-two conversation, not a deliverable. It requires independent coverage that already exists, and attempting it before that coverage exists produces a deleted article and a poor first impression with editors.

How do you measure something that has no rank tracker?

With a repeatable prompt audit. We agree a set of about 30 prompts covering how buyers actually ask for what you sell, run them across ChatGPT Search, Perplexity, Gemini, Claude and Google's AI Mode, and log each result as cited, mentioned without a link, or absent, together with which competitor was named instead. Repeated quarterly, that becomes a trend line rather than an anecdote.

Answers vary between runs, so a single check proves nothing. The value is in the deltas: which prompts moved from absent to mentioned after a cluster shipped, which competitor stopped appearing after your comparison page went live, and which prompts have never returned you at all, which is where the next quarter's content plan comes from.

Paid monitoring tools exist and are worth their price at scale: entry tiers start around $29 a month for a small prompt set across the major engines, mid-tier tools sit near $85, and enterprise platforms run from roughly $399. We will set one up in your account if the prompt set justifies it. For most Cambodian SMBs the quarterly manual audit is sufficient and the budget is better spent on the pages that fix what it finds.