What Should I Ask an AI Visibility Vendor Before Signing Up?
September 10, 2026
Ask how many times each prompt is run per assistant, whether you can see the cited sources, and what happens after the tool finds a gap. Those three questions eliminate most of the field. The rest of this list is the detail behind them.
Table of Contents
The eleven questions
1. How many times do you run each prompt, per assistant? Assistants are not deterministic. The same question asked three times can produce three different answers. A tool that asks once reports a coin flip as a fact. You want a rate, for example named in seven of nine responses, not a yes or no.
2. Which assistants, and can I see the raw response? Coverage lists are marketing. Seeing the actual answer text is what lets you argue with the score.
3. Do you show the sources the answer cited? This is the single most actionable field in the whole category. The cited domains are your publishing target list. A tool that hides them is telling you the score and withholding the plan.
4. Can I segment by buying stage? Discovery, evaluation and decision questions behave differently, and presence is rarely even across them. Without stage segmentation, an average hides where your actual opening is.
5. Can I use my own prompts? Vendor prompt sets are generic. Your buyers ask specific things. If you cannot add your own questions, you are measuring someone else's category.
6. Does the prompt set stay stable over time? If prompts change silently, your trend line is meaningless. Stability is what makes a before and after reading credible.
7. What happens after you find a gap? Most tools stop here. Ask directly what the product does next, and be suspicious of an answer that is really a report.
8. Can I re-scan on demand after publishing a change? Verification is the only way to know a change worked rather than the answer drifting on its own.
9. Do you track competitors on the same prompts? Absence is only half the story. Knowing who took the slot tells you what the model considers a credible answer in your category.
10. What is the real price at my volume, and what is metered? Ask what happens when you add prompts, competitors or re-scans. Credit models can be fine, but you should know the shape before you sign.
11. Can I export my history if I leave? Your baseline is the asset. Do not let it live somewhere you cannot take it from.
What good answers look like
A good vendor will answer question 1 with a number, question 3 with a screenshot, and question 7 without changing the subject. A vendor who deflects on repeat runs is selling a single-shot query dressed as measurement.
If you want a baseline to bring into those conversations, run the free AI visibility check first. It costs nothing, needs no signup, and gives you a number to hold every demo against. The tool comparison covers pricing and coverage across the main options, including ours, and how Monroya works answers question 7 in detail.
FAQ
Should I run a pilot with two vendors at once? It rarely helps. Different prompt sets and scoring produce different numbers and neither is wrong. Pick one and measure against your own baseline.
How long a trial do I need? Long enough for one full cycle: baseline, publish one change, rescan. For most teams that is two to four weeks.
Is a cheaper monitoring tool enough to start? Often yes, if you will actually look at it. The upgrade trigger is when you stop asking whether you are missing and start asking what to publish.