Generative engine optimisation agency: the GEO work that gets you cited

Generative engine optimisation, or GEO, is the work of making a brand's facts readable, retrievable and corroborated enough that a generative engine repeats them. These engines write an answer out of several sources rather than handing over a list of links, so the target is a citation rather than a position. In practice it comes down to three jobs: publishing your facts in one consistent form, getting independent sources to repeat them, and making sure the crawlers that build the answers can fetch your pages.

They overlap, and no standards body defines either term, so plenty of practitioners use them for the same work. The split we work to is selection against synthesis. Answer engine optimisation is about being the answer: structuring a page so the question is the heading and the answer sits directly under it, which is mostly work on your own site. Generative engine optimisation is about being the citation inside an answer the model writes itself, which is mostly work on properties you do not own. Most brands need both, and the technical foundations underneath them are shared.

ChatGPT, Perplexity, Gemini, Copilot and Google's AI Overviews, and they do not share a pipeline. Copilot answers are grounded in Bing's index, AI Overviews sit inside Google Search, and ChatGPT, Claude and Perplexity each run their own retrieval crawler. That is why one tactic rarely moves all of them at once.

If you want to be cited, the retrieval agents at least. GPTBot, ClaudeBot and Google-Extended mostly feed model training, while OAI-SearchBot, Claude-SearchBot, PerplexityBot and Bingbot fetch the pages an answer is built from. Block that second group and your pages cannot be retrieved, so they cannot be cited. Blocking the training agents is a legitimate business decision and we will walk you through what each one costs before anything is changed.

Not for rankings. No major model provider has committed to reading llms.txt in production, and Google has said the file has no effect on Search. We publish one on this site anyway, because it costs nothing to generate and agent tooling does read it. Treat it as housekeeping rather than a ranking factor, and be wary of anyone selling it as one.

With a reading of what the assistants say about you today and where each claim came from. Then one agreed wording per fact, your own pages and structured data first, and outwards through listings, profiles and stockists. The conflicts are usually old rather than malicious: a discontinued spec on a marketplace, a retired warranty on a stockist page.

The plumbing moves in weeks: access, rendering, structured data and the fact sheet. Corroboration takes longer, because it depends on other people publishing and on the engines refreshing their indexes to their own schedule. You get the same monthly readout from month one, so the movement is visible while it happens.

No, and nobody honestly can. Nobody controls what a model writes, none of the engines publish a visibility metric we could hold up as proof, and the surfaces change month to month. What we can do is remove the reasons you are not being cited, then show you the same measurement every month whether it flatters us or not.