Costs, returns and the business case · Guide 6 of 11

How do we choose a firm to help us get recommended by ChatGPT?

Every agency now claims it can get you named by ChatGPT. The questions to ask, the red flags, what can fairly be guaranteed and what a report worth paying for looks like.

A managing director in her fifties and a finance director in his forties at a meeting table, three printed proposals spread out and one marked with a lime-green sticky note, natural daylight.
On this page · 11 sections
  1. What are you actually buying?
  2. Which kind of provider covers which part?
  3. What questions should you ask every provider?
  4. What can a provider fairly guarantee?
  5. Which red flags should make you wary?
  6. What does good reporting look like?
  7. How do you compare proposals side by side?
  8. How should the first three months be structured?
  9. What to do this month
  10. Questions founders ask
  11. Sources

If you run a B2B firm, your inbox has probably filled up with offers to "get you into ChatGPT". Your SEO agency has added a paragraph about AI to its proposal. A software company wants to sell you a dashboard. A PR firm says coverage is what AI assistants read. Some call the work GEO (generative engine optimisation) or AEO (answer engine optimisation); others call it AI SEO. They all mean the same thing: getting your firm named when buyers ask ChatGPT, Gemini, Copilot, Claude or Perplexity who to hire.

The labels matter less than they seem. Google's own guide to optimising for its generative AI features (May 2026, updated July 2026) names both GEO and AEO and says that, from Google's side, the work is still SEO. UK searchers mostly use the plainer term: over the 12 months to 24 September 2026, "AI SEO" drew about nine times the UK search interest of "generative engine optimization" on Google Trends. Whatever a provider calls it, judge what it will do.

Most founders we speak to have been here before with other agencies: a retainer, a monthly report full of charts, and no customer anyone could trace to it. We sell this kind of work ourselves, so read what follows with that in mind. It is the guide we would want on your side of the table, and it applies to any provider, including us.

What are you actually buying?

Three kinds of work and a way of measuring them:

  • A website AI assistants can read and understand. Pages that load properly for automated readers, say plainly what you do and for whom, and state key facts consistently, including in structured data (a small block of code that labels facts for machines).
  • Content that answers your buyers' real questions. Service pages, guides and case studies that a buyer, and an assistant, could quote.
  • Independent mentions. Trade coverage, directory profiles, reviews and awards that confirm what your website says. This is where most citations come from: Muck Rack's May 2026 study of more than 25 million links cited by ChatGPT, Claude and Gemini found 84% were earned media.

On top of that you need a consistent way of checking what assistants say about you. Our guide on whether you still need SEO now buyers ask ChatGPT explains how the old work and the new overlap.

Which kind of provider covers which part?

Most providers grew out of one of the three, which is fine as long as nobody sells their part as the whole job.

Provider type Usually strong on Usually misses How they usually charge
SEO agency Website structure, technical fixes, rankings Earned coverage and reviews; reporting on AI answers Monthly retainer
PR firm Trade coverage and journalist relationships Website fixes; measuring AI answers Monthly retainer or project
Specialist GEO or AI search agency Baselines of AI answers, content built around buyer questions Varies widely: some are rebadged SEO, some have no PR capability Audit, then monthly retainer
Content agency Volume of articles and pages Independent mentions; measurement Per piece or monthly
Tracking software Asking the questions and recording answers at scale Doing any of the work Monthly subscription
Freelancer Specialist skill at lower overhead Breadth and cover when they are away Day rate or monthly

The first test of any proposal is whether it covers all three strands, or says clearly who else will. For what each of these typically costs in the UK, see how much AI search optimisation costs.

What questions should you ask every provider?

Ask these seven in the first meeting, or copy them into an email to each firm on your shortlist. The quality of the answers will tell you more than any credentials slide.

1. What do AI assistants say about us today?

A capable provider will have checked before the meeting, or will offer to. They should show you the actual answers to questions your buyers ask, who was named instead of you, and which sources were cited. A good baseline looks like the one we built for TCC Group: "36 recorded answers from ChatGPT, Gemini and Claude, every conversation kept and reproducible." If a provider cannot tell you where you stand now, it cannot show you progress later.

2. Which buyer questions will you track, and who chose them?

The questions matter more than the tool. They should be the ones your buyers type, such as "best employment lawyers for tech firms in Bristol", not questions containing your own name. Agree a list of 20 to 50 together and keep it stable, so that month-to-month comparisons mean something.

