Choose a provider by asking eight questions: how it tests, which prompts it tracks, how it reports, who owns the content, how long the contract runs, what proof it can show, what it refuses to promise, and who approves what goes live. Walk away from anyone who guarantees a position in ChatGPT or Google AI answers.
AI visibility work is new enough that nobody has decades of results to show, which makes the questions more useful than the brochure. A good provider answers each one plainly and in writing. A weak one answers with a dashboard screenshot. The questions below are written for a small firm, say a nine-person HR consultancy, where one partner will read the proposal at the weekend and has to decide on Monday.
First, know what "visibility" means before you buy it
AI visibility is how often an AI assistant names your firm, links to your pages or describes you accurately when someone asks a question your clients ask. It is measured by running the same set of questions repeatedly and counting. That is all, and it is why the method matters more than the logo on the proposal.
Three things can move it, and a provider should be able to say which of them it works on. Your own pages can answer questions more clearly. Other sites can mention and describe you consistently, such as directories, professional bodies, press and review sites. And your facts can be easy to find and free of contradictions: opening hours, services, team, prices if you publish them. Nobody can edit the assistant itself.
If you want the basics before you read proposals, what generative engine optimisation is and whether your business needs it covers the idea in plain terms, and how to improve AI SEO so your firm is cited goes through the page-level work.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowEight questions, with what a good answer sounds like
1. How do you test, and what do you do about answers that change?
The same prompt can return different answers on different runs, on different assistants and for different users. A single screenshot of your firm being named proves almost nothing, because another run might not name you. A good answer describes repeated runs: each prompt asked several times, on each assistant you care about, with the share of runs that mention you reported as a rate.
Weak answer:
"We check ChatGPT regularly and send you screenshots of where you appear."
Good answer:
"We run each of your 40 prompts five times a week on three assistants, log
every answer, and report the percentage of runs that name your firm, the
percentage that link to your site, and which competitors appear instead.
Results change between runs, so we report trends over four weeks, not single
screenshots."Ask also whether the checks are run logged out or through an account with history, because personalised answers differ. There is no single right setting, but the provider should know which one it used.
2. Which prompts do you track, and who chooses them?
The prompt set is the whole measurement. A provider that tracks questions nobody asks will report improvements nobody feels. For a small firm, thirty to sixty buyer questions is enough. As a yardstick, the Starter plan in Semrush's SEO + AI Search range ($199 a month, or $165.17 a month billed annually) tracks 50 prompts daily for one domain, and Profound's published trial runs 50 unique prompts daily for seven days across ChatGPT, Gemini and Google AI Overviews, from a recommended prompt set that trial users cannot change. Neither number is a target, but a proposal that tracks five prompts, or five hundred, needs an explanation.
Here is a filled-in sample set for the HR consultancy. Provider templates tend to be all category one. Yours should include real enquiries.
| Type | Example prompt (illustrative) | Why it is in the set |
|---|---|---|
| Problem | "How do I handle a grievance in a team of twelve without a full-time HR person?" | Matches the first email a prospect sends |
| Comparison | "Outsourced HR consultancy or hire an HR manager for a company of forty?" | Buyers weigh options before they shortlist |
| Shortlist | "Which HR consultancies work with small manufacturers?" | This is where a firm is named or is not |
| Brand | "What does [firm name] charge and who is it for?" | Checks accuracy of what the assistant says about you |
| Competitor | "[competitor] alternatives for small business HR" | Shows who is named in your place |
| Process | "What happens in a first meeting with an HR consultant?" | Early research, cheap to answer well |
Ask the provider to show your set before work starts and to let you add and remove prompts. If it will not share the prompt list, you cannot judge the reports.
3. What will the report contain, and how often?
Ask for a sample report from a current client with the name removed. Look for definitions: what "visibility score" means, how it is calculated, what counts as a mention, and what counts as a citation (the assistant linking or naming one of your pages). A single score with no definition is the commonest way to make flat results look like progress.
