Grow · Intermediate

AI search optimisation (GEO, AEO): what it is and what a company should do

Three names are sold for one job: making a website usable as a source when an AI assistant answers a question. Here is what that job consists of, what nobody can promise, and the order in which to do it.

6 min read

The glass lens of a lighthouse lamp.

Photo: ChrisEdwardsCE, public domain (CC0)

In this guide

Short answer

AI search optimisation means making your website usable as a source for the answers AI assistants give. GEO (generative engine optimisation) and AEO (answer engine optimisation) are other names for the same work. It consists of five things: readable pages, clear company facts, structured data, direct answers and mentions on other sites.

3names for one job: AI search optimisation, GEO and AEO
5steps of work, in a fixed order, from access to mentions
2ways an assistant answers: from training memory or from pages fetched live

What is AI search optimisation, and are GEO and AEO something different?

They are three names for one task. AI search optimisation is the work of making a website easy for AI assistants to read, understand and cite. Generative engine optimisation (GEO) is the same work, named after the systems that generate an answer instead of listing links. Answer engine optimisation (AEO) is the same again, named after the result: the reader gets an answer, not a list.

Offers under each name contain much the same tasks. Ask what will be changed on the site, and compare that.

It does not replace search engine optimisation either. An assistant that searches the web while it answers can only use pages that can be found and fetched, so a site that is blocked, thin or unclear fails in both places. Most of the work below is good SEO with stricter demands on clarity.

What do AI assistants need from a website?

They need five things, and each can be checked. An assistant can only repeat what it was able to fetch and understand, so the list starts with access and ends with reputation.

What an assistant needs and how to check it
Need What it means How to check
Crawlable pages Not blocked in robots.txt or by a firewall, and the content is in the HTML Open a page with scripts switched off: is the text there?
Clear company facts Who you are, what you sell, for whom and where, stated the same way everywhere Read the home page as a stranger, then compare it with your directory and map entries
Structured data The same facts as schema.org markup: organisation, products, questions Run a page through a structured data testing tool
Direct answers A real question as the heading, the answer in the first sentence below Take five customer questions and look for the page that answers each
Mentions by others Trade press, customers, associations and directories that name you Search for your company name and ignore your own site

Two of these need a word of explanation.

Content in the HTML. Some websites send an almost empty page and let a script fill in the text in the visitor’s browser. A person sees a complete page. A crawler that does not run scripts sees an empty frame. Do not count on a crawler running your scripts: the main text, the product data and the contact details belong in the page as it leaves the server.

Structured data. Markup adds nothing new. It repeats the facts of the page in a form that needs no interpretation: this is the company name, this is the address, this is the product and its article number. The vendors of assistants do not publish how much weight they give it, but search engines document that they use it, and it is cheap to add.

What nobody knows: ranking rules and guarantees

Nobody outside the vendors knows how an assistant chooses its sources, and nobody can guarantee a mention. The rules are not published, the systems change often, and the same question can produce a different answer on another day. Anyone who claims to have the formula is generalising from their own tests.

Treat an offer of a guaranteed place in an assistant’s answers as a warning sign, and spend on work that is useful anyway: a site that people, search engines and assistants can all read.

What should a company do first? Five steps in order

The order is access, facts, markup, answers, mentions. Each step makes the next one worth doing: there is no point in writing good answers on pages a crawler cannot fetch.

  1. Open the doorCheck robots.txt, the firewall and the hosting settings for blocks on search and AI crawlers. Make sure the main text of each page is in the HTML.
  2. State the facts once, clearlyWrite on the home page and the about page what you make or sell, for whom and where. Use one spelling of the company name and one address everywhere.
  3. Add structured dataMark up the organisation, the products and the question pages with schema.org, carrying the same facts as the visible text.
  4. Answer real questionsCollect the questions customers ask sales and support. Give each a page or a section with the question as heading and the answer directly beneath it.
  5. Earn mentionsBring directory and association entries into line with the site. Then work on case studies with customers and articles in the trade press.

Steps one to three are technical and finite: once done, they stay done until the site changes. Steps four and five are ongoing work, and step four is mostly writing. If nobody in the company has time for it, that is the part to hand to someone who does content and copywriting for a living.

How we do it

We begin by reading the site the way a crawler does: what is blocked, what is missing from the HTML, which facts contradict each other. Then we fix access and markup, and rewrite or add the pages that should answer your customers’ questions, in every language you sell in. We also say plainly what we cannot promise, which is the mention itself.

How do you know whether AI search optimisation worked?

You ask the assistants the questions your customers ask, before and after, and compare. A simple protocol for that, with ten questions and a spreadsheet, is described in how to check whether AI assistants mention your company.

Expect two speeds. An assistant that fetches pages while it answers can pick up a changed page once its crawler has visited again. What an assistant knows from training changes only when its vendor trains a new model, and that is outside your control.

  • Search and AI crawlers are not blocked in robots.txt, the firewall or the hosting settings.
  • The main text of every important page is in the HTML without scripts.
  • The home page says in its first lines what the company does, for whom and where.
  • Name, address and product range read the same on the site and in every profile.
  • Organisation and product markup is present and passes a testing tool.
  • Each frequent customer question has a heading and a direct answer on the site.

Questions we hear often

No, but they overlap heavily. SEO aims at a place in a list of links, GEO at being used as a source inside an answer. Crawlable pages, clear content and mentions by others serve both.

No. The vendors decide which sources an answer uses and do not publish the rules. An agency can only remove the reasons you are left out.

Usually not. Most of the work is done on the existing site: unblocking, clearer texts, markup and question pages. A rebuild is only worth considering when the content cannot be delivered in the HTML.

It differs by assistant. Answers based on live web search can change once crawlers have revisited your pages; knowledge from training changes only with a new model.

Not by itself. It states your facts without ambiguity, which helps any machine reader, but readable pages and direct answers matter first.

Want this sorted without doing it yourself?

Tell us what you are working on. A person replies within one working day.