Last checked: 10 June 2026. Disclosure: we are Fortitude Media. We compete for some of the same budgets as the agencies this guide will help you interview, so read it knowing that. We have written the questions to be useful even if you never speak to us, and there is exactly one paragraph about us, clearly marked, at the end. Facts and prices come from public sources checked on the date above. Corrections welcome by email.
The UK GEO services market is reported to be growing at more than 30% a year, which means plenty of good firms and plenty of rebadged SEO decks. These questions separate them. If you are still deciding whether an agency is the right shape of purchase at all, read tool, agency or both first, and our UK B2B roundup covers the alternatives by name.
1. Which engines do you measure, and how often?
A credible agency names its engines and its cadence without checking. The major engines are ChatGPT, Claude, Gemini, Perplexity and Google AI Overviews, and they answer differently, so coverage matters: a firm tracking two engines is reporting a fraction of the picture. Why it matters: AI answers vary between engines and over time, so a single engine measured occasionally tells you almost nothing. A good answer names at least four or five engines, states a regular cadence, and admits that answers fluctuate run to run. A weak answer is "we check ChatGPT" or any hesitation about how often.
2. Can I see my number in your own platform, or in a third-party tool?
This question exposes the biggest structural split in the market. Most GEO agencies have no proprietary platform: reporting is assembled from third-party tools they rent. That is not dishonest, but it means the measurement can change or vanish when their tool subscription does, and the agency cannot vouch for the methodology behind it. A good answer is either "our own platform, here is a login" or a frank "we use [tool], here is its methodology, and you could subscribe to it yourself to verify us". A weak answer is a monthly PDF with no stated source.
3. Do you have history on my category from before our engagement?
Almost no agency does, and the answer tells you how your first three months will be spent. Without prior history, the engagement starts blind: the baseline is built during the time you are paying for, and you can never see whether your category was already rising or falling before you signed. A good answer is honest about this ("no, we will establish a baseline in month one") or, rarely, shows real longitudinal data on your category. A weak answer dodges the question or presents general market stats as if they were your category's history.
4. How do you separate your work from market noise?
Attribution is the hardest problem in this field, and the best agencies say so. AI answers move because of model updates, competitor activity and seasonality, not just because of anyone's work. A good answer describes a comparison method: tracking named competitors alongside you, so your movement is read against the category rather than in isolation, and being upfront that samples are samples. A weak answer claims every uptick as theirs and never mentions confounds. Our piece on benchmarking AI visibility against competitors explains why the competitor baseline is the part you should insist on.
5. What does "done" look like each month?
You are buying a retainer, so ask for the shape of a month before you sign, not after. Vague retainers drift. A good answer lists concrete outputs (content pieces, outreach placements, technical fixes) and connects each one to the metric it is supposed to move. A weak answer is "ongoing optimisation" or hours without deliverables. Ask to see a real (anonymised) monthly report from an existing client; the report they show you is the best report they produce.
6. Who writes the content, and who is accountable for the number moving?
These are two different jobs and you want one name attached to each. Content quality is the core of this work: AI engines cite pages that answer buyer questions with substance, as we set out in the anatomy of an article AI engines cite. A good answer tells you who writes (in-house, freelance, AI-assisted and edited, or AI-generated), who has subject-matter review, and which person owns the visibility number itself. A weak answer separates the writers from the metric entirely, so that when the number stalls, content blames strategy and strategy blames content.
7. What happens to my data and reporting if I leave?
Exit terms reveal how confident an agency is about retention. If the measurement lives in their third-party tool subscriptions, your history may evaporate the day the retainer ends. A good answer states plainly what you keep: exported reports, the content (you should own all of it), and ideally continued access to the measurement at some price. A weak answer has never been asked the question before. Get whatever they say into the contract.
8. How is your pricing structured, and what is in it?
Typical credible mid-market GEO retainers in the UK run £2,000 to £10,000 a month, and the structure matters as much as the figure. Ask what is inside: how much content, how much outreach, how much measurement, and what costs extra. Hourly metering can quietly inflate; flat retainers can quietly dilute. A good answer maps money to outputs without prompting. A weak answer is a single number with "it depends" behind it. If the quote is far below £2,000 a month, ask what is being left out; if far above £10,000, ask what evidence justifies it.
