Generative Engine Optimisation
People now ask an AI assistant which supplier to use, and read the summary above the search results instead of the results. Generative engine optimisation is the work of making your business easy for those systems to find, understand and cite.
This field is new. The companies that run the assistants publish little about how sources are chosen, the answers change from one day to the next, and two people asking the same question can receive different replies.
So we say this at the start: nobody can promise that an assistant will mention you, quote you or recommend you. We can remove the obstacles that stop it happening, and report honestly on what we observe.
How an assistant arrives at an answer
Two separate routes lead to a mention. The work differs for each.
From training data
A language model learned from text collected up to a cut-off date. What it says about you from memory reflects what was written about you before then, and it cannot be edited afterwards.
From live retrieval
Many assistants run a web search, read a handful of pages and write an answer with citations. Here a page published this week can be used.
Through search indexes
Retrieval often relies on a conventional search index. A page that is not indexed by Google or Bing is unlikely to be fetched by an assistant built on top of them.
Through other people
Assistants repeat what independent sources say. Trade publications, directories, reviews and reference sites describing you accurately matter as much as your own pages.
What we change
Most of it is careful publishing. There is no trick that forces a citation.
Crawler access
We read robots.txt, firewall rules and server logs to see which AI crawlers reach the site, such as GPTBot, ClaudeBot and PerplexityBot. Whether to allow each one is your decision; we explain the trade-off.
Content in the HTML
Several AI crawlers do not run JavaScript. Text that appears only after scripts execute may be invisible to them, so key facts are rendered on the server.
Direct answers
Pages open with the answer in plain sentences, followed by detail. A passage that makes sense when lifted out by itself is easier to quote.
Facts stated once, consistently
Company name, what you do, where, prices and specifications are written the same way on your site, business profiles and directory entries.
Structured data
Schema.org markup for the organisation, products, articles and authors. It helps machines read the page. Whether assistants use it directly has not been confirmed by their makers.
Material worth citing
Original figures, test results, definitions and named expert opinion. A page that repeats what ten others say gives an assistant no reason to pick it.
What nobody can promise here
Read this list before paying anyone, including us, for this service.
- That an assistant will name your business in its answer
- That a citation, once gained, will still be there next month
- That what the model says about you will be accurate
- That visits from assistants can be counted completely
- That a proposed convention such as llms.txt is read by any major assistant
- That methods which help today will help after the next model release
How an engagement runs
Measurement is done by sampling, and we present it as a sample.
Question set
Together we write the questions a buyer might put to an assistant. The list is fixed so that later checks compare like with like.
First reading
We put each question to the main assistants several times and record whether you appear, who else appears and which sources are cited.
Changes
Access, page structure and facts are corrected on your site. Errors on third-party sites are reported to their editors, who may or may not act.
Repeat readings
The same questions are asked again each month. You receive a written note of what changed, with the caveat that answers vary between runs.
What we check with
Dedicated tracking products for this field exist, but are young and disagree with each other. We rely mainly on sources we can verify.
- Assistants sampled
- ChatGPTGeminiGoogle AI OverviewsMicrosoft CopilotPerplexityClaude
- Site evidence
- Server log filesGoogle Search ConsoleBing Webmaster ToolsGoogle Analytics 4 referrals
- Page checks
- Rich Results TestSchema Markup ValidatorScreaming Frog SEO Spider
Fair questions about a young discipline
Is this different from SEO?
It overlaps heavily. A site that is crawlable, clearly written and well regarded by others is in a good position for both. The additions are crawler access for AI agents, quotable passages and sampling of answers.
Can you measure visits from AI assistants?
Partly. Some assistants pass a referrer that analytics can record, and others pass none. Many readers take the answer and never click, so visits understate exposure.
An assistant states something false about us. Can you correct it?
Not directly. We can correct the sources it is likely to read and use the feedback channel the provider offers. Whether the answer changes is outside our control.
Should we block AI crawlers?
That is a business and legal choice. Blocking protects your content from reuse and may reduce the chance of being cited. We set up whichever policy you decide on.
Who owns what you produce?
You do: the pages, the markup, the question set and the recorded readings. Search Console and analytics stay under your account.
Start with five questions
Send five questions you would like an assistant to answer with your name in the reply. We will write back within one business day with what we found.