A useful monthly report has five parts: mention rate by assistant, which of your pages are cited, which competitors appear in your place, inaccurate statements about your firm and what was done about them, and the work completed. Beyond the report, ask how they connect it to enquiries. The honest answer is partial: many visits from assistants show up as referral traffic in analytics, and a "how did you hear about us" field on your contact form catches some of the rest. Google Search Console's generative AI performance report shows impressions only, with AI Overviews and AI Mode combined, so a provider quoting clicks from AI Overviews should say where that number comes from. How to track whether AI search is sending you customers sets up both.
4. Who owns the content, the accounts and the data?
The proposal should say, in plain words, that you own every page, schema file, article and image produced, with copyright assigned to you on payment. It should also say the work lives on your website and your accounts, not on a subdomain or network of sites the provider controls. If you leave, you should keep everything, including the prompt set and the historic tracking data.
Ask three specific things. Who holds the admin login to your site changes, analytics and tag manager? Does the provider publish anything on sites outside your control, and if so which and why? And will the provider use client data, such as case details or staff information, in any AI tool? For an HR or advisory firm, the last one is a confidentiality issue before it is a marketing one.
5. How long is the contract, and how do I leave?
Changes to pages take weeks to show up in assistants, so providers like long terms. That is a fair reason but not a fair excuse for a twelve-month lock-in before you have seen a report you trust. A reasonable shape for a small firm is an audit with a fixed fee, then a three-month term with a one-month notice period, then monthly.
Look at the exit clause as closely as the price. You want written answers to: what is the minimum term, what is the notice period, are there cancellation fees, and what is handed over on exit. I will not give a market price here, because published retainer prices are rare and vary widely. Get at least three written quotes on the same scope, then compare the cost to a do-it-yourself route using the sum later in this tutorial.
6. What proof can you show?
Be careful with proof, because it is easy to fake and hard to check. Ask for one case where the provider can show the same prompt set measured before and after, with dates, the specific page or mention changes made in between, and a client willing to take a reference call. "We grew a client's AI traffic 300%" with no starting number is not evidence. Starting from zero, almost any change is a large percentage.
Also ask for a failure. A provider with real experience can describe a project that did not move and why. One that has never had a miss is either new or not telling you.
7. What will you refuse to promise?
This is the question that sorts providers fastest. The right answer includes rankings, mentions in a named assistant, a date by which you will be named, and any percentage increase. A provider can promise activity and quality: pages fixed by a date, a tracked prompt set, reports on time, accurate descriptions corrected. It cannot promise the outcome inside someone else's system.
- "Guaranteed first-page placement in AI answers" is a red flag, since there is no page and no position to guarantee.
- "We have a direct relationship with the AI companies" deserves a request for evidence, because there is no paid route to being named in an ordinary answer.
- "Our proprietary AI visibility score" with no definition should be treated as marketing.
- Bulk-produced pages, bought mentions or invented reviews risk harm that outlasts the contract, including a poor reputation with search engines.
8. Who approves what goes live, and how are regulated claims handled?
A small advisory firm, whether HR, financial advice or mortgages, carries responsibility for every claim on its site. Ask who signs off on each page, how long that takes, and what happens to wording about qualifications, outcomes and fees. The safe answer is that nothing is published without your written approval, and that the provider flags any sentence that makes a claim of result or a regulated statement for you to check with your compliance adviser.
The same question applies to AI-written drafts. Ask whether they use AI to draft content, whether a named human edits it, and whether the draft passes a check for invented facts before you see it.
A short email that sends the eight questions
You do not need a formal tender. Send each shortlisted provider the same message and compare answers side by side.