3. What will you change on our website, and why?

Expect specifics: which pages, what is missing, whether AI crawlers can reach the site, which facts disagree between pages. Vague answers about "improving content" suggest a template.

4. How will you earn independent mentions?

Ask which publications, directories and review sites matter in your sector, who writes the pitches, and for linked examples of coverage won for similar firms. Paid placements and bulk outreach are a different thing, as our guide to earned, owned and paid coverage explains.

5. Who will actually do the work?

Ask for names, experience and hours a month on your account, whether any work is subcontracted, and how AI writing tools are used. Using them to help draft is normal. Publishing their output without an expert's input is not.

6. How will we know it is working, and when?

A good answer separates early signs (fixes made, coverage secured) from later results (being named more often, buyers mentioning AI when they get in touch). Assistants that search the web live can reflect a new page or article within weeks; answers drawn from a model's memory change only when it is retrained, which takes months.

7. What happens if we leave?

You should own your content, website changes, accounts and tracking data. Check the notice period and whether anything stops working when the contract ends. The best outcome is work your own team can carry on: for TCC Group, "TCC chose to build it with its own team, and the new tcc.group is live, built to the plan, with nothing about it depending on Fortitude."

What can a provider fairly guarantee?

Only what it controls. Nobody controls what ChatGPT or Gemini says, so be wary of a guarantee of a set number of citations, rankings inside AI answers or AI-sourced leads: it usually means the provider misunderstands the work or plans to measure something easy.

A provider can fairly guarantee effort (named people, hours, pages and pitches delivered), reporting (the same questions, every month, losses included), your ownership of everything produced, and outcomes it controls directly. Our outbound service, Pipeline That Fills, carries a guarantee tied to meetings held, because we run that work end to end. We make no guarantee about AI answers, and neither should anyone else.

Which red flags should make you wary?

  • A guaranteed place in AI answers. Ask exactly what is guaranteed and how it will be measured. Guarantees of effort, reporting and meetings held are fair; nobody can promise what an assistant will say.
  • "We'll get you into ChatGPT's training data." Models are trained on large snapshots of the web taken before release. You can influence future snapshots by being well covered and easy to read, but nobody can insert you.
  • A special relationship with an AI company. ChatGPT has shown labelled ads to UK users on its Free and Go plans since 11 August 2026, and OpenAI says the ads do not change the answer itself. Buying an advert beside the answer is not the same as being named in it.
  • llms.txt or special markup sold as the fix. Google says its search ignores llms.txt and that its AI features need no special markup. Both are cheap housekeeping, not a strategy.
  • A page for every variation of a question. Google's guide warns that mass-producing pages for query variations, mainly to influence AI answers, breaks its spam policy on scaled content.
  • Content at volume with no expert input. Dozens of AI-drafted articles a month give assistants nothing new to quote.
  • Paid placements sold as PR. Sponsored articles and pay-to-enter lists are advertising.
  • The old SEO report with a new name. Keyword rankings and traffic charts do not tell you what AI assistants say.
  • Long lock-ins. A twelve-month minimum with exit penalties puts all the risk on you.

What does good reporting look like?

The same buyer questions every month, what each assistant said, and what the provider did that might explain any change. The method we use for ID-Pal is the standard to ask for: "The same buyer questions run every month in ChatGPT, Gemini and Claude, in a fresh chat. A win only counts when it repeats, and losses are reported too."

A report worth paying for A report to be wary of
The agreed questions, asked more than once in each assistant A single "visibility score" with no working
For each question: named or not, how described, which competitors instead Screenshots of one good answer
The sources the assistants cited Share of voice with no questions behind it
Work done that month, with links and reasons Long activity lists linked to nothing
Enquiries where the buyer mentioned an AI assistant Impressions and reach as the headline
A short plan for next month you can challenge Every month green, nothing ever slipped

Our guide to building an AI visibility dashboard and board report shows the format in full, whether you run the reporting or a provider does.

How do you compare proposals side by side?

Put the proposals on one page and score them on the same things. Price matters, but only once you know what each price buys.

