Generative engine optimisation: a working guide for UK businesses
Generative engine optimisation (GEO) is the work of earning citations in the answers that AI assistants give. When someone asks ChatGPT, Claude, Perplexity or Google's AI results who does what you do, the assistant assembles an answer from sources it can read, parse and trust. GEO is making your site one of those sources.
This is a working guide, not a theory piece: we do this for our own two sites and for clients, and the honest headline is that GEO rewards almost embarrassingly unfashionable things. Clear answers to real questions. Machine-readable structure. Claims a sceptical reader could verify. And entity clarity, meaning a machine can work out who you are, where you operate and what you actually sell without guesswork.
If you want the beginner's version first, read our plain English explainer on GEO and come back.
What do assistants actually reward?
Four things keep showing up when a page gets cited.
First, being the clearest answer to a specific question. Assistants assemble responses from passages, not from domains. A page that opens by directly answering "does a late notice bar a claim under NEC?" can be lifted into an answer; a page of atmospheric brand copy cannot.
Second, structure a machine can parse. Question-form headings, short definitional paragraphs, tables for comparisons, and structured data using schema.org vocabulary all reduce the work a model has to do to understand the page. The Google Search documentation is explicit that structured data helps systems understand content, and the same markup serves the assistant crawlers.
Third, verifiable claims. A statement with a source attached is safer for an assistant to repeat than a bare assertion, because these systems are increasingly tuned to avoid repeating unsupported numbers. Pages full of round marketing statistics are not just unpersuasive to humans; they are radioactive to a system trying not to hallucinate.
Fourth, entity clarity. Consistent business name, consistent services, consistent location details across your site, your Companies House record and your directory listings. Ambiguity about who you are is a reason to cite somebody else.
What does the practical checklist look like?
| Action | What good looks like | Effort |
|---|---|---|
| Answer real questions | One page per genuine customer question, answer in the first paragraph | Ongoing |
| Question-form headings | H2s a person would actually type or ask | Low |
| Structured data | Organisation, Service, FAQPage and Article markup that matches the visible page | Low to medium |
| Tables for comparisons | Any either-or decision presented as a table, not prose | Low |
| Cite your sources | Named, linked, authoritative; legislation and regulators over blogs | Low |
| Entity consistency | Same name, services and location everywhere a machine might look | Low |
| Crawlability | Assistants' crawlers not blocked; clean HTML, content visible without JavaScript | Medium |
| Honest scope | Say what you do not do; assistants reward precision over breadth | Low |
None of this is exotic. The barrier is not knowledge, it is the discipline to write for a reader with a question rather than for your own brochure.
How is GEO different from the SEO you already know?
The overlap is large, which is good news: nothing above hurts your conventional rankings. The differences are emphasis. Classic SEO competes for position on a results page a human then scans. GEO competes for inclusion in a synthesised answer, which makes passage-level clarity and citation-worthiness matter more, and makes the click matter less. That shift is the same one we described in the zero-click automation argument: value increasingly flows to whoever the machine consults, not whoever the human visits.
There is also a structural point. As assistants connect to tools and data rather than just web pages (the direction MCP and connected office tooling points in), being machine-legible stops being a marketing tactic and starts being basic plumbing for how your firm gets found and used.
Can you actually measure it yet?
Honestly: partially, and anyone claiming precision is selling something. The field is young. Assistant platforms do not publish citation analytics the way search consoles publish rankings, and the same question asked twice can produce different sources.
What you can do today is modest but real. Ask the major assistants the questions your buyers ask, monthly, and record who gets cited. Watch referral traffic from assistant domains in your analytics. Watch for the tell-tale enquiry that arrives already informed ("the AI said you deal with X"). Treat all of it as directional evidence, not a dashboard. We do not publish traffic claims for exactly this reason: the honest measurement story is early, and inventing numbers would fail our own verifiability test above.
Where should a UK firm start this week?
Pick the ten questions your best customers actually ask before they buy, check whether your site answers any of them in the first paragraph of any page, and be honest about the result. For most UK SMEs the answer is somewhere between one and zero, which is also the size of the opportunity, because your competitors are in the same position and mostly not moving.
Then fix entity basics (one afternoon), add Organisation and Service markup (one more), and start answering one real question per week in plain English with sources. Twelve weeks of that beats any amount of tooling.
GEO is not a trick to be applied after the fact. It is a decision to become the clearest, best-evidenced answer in your niche, and then to let the machines notice. The firms that treat it that way will be quoted; the firms that treat it as a hack will be filtered.