Subject: Questions before we decide on AI visibility support
Hello [first name],
Before we go further, please answer these in writing:
1. How do you test, and how do you handle answers that change between runs?
2. Which prompts would you track for us, and can we add our own?
3. Please send an anonymised sample report and your definition of each metric.
4. Who owns the content, the accounts and the tracking data, including on exit?
5. What is the minimum term, the notice period and any cancellation fee?
6. Can you show one before-and-after case with dates, and one that did not work?
7. What do you not promise?
8. Who signs off content before it is published, and how are regulated claims handled?
Thank you,
[your name]Treat the quality of the reply as part of the test. Slow, vague or defensive answers before you have paid are a fair preview of the service after you have.
The do-it-yourself audit: a benchmark in about 90 minutes
Run this yourself before you hire anyone, or alongside a provider's first report. It costs nothing but time and shows you what a provider should be improving.
- Write 30 prompts in the six types from the table above, using real wording from client emails and enquiry calls.
- Pick three assistants your clients are likely to use. Open each in a fresh session, logged out if you can.
- Ask each prompt once on each assistant, which is 90 answers, and paste them into a sheet with columns for prompt, assistant, firm named (yes or no), your page cited (yes or no), competitors named and anything wrong about you.
- Count: the share of answers that name you, the share that cite your site, and the three competitors named most often.
- Repeat the same sheet a week later on ten prompts to see how much the answers move on their own. That movement is your noise level, and any provider's claimed gain has to be bigger than it.
What the sheet might look like (illustrative)
Say the HR consultancy runs the audit and finds it is named in 7 of 90 answers (about 8%), cited in 3, and that two competitors appear in 31 and 24 answers. Four answers contain errors: one says the firm only serves large companies, and one quotes an old fee. A week later, on ten prompts, the firm's mentions moved from 2 to 1 with no changes made. That tells the partner three useful things: the baseline is low, two competitors dominate the shortlist prompts, and a one-mention swing is noise, not progress. It also gives her a concrete first fix, the errors on her own pages, which costs nothing.
The cost sum: do it yourself, tool, or provider
Put the three routes side by side for the same firm. The provider figure is an invented quote; replace it with your own.
| Route | Monthly cost (illustrative) | Six-month cost | What you get |
|---|---|---|---|
| Do it yourself, spreadsheet only | 3 hours of a partner's time at an assumed $90 an hour = $270 | $1,620 | Baseline, trend, list of fixes, no tooling |
| Tracking tool plus 2 hours a month | $199 (Semrush SEO + AI Search Starter, monthly billing) + $180 time = $379 | $2,274 | Daily tracking of 50 prompts, one domain, automated reports |
| Provider retainer on a six-month minimum | Assumed quote of $1,500 | $9,000 | Tracking, page work, outreach, reports, if the scope is as promised |
The tool row uses monthly billing because the comparison runs six months; annual billing brings Semrush to $165.17 a month but commits you for a year. The provider is not automatically the wrong choice. It is worth paying for when the work needed is more than your team can do, such as rewriting a dozen service pages, fixing structured data and getting the firm described consistently across third-party sites. It is poor value when the main output is a report you could produce yourself. A useful test: ask the provider to list the five changes it would make in month one. If none of them touches your pages or your off-site mentions, you are buying monitoring.
For the cost of monitoring alone, the comparison of AI brand visibility software for small businesses sets out the tools side by side, and how to choose an SEO agency that understands AI search helps if your current agency is about to add this service to the invoice.
Reading one proposal line by line
Here is a paragraph of the kind that turns up in proposals, followed by what to ask about each phrase. The text is invented for illustration.
"Our AI Authority Package places your firm at the centre of AI-generated
answers within 90 days. We build a network of authority content across
trusted platforms, monitor your AI Visibility Score weekly and deliver a
monthly performance report. Minimum term 12 months.""Within 90 days" is a promise about an outcome inside someone else's system, so question 7 applies at once. "A network of authority content across trusted platforms" does not say which platforms, who owns the posts, or whether they carry your firm's name, so questions 4 and 8 are open. "AI Visibility Score" has no definition, which is question 3. "Minimum term 12 months" with no exit wording is question 5. In total, one sentence fails four of the eight questions before a price is mentioned.
