GEO, without the hype
Generative engine optimisation
Generative engine optimisation is the work of getting your company named inside an AI's answer, rather than listed in the links beneath it. It matters because the answer is increasingly where the decision happens — and being on page one of a search result the buyer never scrolls to is not the same as being seen.
GEO in one paragraph
Definition. Generative engine optimisation (GEO) is the practice of making a brand, product or source more likely to be selected and cited by generative AI systems when they answer a question. It overlaps with SEO — both care about being findable — but the unit of success is different: SEO wins a position in a list, GEO wins a sentence inside an answer.
What actually changes from SEO
| Classic SEO | GEO | |
|---|---|---|
| What you win | A position in a list of links | A mention inside the answer itself |
| Who decides | A ranking algorithm over an index | A model choosing what to say, sometimes with retrieval |
| Unit of content | A page targeting a query | A passage that answers a question and can be quoted intact |
| Where the credit goes | Your domain | Whoever the model cites — which may be a directory, not you |
| How you measure | Position, clicks, impressions | Mention rate across runs, share of voice, which sources get cited |
| Feedback speed | Weeks | Immediate to measure, slow to move: models update on their own schedule |
The two are not rivals. Retrieval-based answers still lean on the open web, so pages that rank well remain the raw material an AI quotes.
What can honestly be measured today
- Mention rate — how often the brand appears in answers to your buyers' questions, with a confidence interval. Model answers vary between runs, so a figure without an interval is a guess with a decimal point.
- Share of voice — your mentions against the competitors you track, counted only on questions that do not contain your brand name.
- Role and sentiment — recommended, listed, or mentioned in passing; and how the model talks about you.
- Cited sources — the sites the model links to in your field. This is the most actionable output: it names the places you need to exist on.
What cannot honestly be measured: a causal link between a GEO change and revenue, or a guaranteed position in an answer. Anyone selling you that is selling a forecast, not a measurement.
Where to start
- Measure firstWithout a baseline you cannot tell a real change from the noise between two runs of the same question.
- Answer real questionsPages built around the question a buyer types, answered in the first paragraph, in language a model can quote whole.
- Get into the cited sourcesIf the model keeps citing three directories and a comparison article, those are your targets — not another blog post on your own domain.
- Fix what the AI gets wrongOutdated positioning is the most common reason a model recommends someone else.
- Re-measureSame questions, same models. The interval tells you whether anything moved.
Start with the measurement
Free, no sign-up. 16–18 questions, commercial ones repeated 3 times, every raw answer shown.
Run a free checkCommon questions
- Is GEO a real discipline or a rebranding of SEO?
- The techniques overlap heavily; the measurement does not. You cannot read AI mentions out of a rank tracker, and that gap is what makes it a separate practice rather than a new label.
- Can you guarantee my brand will be named?
- No, and neither can anyone else. Models change, retrieval changes, and the answer to the same question differs between runs. What is achievable is a measurable improvement in how often you are named.
- Does this only apply to Google's AI Overviews?
- No. AI Overviews are one surface; assistants people open directly are another, and they answer from different data. We measure across several.
- Where is the formula?
- On the methodology page — the same one the code runs.