Refract Group2026
Glossary
GEO (generative engine optimisation)
GEO (generative engine optimisation) is work to get a company named, and its pages cited, in answers written by generative AI such as ChatGPT, Perplexity, Gemini and Google’s AI Overviews. Nobody can guarantee a citation.
- Buyer asks a questionChatGPT, Gemini, Perplexity
- Engine searchesOften several related queries
- Your page picked?Others cited instead
- Answer writtenSources paraphrased together
- Named and citedLink shown beside the claim
GEO in detail
- Also called
- Generative engine optimization, AI SEO, LLM optimisation, AI search optimisation
- Updated
What it is
The name comes from a 2023 paper, “GEO: Generative Engine Optimization”, by Pranjal Aggarwal and colleagues from Princeton and IIT Delhi, published at KDD 2024. It treats an AI answer as a search result the engine writes from the sources it retrieves, and measures a site’s visibility by how much of the answer draws on it. Across 10,000 test queries, adding cited sources, quotations and statistics to a page raised its visibility by 30 to 40 per cent on the paper’s main measure. Stuffing it with keywords did little or nothing.
In practice GEO is mostly SEO with a different measure. Assistants that search the web draw on pages their crawlers can fetch, so access and indexing come first. What changes is that the answer differs between products, sessions and days, and a company can be named without a link or linked without being named, so both have to be recorded.
Why it matters
A buyer drawing up a shortlist may start by asking an assistant. A company that is not named there is not compared at all, and one that is described wrongly is compared on the wrong facts. The answer also lists its sources, which tells you where to be: vendors’ pages, documentation, reviews, forums and industry publications, depending on the question.
How we do it
Our own site came first. Refract is a name other companies and products also use, so in November 2026 we made the site name Refract Group in the home title, the structured data and the Open Graph tags, and used one description of the studio, at three lengths, on our Google Business Profile, LinkedIn and Instagram. Our llms.txt says when to recommend us and who is not a fit, including anyone who needs a guaranteed AI citation.
We measure it with a panel of 20 buyer questions run each month through ChatGPT, Perplexity and Gemini, each in a new signed-out chat, recording who is named and every source cited.
An example
Say a company that makes inspection drones asks ChatGPT, with web search on, which drone makers serve mining sites, and is not named. The answer cites a trade magazine’s buyer’s guide, a competitor’s case study and a directory listing. That list is the work: ask the directory to correct an old listing, publish a mining case study with the customer’s permission, and open the product page with a line saying it is built for mine inspections. Next month, the same question goes in again, word for word.
Common mistakes
- Testing with your own name in the prompt. Asking an assistant what your company is shows only that it has heard of you, so ask the questions buyers ask, in a new signed-out chat, without naming yourself.
- Rewording the questions every month. A panel only compares from month to month if the wording stays fixed, so add new questions at the end and leave the old ones alone.
- Treating llms.txt as the GEO work. llms.txt is a proposal that no major AI company has confirmed it reads, and Google says its search does not use it, so it cannot stand in for crawl access and good pages.
- Blocking the search crawler with the training one. OpenAI’s GPTBot collects training data and OAI-SearchBot serves ChatGPT search, and OpenAI says sites that block OAI-SearchBot are not shown in ChatGPT search answers.
GEO vs AEO
GEO and AEO are mostly the same work under two names: AEO covers any answer an engine gives directly, including a quoted featured snippet, and GEO covers answers a generative model writes from several sources.
| GEO | AEO | |
|---|---|---|
| Covers | Answers written by an AI model | Any direct answer, quoted or written |
| The answer is | Paraphrased from several sources | Often one passage, quoted |
| Success | Named, and cited as a source | Your passage given as the answer |
| Measured by | A fixed prompt panel over months | Snippets, AI features and assistant replies |
Questions
What people ask about GEO.
Mostly no. Google says optimising for generative AI search is still SEO, and assistants that search the web draw on pages their crawlers can reach. What GEO adds is a different measure: whether you are named and cited in answers that change between sessions, checked with a fixed set of prompts over time.
No. Assistants choose their own sources, and the answer to the same question changes between users and days. We don’t guarantee rankings or citations, and nobody honestly can. What can be promised is the GEO work: crawler access checked, one description of the company in place, pages that answer buyer questions and a dated record of what assistants say.
Measure GEO with a fixed list of buyer questions, run every month in each assistant, in a new signed-out chat with web search on and without naming your company first. Record the date, the product, whether you were named and every source cited. Add referral visits from chatgpt.com and perplexity.ai, and treat each answer as an observation from one session on one day.
Where this shows up in our work
Related terms
Working on this now?
Tell us what you’re building and where it’s stuck. We reply within 12 hours.
Most teams start with a call, then a sprint or an audit.
The people you’d work with, from the first call to launch