9. Can you show a measured before-and-after for a real client?
Case studies in this market are young everywhere, so the bar is evidence, not volume. A good answer shows one real client with a dated baseline, the work done, and the movement afterwards, ideally in whatever tool measured it, and is candid about what did not work. A weak answer is traffic screenshots in place of AI visibility data, or claims that cannot be tied to dates. One verifiable example beats ten adjectives.
10. Do you publish your methodology?
A published methodology is the cheapest trust signal an agency can offer and most cannot. If a firm sells movement in a number, you are entitled to know how the number is made: which prompts, which engines, what counts as a mention, how sentiment is judged. A good answer is a public page you can read before the meeting. A weak answer is "it's proprietary", which in measurement usually means "undocumented". You can read ours as an example of the standard to hold everyone to, including us.
The ten questions at a glance
| # | Question | A good answer sounds like |
|---|---|---|
| 1 | Which engines, how often? | Four or five named engines, regular cadence, candour about variance |
| 2 | Your platform or third-party? | A login you can check, or a named tool plus its methodology |
| 3 | History on my category? | Real prior data, or an honest "no, baseline starts month one" |
| 4 | Your work vs market noise? | Named-competitor baseline, no claiming every uptick |
| 5 | What is "done" monthly? | Concrete outputs tied to metrics, sample report offered |
| 6 | Who writes, who owns the number? | Named writers, named owner of the metric |
| 7 | What do I keep if I leave? | Content is yours, exports provided, terms in the contract |
| 8 | Pricing structure? | Money mapped to outputs, within £2,000 to £10,000/mo norms |
| 9 | Measured before-and-after? | One dated, verifiable client example with caveats |
| 10 | Published methodology? | A public page, not "proprietary" |
How Fortitude answers these, for the record
Since we would face the same ten questions, here are our answers in one paragraph. We measure five engines (ChatGPT, Claude, Gemini, Perplexity, Google AI Overviews) in our own platform, Fortitude Sentinel, which you can log into at any tier including the free one. Category history comes from the Registry, our longitudinal record of how AI engines answer canonical buyer-intent prompts per category, and new customers receive up to twelve months of their category's history at signup. Our methodology is published. The work is done by Fortitude Forge through a monthly pool of delivery hours, and the people doing it are accountable to the same chart you see, which is the core of our case. Plenty of agencies will serve you well; we simply think every one of them should be able to answer this page.
FAQ
How much does a GEO agency cost in the UK? Credible mid-market GEO retainers in the UK typically run £2,000 to £10,000 a month. Quotes far below that range usually exclude content production or measurement, so ask what is inside the number.
What should I ask a GEO agency before hiring them? Ask which engines they measure and how often, whether you can see your number in their own platform, whether they hold history on your category, how they separate their work from market noise, what a month of delivery looks like, who writes the content, what you keep if you leave, how pricing is structured, whether they can show a measured before-and-after, and whether they publish their methodology.
Do GEO agencies have their own measurement data? Most do not. The typical UK GEO agency has no proprietary platform and assembles reporting from third-party tools, which is workable but means you should ask which tools, what methodology sits behind them, and what happens to your history if you leave.
How do I know if a GEO agency is actually working? Insist on a dated baseline before work starts and measurement against named competitors, so your movement is read against the category rather than in isolation. An agency that claims every uptick as its own, without controlling for market noise, is not measuring seriously.
Can a GEO agency promise a timeline for results? No serious provider should promise a fixed timeline, because AI answers vary between engines and over time and every measurement is a sample, not a census. A good agency commits to concrete monthly outputs and transparent measurement rather than a guaranteed date.
Before you interview anyone, know your baseline. Our free AI Visibility Check shows how the five major AI engines answer your category's buying questions today. Take it into the meeting; it makes question three much more interesting.