Answer Engine Optimization (AEO)

An answer engine does not hand you ten links and let you choose. It reads the web, decides what is true, writes a paragraph, and names a handful of sources. Your brand is either in that paragraph or it is not, and there is no second page to be on.

Answer engine optimization is the work of getting into that paragraph. It borrows from SEO — clear writing, crawlable pages, credible sources — but the unit of success changes. You are no longer optimizing for a rank. You are optimizing for a mention, a citation, and the sentence the model writes about you.

The discipline is young enough that it goes by four names. AEO, GEO (generative engine optimization), LLMO (large language model optimization), and generative SEO all describe the same work. Choose one and move on; the practitioners arguing about the acronym are not the ones getting cited.

AEO stands for answer engine optimization. An answer engine is any system that responds to a question with a synthesized answer rather than a ranked list — ChatGPT, Perplexity, Claude, Gemini, Copilot, and Google's own AI Overviews and AI Mode.

The term matters mostly because it marks a change in what marketing teams can control. For twenty years the deliverable was a position. Now the deliverable is a description: how the model characterizes your product, which competitors it names in the same breath, and whether it gets your pricing right.

Most of SEO still applies. Answer engines read the same web crawlers index, prefer the same well-structured pages, and trust the same authoritative domains. If your site is unindexable, AEO is not your problem yet.

Four things are genuinely different.

Two mechanisms operate at once, and conflating them is the most common mistake in AEO.

The first is retrieval. When an engine runs a live web search, it issues one or more queries, retrieves pages, and grounds its answer in what it finds. Here, classic SEO is doing the heavy lifting — if you do not surface for the query the model issues, you cannot be cited. The queries models issue are frequently not the queries humans type, which is why keyword research and prompt research are different exercises.

The second is the model's own parametric memory — what it absorbed during training. This is why a brand can be recommended confidently in a chat with web access turned off. Parametric memory moves slowly, is shaped by how widely a brand is discussed across the open web, and cannot be influenced on a campaign timescale.

Practically: retrieval is what you can move this quarter. Parametric memory is what you move over years, by being the thing people write about.

Four metrics carry almost all the signal. Everything else is a slice of one of them.

A workable programme has five parts. None of them are exotic; the discipline is in doing them repeatedly rather than once.

Three patterns come up repeatedly in citation data, and they are not the patterns that win at SEO.

Comparison and alternatives pages punch far above their traffic. A page that honestly compares five tools in a category gets pulled into answers constantly, because that is the shape of the question people ask an answer engine. The same page may attract very little search traffic, which makes it easy to under-invest in.

Definitional content — clear 'what is X' pages with a short, extractable definition up top — is the cheapest citation you can buy. It is also the shape that compounds: once a model has a good definition of your category, your framing travels with it.

First-party data is the only genuinely defensible asset. Anyone can restate a study. If you publish the study, every article about it is a path back to you, and models cite the origin more often than the commentary.

An AEO tool does one core job: run your prompts across the engines on a schedule, and record what came back. Everything else — dashboards, alerts, competitor tracking, reporting — is built on that log.

The differences that matter when choosing one are which engines are covered and how, how many runs per prompt you get (sampling noise is the whole problem), whether you can see the raw answer rather than a score, and whether the methodology behind the score is inspectable.

That last point is why Elmo is open source. A visibility score is a number somebody chose how to calculate. When two tools disagree about your brand — and they routinely do — the only way to know which is right is to read how each one counts. Elmo can be self-hosted, and the counting is in the repository.

Retrieval-driven prompts respond within weeks. If you get added to a roundup the engines already cite, or you publish a page that ranks for the query the model issues, you can see the mention rate move on the next measurement cycle.

Parametric prompts — where the engine answers from memory without searching — move on a timescale of quarters to years, and only as a by-product of being genuinely well known. No tactic accelerates this. It is the accumulated residue of being written about.

The honest framing for a marketing team: AEO is a measurement discipline first and a content discipline second. Most of the early wins come from finding out you are invisible on a question you assumed you owned.