10 Questions to Ask Any AI Visibility Vendor Before You Pay

AI search visibility is a young market. Dozens of tools now sell a version of the same promise: we will tell you whether ChatGPT, Perplexity, Google AI, and the other engines mention or cite your brand. The promise is real and worth paying for. The problem is that the numbers behind it are produced in very different ways, and most vendors do not disclose which way they use.

Two tools can give the same brand very different visibility scores for the same week, and both can look precise on a dashboard. The difference is almost always methodology: what was asked, of which engine, through what access, how often, and how the answer was counted. None of that is visible on a pricing page, which is why you have to ask.

These questions are vendor-neutral. Any honest tool, including ours, should be able to answer every one of them in writing. The goal is not to find a perfect vendor. It is to find one whose numbers you could defend to your own boss or client.

The single most important thing to establish is the data source per engine, because it determines whether the numbers are stable, legitimate, and reproducible.

• +1. Which engines do you actually query today, and which are estimated, modeled, or planned? Ask for the list in writing. Many tools advertise a headline engine count where only some engines are queried live and the rest are inferred.
• +2. For each live engine, what is the access method: an official API, licensed data from the platform, or scraping the consumer interface? Official APIs and licensed data are stable and permitted. Scraping a consumer product like the ChatGPT web app typically violates the platform's terms of use, breaks without warning, and can misrepresent what real users see.
• +3. If an engine has no official citation API, how do you cover it, and do you label that coverage differently? The honest answers are licensed data, clearly labeled estimates, or we do not cover it yet. The dishonest answer is silence.
• +4. How often is each engine re-checked, and does the cadence differ by plan? Visibility moves. A number measured monthly is a different product from one measured weekly, and you should know which you are buying.

AI engines are probabilistic: the same question can produce different answers on different runs. A serious measurement tool manages that; a dashboard that ignores it is printing noise with confident typography.

• +5. Are your results reproducible? If you probe the same prompt twice in the same week, how much do results vary, and do you disclose that variance? Vendors who sample multiple runs, or fix model settings where the API allows it, will say so proudly. Vendors who do not will change the subject.
• +6. Can I see the actual prompts you probe for my brand, or are they aggregated away? You cannot judge whether a prompt list represents your buyers unless you can read it. Tools that hide prompts behind keywords or topic scores make their numbers unfalsifiable.
• +7. Are your visibility numbers direct measurements, statistical estimates, or model outputs, and are they labeled per engine? All three can be legitimate. Mixing them in one score without labels is how a marketing number gets dressed up as a measurement.

The last three questions are about incentives: what the vendor claims, what they guarantee, and whether you are locked in.

• +8. Do you claim to see the real prompts people type into ChatGPT or other assistants? Nobody outside the platforms can see private user conversations. Vendors estimate demand from search data, clickstream panels, and their own probes. Estimation is fine; claiming to read real user prompts is not.
• +9. What exactly do you guarantee? No honest vendor can guarantee a citation, a ranking, or a number one spot in any AI answer, because the engines are not theirs to control. What can be guaranteed is the work: what gets measured, how often, what gets delivered, and what happens if it is not.
• +10. Can I export my data and history if I leave? Visibility history is most valuable as a long time series. If the answer to can I take it with me is no, the tool owns your baseline, not you.

A few patterns reliably signal marketing outrunning measurement. Each has an innocent version. Together they mean the burden of proof has shifted to the vendor.

• +A guaranteed number one spot in ChatGPT, or guaranteed citations. The engines are not theirs to promise.
• +A big engine count with no per-engine methodology page. Coverage claims are cheap; sourcing disclosures are not.
• +Claims of access to real user prompts or real AI conversations without naming the data source and its consent basis.
• +Scores that never show variance, sample size, or a measured-on date. Real measurements have error bars and timestamps.
• +No way to see the underlying prompts and citations behind your score. If you cannot audit it, you cannot trust it.

We sell one of these tools, so here is Aeody against its own checklist. We publish a per-engine coverage table on our methodology page showing exactly what is live and what is planned, and we update it as engines come online. Every live engine is queried through its official API. We never scrape the ChatGPT consumer interface or any engine's logged-in product. Deterministic checks are reproducible by design, scoring runs at fixed model settings where an API allows it, results are labeled per engine, you can read every prompt we probe for your brand, and your data is yours to export.

And the two answers that matter most: we do not claim to see the real prompts people type into AI assistants, because nobody outside the platforms can. And we guarantee the work, never the outcome. Estimates, not guarantees. If a vendor you are evaluating answers these ten questions as plainly, they are worth considering, whoever they are.