What to compare What good looks like Warning sign
Starting point Real answers from several assistants, shown to you A generic audit or a score with no working
Buyer questions Agreed with you, in your buyers' words Chosen by the provider, or containing your name
Website work Named pages and specific changes General promises with no detail
Independent mentions Named publications and linked past examples Paid placements or bulk outreach only
Team Named people with relevant experience A sales lead you never see again
Reporting Sample report shown before signing "We will tailor it once you start"
Guarantees Effort, reporting and ownership A guaranteed place in AI answers
Terms Short notice period, and you own everything Long minimum term and exit penalties

Score each row from one to three and add them up. When two proposals score the same, the tiebreaker is usually the people, so meet whoever will run your account.

How should the first three months be structured?

Start with a defined first phase rather than a long commitment. Three months is enough to test whether a provider does what it says, though not to judge the full effect on your pipeline. By the end you should have a baseline, the main website fixes done and explained, a content plan tied to buyer questions with the first pieces live, the first independent coverage or a named list of targets, and monthly reports asking the same questions.

If you want the diagnosis before choosing anyone, a fixed-scope piece of work such as our Paid Discovery gives you every finding ranked and fixes written so your own team, or any provider, can ship them.

What to do this month

  1. Write down 20 questions your best clients would have asked before they found you.
  2. Ask them in ChatGPT, Gemini, Copilot, Claude and Perplexity, more than once each, and save the answers as your own baseline.
  3. Send the seven questions above to each provider you are considering, and ask each for a sample monthly report.
  4. Ask each for two client references, including one who has left.
  5. Score the proposals in the table and meet the person who would run your account.

Frequently asked questions

How much does a GEO agency cost in the UK?

It depends on how much of the work is included: an audit, a content programme, earned coverage, tracking or all four. Compare what each proposal buys each month in named people's hours, pages changed and coverage pursued. Our guide to what AI search optimisation costs in the UK sets out published market ranges, and our own prices are on what it costs.

Can a GEO agency guarantee ChatGPT will mention us?

No. Nobody controls what an AI assistant says, and answers vary from one run to the next. An agency can guarantee effort, reporting and your ownership of the work. Treat a guaranteed mention as a reason to ask how they will measure it, because it usually means they plan to measure something easy.

What contract length is normal for an AI search agency?

Terms vary, and there is no industry standard. In our view, a defined first phase of about three months, then monthly terms with a short notice period, suits this work better than a twelve-month minimum. Judge the work at three months and the results over six to twelve.

Should we hire a GEO specialist or train our own marketer?

If you have a capable marketer with time, a clear plan and a tracking method may be enough, with outside help for press relationships and technical fixes. If you have no marketing team, you will need someone to do the work. A middle route is a written plan your team builds from: for TCC Group, the result was "1 week to a plan TCC's own team built the new site from."

Can our existing SEO agency do this work?

Possibly. Much of the website work overlaps with good SEO. Ask them the same seven questions, especially how they will earn independent mentions and report what AI assistants say. If the answer is a renamed keyword report, you need a different provider or a specialist alongside them.

Is a software tool enough on its own?

A tool tells you what assistants say, which is valuable, but it does not change it. For some firms a tool plus a capable in-house marketer is enough. For most founder-led firms without a marketing team, the tool is the starting point and the work still needs doing.

How do we check a provider's claims about past results?

Ask for the client's name, the questions tracked and the answers before and after, then speak to that client yourself. Ask to speak to a client who has left, and find out why. Be wary of percentage improvements with no baseline, and of results you cannot trace to a named firm.

Whoever you choose, start with your own baseline so you can judge every proposal against it. Book a free Growth Review: 30 minutes with David Adams, and three fixes in writing that are yours to take into any conversation, including with other providers. For more on when outside help is worth paying for, see whether you still need a marketing agency.

Sources

David Adams
Written by

David Adams

Co-founder and CEO, Fortitude Media · AI and automation

David spent thirteen years in SaaS, from retention calls to COO and then CRO of a £30 million partner-led business. He runs Fortitude, builds the tools behind it and runs Pipeline That Fills.

  • Websites
  • Getting named by AI
  • The business case
  • Meetings and pipeline
More from David

fortitudemedia.ai/insights/how-to-compare-ai-optimisation-providers · Reviewed 24 September 2026 · Book a Growth Review at fortitudemedia.ai/growth-review

Ready to find out where you stand?

Book a free Growth Review.

Book your Growth Review ↗