Generative engine optimisation, explained without the hype

Generative engine optimisation is a clumsy name for a simple problem. When somebody asks ChatGPT for the best accountant in Leeds, or which competitor monitoring tool to buy, an answer comes back naming two or three companies. GEO is the work of being one of them.

It matters because the question now ends there. There is no page of ten blue links to be fifth on. You are named or you are not.

How the answer is actually built

It helps to know what is happening under the hood, because it explains why the usual SEO advice only half works.

When you ask a modern assistant a buying question, roughly this happens:

1. It decides whether to search. Some answers come from what the model already absorbed in training. Others trigger a live web search. Buying questions with local or current intent almost always trigger a search.
2. It runs several searches, not one, and usually not your exact phrasing. It rewrites your question into a handful of queries.
3. It reads a small number of results. Often five to ten pages, heavily weighted towards ones that look like lists, comparisons and reviews.
4. It writes a summary naming a few options, and cites some of what it read.

Two consequences fall out of that, and they are the whole game.

Your own site is usually not the source. The pages that get read are third party: directories, comparison articles, review sites, forum threads. You can have the best page on your own website and still be absent, because the assistant never read it.

Being cited and being named are different things. An assistant can quote your blog post as a source and then recommend three other companies in the answer. Citation is nice. Being named is the outcome.

What actually moves the needle

Ranked roughly by what we see move fastest.

1. Be in the lists that get read

Every category has a handful of pages assistants lean on. Directories, "best X in Y" articles, review platforms, a strong subreddit thread. Find them by asking the assistant your buying question and reading which sources it cites.

Then get listed. This is unglamorous work: submissions, outreach to the writer of the listicle, claiming your profile on the review site. It is also the single highest leverage thing in GEO, because one listing can put you in dozens of answers.

2. Make the facts easy to extract

Assistants summarise. They cannot summarise what is not written plainly.

Put the answerable facts in text on your own pages: what you do, who for, where, what it costs, how long it takes, what makes you different in one sentence. Not in an image, not in a video, not implied by a testimonial. Written down.

Most homepages fail this badly. Read yours and ask whether a stranger could answer five factual questions about your business from it alone.

3. Get third parties to say the same thing

Consistency across sources is what turns a maybe into a name. If your own site, your Google listing, your LinkedIn page and three directories all describe you the same way, the model has a confident, repeated signal. If they contradict each other, you get dropped in favour of a competitor who is boringly consistent.

4. Reviews, properly

Review count and rating come up constantly in local and service answers. Assistants read them because they are a cheap proxy for quality. This is one of the few areas where classic local SEO work maps straight onto GEO.

5. Structured data, modestly

Schema markup helps machines parse you and is worth doing. It is not the lever people claim. A site with perfect schema and no third party coverage stays invisible.

What does not work

• Keyword stuffing for robots. Models read like readers. Text written for machines reads badly to both.
• Buying "AI optimisation" packages that promise placement. Nobody controls an organic answer.
• One off checks. Answers vary between runs, models and phrasings. A single lucky result means nothing.
• Obsessing over your own blog. Useful for citation, rarely the reason you get named.

How to measure it

This is where most GEO effort falls apart. Without measurement it is astrology.

The method that works is boring and repeatable:

1. Write ten to twenty questions a buyer would actually ask. Not your brand name. Things like best competitor monitoring tool for agencies or who should I use for X in Y.
2. Ask all of them across ChatGPT, Claude, Gemini and Perplexity, with web search on.
3. Record, for each answer, which businesses were named and which sources were cited.
4. Repeat weekly, same questions, and watch the share move.

The number that matters is simple: of the questions asked, what share named you. Track it for your competitors at the same time or you have no idea whether a change is you or the category.

Doing this by hand across four assistants and twenty questions is about two hours a week. Figo does it automatically and keeps the history, which is why we built that part of it, but the method above works with nothing but a browser and a spreadsheet.

A realistic first month

• Week 1. Write your questions. Run the baseline by hand. Note every source cited.
• Week 2. Fix your own pages so the basic facts are extractable text. Claim and complete every profile you already have.
• Week 3. Get into three sources the assistants actually cited in your baseline.
• Week 4. Re run the same questions. Compare. Expect small movement, not a transformation.

If your share has not moved after eight weeks of that, the problem is usually that you are absent from the sources being read, not that your own site is wrong.

The honest summary

GEO is mostly public relations and distribution wearing a technical costume. Be in the places that get read, say the same true thing everywhere, make your facts easy to lift, and measure it weekly so you know whether any of it worked.