How to choose an AI visibility tool

What the tool actually replaces

An AI visibility tool replaces you, or a junior marketer, asking the same questions in five chat windows every week and writing the results into a spreadsheet. Everything beyond that is a question of how trustworthy and how convenient the replacement is.

Framed that way, the purchase criteria write themselves: does it ask the questions you care about, on the engines your buyers use, often enough that the numbers are a sample rather than a mood, and does it show you enough raw material that you believe the summary? A tool that is weak on any of those will quietly reintroduce the manual work you were trying to remove.

The eight-point checklist

Run every candidate through this list before you look at dashboards. Dashboards are the easy part; these eight decide whether the dashboard can be trusted.

1. Engines covered, by name, per tier. Get the actual list for the plan you would buy, not the vendor's total. Is Claude there? Grok? Are Google AI Overviews or AI Mode included or add-ons?
2. How answers are collected. Official APIs or browser automation. Both are defensible; they answer slightly different questions, and the choice affects cost and repeatability.
3. Sample sizes on every number. If a chart does not say how many answers it summarises, you cannot tell a trend from a slow day.
4. Prompt limits and what happens at the cap. What does the tier include, what does an extra prompt cost, and do engines count separately?
5. Access to raw answers. Export, archive or API. Without the originals, a disputed number ends in an opinion contest.
6. Pricing model. Per prompt, per model, per seat, credit-based, or usage-based. Map it to your real volume before comparing sticker prices.
7. Seats and clients. Unlimited team members is common now; workspaces and client separation often are not. Agencies should check this first.
8. Perception and fact-checking. Mentions are the floor. Whether the answers describe you correctly, and how often they repeat objections, is where the next round of competition is.

Ask how answers are collected

Browser-scraped tools see the consumer interface exactly as a logged-in user does, including UI-only features. API-based tools call the official endpoints, which are cheaper, more repeatable and easier to schedule, but reflect the API surface rather than the app.

Neither is dishonest; they answer different questions. If your concern is “what does a person see in the app”, scraping is closer. If your concern is stable daily measurement you can compare next month against this month, the API path is easier to trust, because pages and layouts change while API contracts do not. The cost difference is real on both sides: scrapers need infrastructure and get blocked; API calls bill per token and need keys.

The full trade-off, with the maintenance angle, is in API vs scraped AI answers.

Look for sample sizes before dashboards

Any vendor can build a beautiful chart. Ask to see the number underneath it: how many answers, over how many days, per engine. If the answer is missing, the chart is a mood ring.

Two habits separate serious tools from screenshot tools. First, they show the count with the figure (“68% based on 18 answers”) instead of a bare percentage. Second, they suppress scores when the sample is too small to read, rather than showing a confident 0 or 100 on day two. Interval display, where a score ships with the range the true value most likely falls in, is the strongest version of this and is still rare.

Price the plan you will actually run

Entry prices are marketing. The bill is a function of prompts × engines × runs, plus whatever the vendor charges for the engines and seats you need, so calculate that first and compare totals.

• Prompt bundles: many tools sell a monthly prompt count (15, 100, 400 are common shapes) and charge extra prompts on top.
• Engine add-ons: some vendors include four engines and charge separately for Gemini, Claude or AI Mode. “Seven engines” may be a paid configuration, not the plan you were quoted.
• Bundled usage vs bring your own key: most tools bake model costs into the subscription. A bring-your-own-key tool instead calls the provider with your own API key, so you pay the provider's published token prices with no markup, and the subscription (if any) is only for the software. The math favours bringing your own key at high run counts and bundled plans at very low ones. More detail in bring your own key vs bundled pricing.
• Seats and workspaces: unlimited team members is now standard at several vendors, including advisible; client workspaces and project caps vary widely.

Questions to ask in every demo

Send these before the call. A vendor who answers them in writing saves you an hour and tells you how confident they are in their own methodology.

1. Which exact engines are included in the plan I would buy, and which are add-ons?
2. Do you collect answers through the official APIs or through browser automation?
3. Show me where the sample size appears on this chart, and what happens when the sample is small.
4. How many prompts does the plan include, and what does an extra prompt cost?
5. Can I export every original answer, and is there an API?
6. What happened to your product roadmap in the last six months? (Look for cadence, not confetti.)

The longer, scored version of this list lives in questions to ask in an AI visibility demo, and the market context for all of it in best AI search visibility tools.

Who should not buy a tool yet

If you have fewer than about ten real buyer questions and one person curious about the answer, do the manual audit first. A tool on top of a vague question produces confident charts nobody acts on.

Do the two-afternoon version: write 10 buyer questions, ask them on ChatGPT and Perplexity this week and next, paste the answers into a sheet, count mentions by hand. You will learn whether AI answers matter to your category and which prompts deserve tracking. The DIY starting point is free ways to track AI visibility.

Buy when one of three things is true: the prompt list has grown past what a spreadsheet holds, more than one person needs the numbers, or a client or executive asks for a report you cannot produce honestly by hand. advisible has no seat limits, which makes the “try it before the spreadsheet rots” path unusually cheap.