What is answer engine optimisation?

Answer engine optimisation is the work of becoming the source an AI engine quotes when it answers a question instead of listing links. It is not a markup exercise and it is not SEO with a new name.

The term describes a real change in buyer behaviour: a question that used to produce ten links now produces a paragraph naming two or three products. Kunzum measures what those paragraphs say. In September 2026, across 1,500 responses from ChatGPT, Claude, Perplexity and Google AI Overviews, the engines produced 12,518 citations drawn from 1,396 domains. AEO is the discipline of being in that set for the questions your buyers ask.

What answer engine optimisation actually involves

Kunzum's version of the work has four parts, and none of them is technical in the SEO sense. The full version is written up on the method page, with the measured evidence for each step.

Establishing what the engines say now. Real buyer questions, asked across four engines, repeatedly, with the answers and their citations recorded. Without this there is no before, so there is no after.

Matching the buyer's vocabulary. What buyers call a category is often not what the vendor's site calls it. An engine retrieving on the buyer's phrasing will not find a page written in the vendor's.

Making each passage survive alone. A model pulls a paragraph and judges it on its own. A claim in one block and its proof three blocks down survives as an unsupported claim, which is the shape a retrieval system discards.

Working off-site, then measuring monthly. Documentation, aggregators and the places buyers actually discuss the category. Which places is not a matter of taste — it is in the citation data, and in this corpus the answer was overwhelmingly not the trade press.

Why retrieval is the whole constraint

Every citing engine is a descendant of the architecture Lewis and colleagues published in 2020 as retrieval-augmented generation: fetch documents, then answer from them. That is why AEO can work at all, and also exactly where it stops.

In edition one of the Kunzum index, ChatGPT ran a web search on 54% of its answers and Claude on 65%. Perplexity retrieved on 100%. The answers produced without retrieval came from training data, and no amount of publishing in September 2026 could have altered them. Any agency describing AEO without this limit is describing something other than how these systems work.

The engines document the split themselves. As of September 2026, OpenAI’s crawler documentation lists OAI-SearchBot as the crawler that surfaces sites in ChatGPT search and GPTBot separately as the training crawler, Perplexity’s crawler documentation says PerplexityBot “is not used to crawl content for AI foundation models”, and Anthropic documents its own crawlers and how to control them.

What AEO is not

It is not structured data. As of September 2026, Google’s Search Central documentation states: “You don’t need to create new machine readable files, AI text files, or markup to appear in these features. There’s also no special schema.org structured data that you need to add.” Kunzum uses schema markup on this site, but as a parsing aid, not as the mechanism, and nobody should be billed for it as the mechanism.

It is not prompt injection. Kumar and Lakkaraju’s 2024 paper Manipulating Large Language Models to Increase Product Visibility demonstrates that text inserted into a product page can move a model’s ranking of it. That technique exists, it works, and it is not what this is. The distinction is worth stating plainly because the field is young enough that both get sold under the same label.

It is not a guarantee of accuracy. Liu, Zhang and Liang’s 2023 paper Evaluating Verifiability in Generative Search Engines found that generative search engines frequently make statements their own cited sources do not support. Being cited and being represented correctly are two different outcomes, and only the first is directly workable.

Common questions

Is answer engine optimisation the same as SEO?

No. SEO competes for a position in a list of links. AEO competes to be the source a model quotes inside an answer, which the reader may act on without ever seeing a list. They share an indexing layer: a page that is not indexed is invisible to both.

Does AEO require special schema markup?

No. As of September 2026 Google's Search Central documentation states there is no special schema.org structured data needed to appear in AI Overviews or AI Mode, and no new machine-readable file. Markup helps a machine parse a page it already retrieved; it does not cause retrieval.

Can AEO work be measured?

Partly. There is no impression log, so the only honest measurement is to ask the engines the same questions repeatedly and record what they name and what they cite. Kunzum runs 50 prompts across four engines and publishes the raw responses; the residual uncertainty is that engines vary run to run, which is why nothing rests on a single reply.

What cannot AEO do?

It cannot reach an answer the engine produced without retrieval. In Kunzum's September 2026 index, ChatGPT ran a web search on 54% of its answers and Claude on 65%; the remainder came from training data, where nothing published afterwards could have changed the response.