What is GEO?

Why it's a separate job from SEO

Ordinary search returns ten links and you scroll. Being tenth is worth something. Being twentieth is worth a little. There's a long tail of consolation traffic.

A generated answer doesn't work that way. Ask an assistant for a plumber in your town and it names two or three businesses, then stops. There is no page two, no scrolling, and no consolation prize for being fourth. You're either in the answer or you're invisible.

That changes the economics. In SEO, incremental improvement gives incremental return. In GEO it's closer to binary, which makes it both riskier and, right now while almost nobody is competing, unusually cheap to win.

How assistants decide who to name

The exact mechanisms differ between tools and change often, but the pattern is consistent enough to work with. Broadly, an assistant answering a "recommend me a supplier" question is drawing on some mix of three things: what it absorbed during training, what it retrieves from a search index when it browses, and what it can verify across multiple independent sources.

That third one is the part most businesses miss. An assistant is far more willing to name a business it can corroborate from several places than one that only asserts things about itself on its own website. Your site claiming you're a Poole electrician is one data point. Companies House, your Google Business Profile, a directory listing and a trade body page all agreeing is something quite different.

What actually works

Be readable by machines

Structured data (schema markup) that says plainly what your business is, where it operates, what it sells and who runs it. Clean HTML. Content that isn't trapped behind JavaScript, tabs or accordions. A lot of local sites hide their best answers inside collapsed FAQ widgets that crawlers handle poorly.

Be consistent everywhere

Your name, address, phone number and description should match across your site, your Google Business Profile, Companies House, directories and social profiles. Contradictions actively lower confidence. If three sources say slightly different things, an assistant hedges, and hedging means naming someone else.

Be corroborated by others

This is the highest-value and slowest work. Directory listings, local press, supplier and partner pages, trade associations, reviews. What other sites say about you matters more than what your own site says. It's also the part no amount of on-page tinkering can substitute for.

Write things that can be quoted

Assistants extract sentences. Content built to be lifted gets used: direct answers in the first line or two, specific numbers, dates, plain definitions, comparison tables. Long meandering intros that circle the point for three paragraphs before answering do not.

What doesn't work, or isn't proven

• llms.txt. A proposed file for telling AI crawlers about your site. There's no public evidence any major AI company uses it as a ranking or retrieval input. It's cheap and harmless to add, but if someone sells it to you as the answer, be sceptical.
• Keyword stuffing. Works even less well here than in SEO, because generated answers are drawn from meaning rather than term frequency.
• Volume of AI-written filler. Publishing forty thin articles gives an assistant more text but no more reason to trust you. Corroboration beats volume.
• Paying for placement. There is currently no advertising slot in these answers to buy. Anyone offering to sell you one is not selling what they claim.

How you measure it

Less precisely than SEO, and it's worth being upfront that anyone promising you a neat dashboard is overselling. Generated answers vary between users, sessions and locations, so there's no reliable equivalent of a rank tracker.

What works in practice is sampling. Write down the ten or fifteen questions a customer might actually ask, such as "who's a good electrician in Bournemouth" or "best bookkeeper for a small business in Poole", then ask them across the main assistants on a fixed schedule. Record two things each time:

1. Are you named at all?
2. Is what it says about you correct?

The second question catches problems the first misses. Being named but described with the wrong services, the wrong area or an old phone number is its own kind of failure, and it's usually traceable to inconsistent information somewhere you'd forgotten about.

Is it worth doing yet?

It depends entirely on whether your customers are using these tools to find suppliers like you. For some trades that's already meaningful; for others it's still marginal.

The honest case for doing it now isn't that the volume is huge. It's that the competition is almost nonexistent. The groundwork is mostly the same groundwork that improves your ordinary search performance anyway, so you're rarely spending money that only pays off in one place. That combination won't last, and it's the cheapest this will ever be.

If you want to know how AI tools currently describe your business, send us your website and we'll check and tell you what we find, free.