What is a GEO agency?
A GEO agency works on being cited inside generated answers rather than ranked in a list of links. Most of what it does is search engine optimisation done well: the genuinely new part is narrow, and it is worth knowing which is which before you pay for it. This guide explains the mechanism, what can and cannot be measured, and four questions that separate a real practice from a repositioned one.
Generative Engine Optimization (GEO) is the work of being the source a generated answer is built from, rather than a link in a list of results. ChatGPT, Perplexity, Google’s AI Mode and Gemini answer a question directly and cite a handful of sources underneath. Being one of those sources is a different objective from ranking first, and it is not reached in quite the same way.
The term is roughly two years old and the market around it is noisy. This guide is written for someone deciding whether to hire for it. It says what is genuinely new, what is search optimisation under a new name, and what you will not be able to measure.
The mechanism, in one paragraph
A generated answer is not retrieved from a ranking. The engine breaks your question into several sub-questions, searches for each of them, reads a set of pages, and composes one answer from what it found. The pages it reads are drawn from ordinary search results, which is why classic optimisation still matters. What decides which of them get cited is different: the engine favours passages that answer a sub-question directly, that state something specific, and that agree with what other sources say. Convergence across several independent sources matters more here than the authority of any single one.
What is genuinely new, and what is not
Most of it is not new. A page that is slow, blocked from crawling, rendered only by JavaScript, or thin, will not be read: by a search engine or by a generative one. If an agency’s GEO offer starts with technical health, semantic structure and credible content, that is because those are the prerequisites, not because it is padding the invoice.
Three things are genuinely different.
The unit of work is the passage, not the page. An answer is assembled from fragments. A page that buries its answer in the fourth paragraph, after a warm-up section, is a page whose answer is harder to lift. Writing the claim first and the context second is a real change in editorial habit, and it is the single change with the most effect.
Being listed elsewhere counts more than it used to. Comparison articles and rankings (the format the industry spent a decade calling low-value) are what generative engines lean on to settle a “who is best at X” question. Appearing in several of them, on sites that are not yours, produces the convergence that triggers a citation. This is uncomfortable to write on an agency site, because it means part of the work happens where you have no control.
Consistency across sources becomes a technical concern. If your company describes itself one way on its own site, another way in a directory, and a third way in an old press release, the engine has three versions and no reason to prefer yours. Reconciling them is unglamorous and it works.
What you can measure, and what you cannot
Start with the uncomfortable part: the direct traffic from being cited is small. In the answer format, most readers get what they came for and do not click. Expect a few percent of your acquisition, not a channel. What a citation actually produces is a brand effect: your name is read, by someone in a buying frame of mind, alongside a recommendation. That is worth having, and it is not a traffic line.
What can be measured, concretely:
Whether you are cited at all. Ask each engine the twenty or thirty questions a prospect would actually ask, record which sources come back, and repeat monthly. It is manual, it is unglamorous, and it is the only measurement that answers the question directly. Any agency selling you GEO should be doing this and showing you the log.
The traffic that does arrive. Analytics tools classify visits from ChatGPT, Perplexity and Claude as referral traffic by default, which scatters them across your reporting and makes them look like nothing. Reclassifying them as organic is a reporting convention, but it is the right one: they are the continuation of search. Set it up before you start, or you will have no baseline.
What cannot be measured: your share of answers across a market, how often you were nearly cited, and whether a competitor paid for anything. Any dashboard presenting those as figures is presenting an estimate as a measurement.
Four questions that separate a practice from a repositioning
Most agencies added GEO to their offer in the space of a year. Some built something; most renamed a page. These four questions tell them apart, and none of them require you to be technical.
“Show me your citation log.” If they measure, they have one: a list of questions, engines, dates and results. If the answer is a slide about how AI is changing search, they are not measuring.
“Which of my competitors are cited today, and on what?” A real answer takes an afternoon of work and is specific. A vague one means the market has not been looked at.
“What part of this is just SEO?” An honest agency will say most of it. One that claims a wholly new discipline is either mistaken or selling.
“What will you do that happens off my site?” If nothing, the offer is incomplete: convergence across sources is where citations come from, and none of it lives on your domain.
Do you need one?
Probably not as a separate line item. If your search visibility is weak, the generative layer will not rescue it: the engines read the same pages. Fix the foundation first, and treat citation as an objective added to that work rather than a service bought beside it.
Where it is worth deliberate effort: markets where buyers research before contacting anyone, where a shortlist forms before the first conversation, and where an assistant is now part of how that shortlist forms. Professional services, software, and considered purchases: the same places where a comparison article has always mattered.
If that describes your market and you would rather have it handled than run it yourself, that is the work we do as a GEO agency: stated here once, at the end, because a guide that sells in its third paragraph is an advertisement.
Frequently asked questions
Is GEO different from SEO?
Partly, and less than the market suggests. Generative engines read pages that ordinary search surfaces, so technical health, structure and credible content remain the prerequisites. What differs is the objective (being cited inside an answer rather than ranked in a list) and three consequences of it: write the answer first rather than the context, appear in comparison articles on sites you do not own, and keep how you describe yourself consistent across sources.
How much traffic does a citation bring?
Less than you would hope. In an answer format most readers get what they came for and do not click, so direct traffic from citations tends to be a few percent of acquisition rather than a channel of its own. The return is a brand effect: your name is read by someone in a buying frame of mind, next to a recommendation. Treat it that way in your expectations and in your reporting, and you will not be disappointed by a number that was never going to be large.
How do I know whether I am cited?
Ask the engines. Write down the twenty or thirty questions a prospect would actually ask, put each one to ChatGPT, Perplexity, Gemini and Google's AI Mode, and record which sources come back. Repeat monthly, keep the log. It is manual and there is no reliable shortcut at the time of writing. Any agency selling generative visibility should be doing this and showing you the record.
Why do rankings and comparison articles matter so much?
Because a generated answer to "who is best at X" is assembled from the sources that already answer that question, and those are comparison articles. Appearing in several of them, on sites that are not yours, produces the agreement across sources that engines lean on. It is the least glamorous part of the work and the one with the clearest effect: which is also why an agency that only proposes work on your own site is proposing half of it.
Should I create a separate GEO budget?
Usually not. If your search visibility is weak, the generative layer will not compensate: the engines read the same pages. The sound approach is to add citation as an objective of the search work you already do, and to judge both on the same number. A separate line item mostly benefits the party invoicing it.
Will my analytics show this traffic correctly?
Not by default. Analytics tools classify visits from ChatGPT, Perplexity and Claude as referral traffic, which scatters them and makes a small channel look like noise. Group them with organic search (they are its continuation) and set that up before you begin, otherwise you will have nothing to compare against later.