Ask These Five Questions Before You Pay

Technical information first: standalone AI-visibility trackers run from $20 a month at the low end to $780 a month at the high end of a published self-serve tier, and neither number tells you what a wrong answer costs you. We've watched that exact confusion happen on sales calls, so we wrote down the five questions we'd want answered before you sign a contract with anybody, us included. We're not going to make you read a spec sheet to get it, and we're not going to rank ourselves first — you don't need a technical background to ask any of these out loud, and you shouldn't have to take a sales deck's word for the answer.

The Price Band You're Actually Choosing From

Rankscale prices its self-serve tiers from $20 a month at Essential to $780 a month at Enterprise, with every tier tracking the same 17-plus engines no matter what you pay (offerings matrix, verified 2026-07-18). Peec AI starts near $95 a month. Otterly starts at $29. We watch this same market from the vendor side, so we know exactly how little any of those three numbers tells you on its own. A WordPress plugin that folds AI visibility into a tool you already run costs less than any of them and does more that isn't AI-visibility at all — but the number on the pricing page is the least useful fact on it, so we're not going to spend this post arguing over it.

Two vendors charging the same $95 a month can be measuring completely different things: one asks three questions of one engine once a week with nothing stored, the other asks twenty questions across five engines, three samples each, daily, with every answer readable afterward. Same sticker, so you're comparing two tabs in a browser with no way to tell which company you're actually looking at until the invoice clears and you go looking for the report.

Question One: Which Model Answered, and When

A visibility number built on a fast, cheap model reads differently from one built on a frontier model, and a vendor who won't say which one is asking you to trust an average across models nobody named. So ask it plainly: which model, which version, and is that written down anywhere you can check later — we store which model answered every question, every time, for exactly this reason. We built that naming rule for our own reports first, so there's nothing here we wouldn't hold ourselves to.

Cost of not asking: you compare this month's report against a competitor's, pulled from a different tool on a different model, and draw a conclusion neither number actually supports — a battle lost to a rounding error nobody flagged.

Question Two: How Many Times Did You Actually Ask

One run of one question is a coin flip wearing a lab coat. Three runs of the same question start to show whether an answer is stable or volatile, which is a fact about the category and not about your brand, so the number is only honest if the vendor tells you how many runs sit behind it. We publish our own run count for this same reason.

Cost of not asking: a single lucky sample gets carried into a boardroom as your company's AI visibility, and next month's number moves for no reason anyone in the room can explain.

Question Three: Is the Search Live, and Is It Labeled

An engine can answer straight out of its training data, or it can search the live web first and answer from what it finds there, and conflating the two is the easiest way to make a stale measurement look current. We label this fact per engine, per question, because the two answers look identical on a chart and aren't identical at all.

Cost of not asking: you fix a page today, and the tracker's number sits still for weeks, because the report was never checking the live web to begin with.

Question Four: Can You Read the Actual Answer

A mention rate is a summary. The sentence an engine actually wrote is the evidence underneath it, and a tool that shows you only the summary is asking you to trust its arithmetic on the strength of its dashboard design alone. We show the sentence itself, every time, for exactly this reason.

Cost of not asking: you can never independently check whether the vendor's definition of a mention matches yours, because you never see what got counted in the first place.

Question Five: What Happens the Day a Call Fails

Every engine times out, rate-limits, or refuses a request on some ordinary Tuesday — that's an outage on the vendor's account, not a fact about your business. A company that counts a failed call as a missed mention is quietly billing its own downtime to your scoreboard, and an outage is not the same finding as invisibility, even though the two look identical on a chart. We exclude those calls from our own numbers too, so a bad day on someone else's server never becomes a bad week on yours.

Cost of not asking: your visibility score drops on a day your brand did nothing differently, and you spend a whole budget meeting fighting a battle that was never actually yours.

Ask It Out Loud

None of the above requires jargon on the actual sales call, before you sign anything. "Which model answered our questions, and can you tell me the version?" "How many times did you ask each one?" "Was web search on when you asked, and how would I know if it wasn't?" "Can I read the actual answer, not just the score?" "What happens to my numbers on a day your own API calls fail?" A company with a clean method answers all five in under two minutes without reaching for a slide. A company without one changes the subject to whatever the dashboard does well instead, and that pivot is the whole answer, so let it be. We ask ourselves these same five questions before we ship a report, so what comes next isn't a sales pitch — it's the scorecard we'd want checked.

A Sixth Question, Even Though It's Ours to Ask

How often does the number actually refresh? A vendor advertising daily monitoring and one running a monthly scorecard are answering two different questions about cadence, and neither is wrong on its own: daily costs more because it asks the engines more often, and monthly costs less because it doesn't. Our bet, and it is a bet rather than a measurement: most companies buying this line item don't need daily, they need the number to move when the underlying page actually changes, and a monthly cadence proves that just as well for a fraction of the price. We publish ours, so you don't have to guess which cadence you're actually paying for.

What We Answer, Because You'll Ask Us the Same Five

So here is our own scorecard against the five questions above, because we'd rather you hold us to it than take our word for it. Each stored answer names the model wherever one was recorded. A free report runs 4 questions across ChatGPT and Gemini, one pass each; a paid tier runs 20 questions across three engines, three samples each — the sample count scales with what you're actually paying for, so a bigger invoice buys a bigger, checkable number, not a nicer chart.

Each check is labeled grounded or not, per engine, in the methodology section, and any lost question opens to the stored answer itself — the same words the engine wrote, not our summary of them. A failed engine call is excluded from the count entirely and disclosed on its own; we don't fold our downtime into your zero. The readiness side of the product re-checks every 24 hours, and the citation scorecard runs once every 30 days — a cadence we say out loud instead of leaving you to guess at it.

The Offer, Not the Pitch

We priced this list to answer honestly, not to win it, so a reader who runs these five questions against a competitor's page instead of ours still walks away with the useful part of this post for free. Read what a grounded answer actually costs before you commit a budget line to anyone's number, ours included, or start with the free scorecard and let the answers make the case themselves. We look forward to earning the rest of the line item, not just the first question on it, and we mean that as a partner, not a vendor waiting for the invoice to clear. We look forward to growing with you.