Send the email above in reply and let the answers decide. A provider with a real method will usually welcome the questions, because it has already answered them for other clients.
Scoring the proposals you receive
Give each proposal a score out of 40: five points for each of the eight questions. Five means written, specific and checkable. Three means plausible but vague. Zero means evasive or missing. The numbers are a prompt for a conversation, not a verdict, but the pattern is revealing: providers tend to score well on reporting and badly on ownership, exit and what they refuse to promise.
| Score out of 40 | Reading |
|---|---|
| 32 or more | Credible. Run a three-month term and check reports against your own spot checks. |
| 24 to 31 | Ask for the gaps to be fixed in writing. Do not sign until they are. |
| Under 24 | Keep your money, run the DIY audit and fix your own pages first. |
The first month with a provider: checks that keep it honest
If you sign, hold the provider to a simple routine so the reports and your own experience do not drift apart. In week one, run your own ten-prompt spot check and save the answers, so you hold a baseline the provider did not produce. In week two, review the prompt set with them and remove anything your clients never ask. At the end of month one, compare their mention rate with your spot check; if the two disagree by a wide margin, ask why before the next invoice.
Ask for a changelog too: every page edited, every mention gained, with dates. Without one, you cannot tell which work moved which number, and you cannot repeat the successes. If you would rather work through the proposal with someone, that is what my AI implementation consultation covers: finding the jobs worth paying for and handing over something your team can run.
Follow-up questions before you sign
Can anyone guarantee my firm will be named in ChatGPT answers?
No. The same prompt can return different answers on different runs, for different users and after model updates, and no outside party controls them. A provider can improve your pages and your mentions elsewhere. Treat any guarantee of a mention or a position as a reason to walk away.
How many prompts should a provider track for a small firm?
Thirty to sixty buyer questions covers most small firms. Semrush's SEO + AI Search Starter plan tracks 50 prompts daily for one domain, which is a useful yardstick. Ask who chooses the prompts, and insist that some come from your own client enquiries rather than the provider's template.
Is a monthly retainer better than a one-off audit?
Start with a one-off audit or a three-month term. Fixes to your pages take weeks to show up, and a short term lets you check the reports against your own spot checks before committing a year. Longer terms should come with a clear exit and ownership of everything produced.
Should I let a provider publish content under my firm's name?
Only if you approve each piece first and own it outright. A regulated or advice-giving firm carries responsibility for every claim on its site, so the contract should give you sign-off, the copyright, and a copy of everything if you leave.
Further reads
- Semrush AI Visibility vs Ahrefs Brand Radar: Which Suits an Agency? — Compares two tracking tools many providers resell.
- AI Visibility Tracking for B2B: Affordable Tools and Free Checks — Cheaper tracking tools and free checks for your own audit.
- How to Get Your Business Recommended by ChatGPT and AI Search — The general steps that make a business easier to recommend.
- What Are the Best AI Visibility Products for a Local Business? — Product comparison for businesses that serve a local area.
- Can Estate Agents Be Found in AI Search? What to Fix Now — A sector example of what to fix first.
- Consultant SEO for ChatGPT Search: Get Your Practice Recommended — Make a consulting practice easy for AI search to identify and cite: entity pages, case studies, reviews, schema and a fixed prompt test you rerun monthly.
- How Does ChatGPT Find a Business? What a Professional Firm Can Control — ChatGPT recommends firms from web search and public pages. Here is what it reads, what you control, and a simple test to see where you stand.
- How to Monitor Brand Mentions in ChatGPT for a Client, Week by Week — A fixed prompt set, repeat runs, a log sheet and a weekday schedule for tracking a client's ChatGPT mentions, with tool costs and wrong-answer fixes.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Semrush pricing page and Profound pricing page (read November 2026); Google Search Console's generative AI performance report as described in Google's documentation. Everything else is method and judgement, with no vendor claims relied on.