How to choose an AI visibility agency: 12 questions to ask before you sign

"AI visibility", "GEO" (generative engine optimisation) and "AEO" (answer engine optimisation) all describe the same goal: getting your business named when someone asks ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews or Google AI Mode who to call. The labels are new, which makes it hard to tell a careful agency from one reselling a dashboard.

Google addresses this directly. Its guide to hiring an SEO, last updated in June 2026, tells site owners to check whether an agency's "AEO" or "GEO" advice lines up with Google's official guidance on generative AI features. That is a good start, but Google's guidance covers its own AI Overviews and AI Mode, two of the seven engines. This page is not a ranking of agencies. It is a checklist you can take into any pitch, ours included.

Why the engines matter separately

A single blended "AI score" hides most of what matters. In our citation-source measurement, we ran 22 buying prompts in one software category across six engines in 23 daily scans between 20 August and 11 September 2026, and counted how many of 25 third-party websites each engine cited at least once.

That is a spread from 23 to 3. In that data, getting onto the roundups and community threads is the direct route into Perplexity or Claude, and barely reaches ChatGPT. An agency that reports one number cannot tell you which of those happened, which is why most of the questions below are about measurement.

Questions 1 to 4: measurement

1. Which engines do you measure, one by one?

There are seven worth asking about: ChatGPT, Perplexity, Gemini, Claude, Grok, Google AI Overviews and Google AI Mode. Coverage is often a plan setting rather than a given. On 24 September 2026, for example, Peec AI's pricing page offered a choice of three models on each plan below Enterprise, with more sold as add-ons. Three can be a sensible budget choice, but you should know which three and why.

Good answer: they name each engine they track and each they don't, and explain the gap.
Red flag: "We track all the AI platforms," with no list.

2. Do you test unbranded prompts, or only our name?

A prompt that contains your business name mostly tests whether an engine can look you up. It flatters everyone.

The same report found a UAE car-servicing directory scoring 62/100 on traditional search and 3/100 on AI answer visibility on the same day, cited on none of its unbranded prompts.

Good answer: reports are built on questions a buyer asks without knowing you, such as "best AC repair in Al Barsha" or "movers in Sharjah for a one-bedroom flat".
Red flag: a sample report full of prompts containing your brand name, with every engine green.

3. How do you choose the prompts, and in which languages?

Prompts should come from how your customers actually ask: sales calls, WhatsApp enquiries, the queries in your Search Console. In the UAE that can mean English and Arabic versions of the same question.

Good answer: they show you the list before measuring, say where each prompt came from, and include Arabic where your customers use it.
Red flag: hundreds of generic prompts that could fit any business, or a prompt list you are not allowed to see.

4. Can I see a sample report with dated, per-engine results?

Answers shift from day to day, so an undated screenshot proves very little.

Good answer: an anonymised real report showing each prompt, each engine, the date it ran, whether the business was named or cited, and who was named instead.
Red flag: a single "visibility score" with no answers behind it.

Questions 5 to 7: the work and who owns it

5. What exactly is the work?

Measurement is the instrument, not the job. The work usually sits in four places: making your business details consistent wherever they appear (entity work), earning mentions on the third-party pages engines cite, writing pages that answer buyers' questions directly, and technical fixes. In our citation study, engines cited third-party hosts 2,061 times and the vendor's own site 106 times, 4.9% of the evidence. An agency that only edits your website is working on the smallest slice.

Keep structured data in proportion. Google's guide to its generative AI features says schema is not required for them and there is no special markup; it also says Google Search ignores llms.txt files.

Good answer: a named list of monthly deliverables across those areas, and who does each.
Red flag: "our platform optimises you automatically", or a plan that is mostly schema and an llms.txt file.

6. Which tool do you measure with, and do I get a login?

Agencies measure with third-party tracking tools; our round-up of AI visibility tools is a starting point if you want to know the options. Google says no third-party tool has access to its internal ranking or AI systems, so treat claims of "inside" data with suspicion.

Good answer: they name the tool, give you read access or regular raw exports, and the data stays yours if you leave.
Red flag: a proprietary "AI index" that nobody outside the agency can check.

7. Who owns the accounts, content and data?

That covers your Google Business Profile, Search Console, website, social accounts and anything published in your name. Google's hiring guide suggests granting only read access to Search Console at the audit stage.

Good answer: everything is created in your name, with the agency added as a manager or user, and content you paid for is yours.
Red flag: profiles or pages set up under the agency's own accounts "for convenience".

Questions 8 to 12: reporting, price and honesty

8. How often do you report, and what is in it?

Good answer: the same prompt set re-run on a fixed schedule, a monthly summary at minimum, changes explained engine by engine, and completed work listed next to any movement. For Google's own features, they also show you the generative AI performance report in Search Console, which Google's guide points site owners to.
Red flag: a prompt set that changes every month, so nothing is comparable.

9. What does it cost, what is the minimum term, and what happens when I leave?

Good answer: a fixed monthly or project price in writing, a clear notice period, and a handover list: reports, prompt set, content and logins.
Red flag: a long lock-in with no break clause, or fees tied to "AI rankings" only the agency can verify.

10. What will you refuse to do?

The answer tells you more than a case study. Google's Maps content policy says a review should reflect a genuine experience with the business. Its spam policies treat paying for links that pass ranking credit as link spam unless the links are marked sponsored or nofollow. Its AI guide warns that mass-producing pages for every variation of a question to manipulate AI answers breaches the scaled content abuse policy, and that chasing inauthentic mentions is less useful than it looks.

Good answer: no fake reviews, no review gating, no undisclosed paid placements, no fake listings, no mass-produced pages.
Red flag: "we have a network of sites that can mention you", with nothing said about disclosure.

11. Do you guarantee results?

Google's hiring guide is blunt: "No one can guarantee a #1 ranking on Google." AI answers re-rank from day to day, which makes promises harder to keep, not easier.

Good answer: they commit to work done and measurement published, and say which prompts they expect to move first and why.
Red flag: "guaranteed placement in ChatGPT", or a percentage uplift written into the contract.

12. How will we know it is working?

Good answer: a baseline before work starts, the same prompts re-measured later, and movement on the prompts they worked on compared with the ones they did not touch. That comparison is the closest thing to proof this field has.
Red flag: success measured in activity ("40 posts published") or in branded prompts.

High Street's own answers to these questions are on our AI search visibility service page.

Questions people ask

Sources

• Google Search Central, Do you need an SEO?: questions about AEO/GEO advice, no guaranteed #1 ranking, read-only Search Console access at audit stage, small businesses doing their own SEO; page last updated 5 June 2026 (read 24 September 2026)
• Google Search Central, Optimizing your website for generative AI features on Google Search: AEO and GEO as SEO, structured data not required, llms.txt ignored by Google Search, scaled content abuse, inauthentic mentions, no third-party access to internal systems, Search Console generative AI performance report (read 24 September 2026)
• Google Search Central, Spam policies: link spam and qualifying paid links; page last updated 28 August 2026 (read 24 September 2026)
• Google Maps user-generated content policy, Prohibited and restricted content: reviews must reflect a genuine experience (read 24 September 2026)
• Peec AI pricing: three chosen models per plan below Enterprise, extra models as add-ons (read 24 September 2026)
• High Street research: Where AI answers come from (11 September 2026) and UAE AI visibility report (8 August 2026)

Drafted with AI assistance from our own research and the sources linked above, and reviewed by the High Street team before publishing. Prices, features and policies change; check the linked source before you act.