Questions to ask before buying an AI visibility tool

Send the list before the call

Every question below works as an email. Send them a day ahead: a vendor who answers in writing saves you the hour, and one who does not has revealed how confident they are in their own methodology before you have taken your coat off.

Read the replies for shape, not just content. A methodology question answered with a feature list, a pricing question answered with a tier name, and an engine question answered with a bare count are all answers to a question you did not ask. One evasion can be a busy week; a pattern across the set means the demo will be the same.

Coverage: which engines and how often

Coverage questions come first because everything else in the demo is built on top of them. You want names and cadence, not counts.

1. "Which exact engines do you track, by name, on the plan I would buy, and how often does each one run?" An engine count without names can mean four engines on your tier and the rest as paid add-ons, sampled monthly instead of daily. A good answer names the engines and the schedule in one breath, without opening a pricing page.

2. "Can I see performance per engine, or is everything blended into one score?" A blended average hides the engine where you are invisible, which is usually the one your buyers opened today. A good answer switches straight to a per-engine view in the product.

Methodology: how answers get collected

Methodology decides what you are measuring. Neither collection method is dishonest, but a vendor who cannot tell you which one they use does not know their own pipeline.

3. "Do you collect answers through the official APIs or through browser automation?" APIs are cheaper and repeatable but reflect the API surface; browser automation sees the consumer interface but needs infrastructure and breaks when pages change. A good answer names the method and the reason for it.

4. "How do you handle the same prompt giving different answers on different runs?" Answer variance is the reason single screenshots mislead, so the collection schedule and sample policy are the methodology. A good answer talks about repeated runs, accumulating answers per prompt, and how noise gets separated from movement.

The numbers: what can be trusted

These two questions decide whether a chart in the demo can survive your boss asking where the number came from.

5. "What sample size sits behind this number, and when do you hide a score?" A percentage with no denominator cannot be distinguished from a slow day. A good answer points at a line like "based on 18 answers" and describes hiding scores until the sample clears a minimum instead of showing a confident 100% on day two.

6. "What does a 95% confidence interval mean in your tool?" Scores move, and only the interval tells you whether the move is real: it is the range the true value most likely falls in. A good answer explains the phrase in plain language and shows one next to a live score; a blank stare means the intervals are decoration or absent.

Pricing: what you will actually pay

Entry prices are marketing material. The bill is a function of your prompt count, your engines and your seats, so ask about those directly.

7. "How many prompts does this plan include, and what does an extra prompt cost?" Prompt caps decide the bill more than the sticker price, and overage rates vary widely between vendors. A good answer gives the cap, the overage, and what happens at the cap in the same reply.

8. "Which engines and features are paid add-ons at my usage level, and what does the plan cost a year from now?" Add-ons discovered after the trial are budget you never agreed to, and annual commitments lock in a volume assumption made on signing day. A good answer walks your real numbers through the pricing page and names every add-on.

Exports and API: getting your data out

Data portability is the question buyers skip and later regret, because the answers you collected are the part with lasting value.

9. "Can I export every original answer, and is there an API?" Without raw answers, a disputed number ends in an opinion contest, and without an API you cannot feed your own warehouse. A good answer shows an answer archive on the spot; advisible keeps every original answer behind its metrics and exposes a read-only REST API.

10. "If I cancel, what happens to my answers and history?" Data held hostage is a switching cost you only discover at exit. A good answer says exports stay available in a standard format and names the retention window without hedging.

The company: who you are buying from

AI visibility tooling is young and the vendor list reshuffles, so two questions about the company itself are worth more than another feature walkthrough.

11. "What did you ship in the last six months, and what is next?" You are buying a moving target, since the engines themselves change quarterly. A good answer is a dated list with a coherent direction, not a roadmap slide with undated bullets.

12. "Is there a seat limit, and do extra workspaces or clients cost extra?" Unlimited team members is now common, but client workspaces and project caps are not, and agencies feel this first. A good answer states the seat policy and workspace pricing as plainly as the headline price.

Red flags worth ending the call over

Four answers should cost the vendor the sale: a blend-only score with no per-engine view, refusal or inability to show raw answers, an engine count without names, and pricing that stays vague about add-ons until after the trial.

Around them sits the softer warning: confidence without denominators. If every chart in the demo is a clean percentage and not one of them says how many answers it summarises, you are looking at a screenshot factory. Ask the sample-size question again and watch whether the second answer improves.

The full evaluation rubric, including the decision on collection method and the prompt math, is in how to choose an AI visibility tool, and the market context with dated prices in best AI search visibility tools.