5 things that actually determine whether AI recommends your client

"Is my client mentioned or not" is the first question worth asking about AI visibility, and it's the one most agencies stop at. It's also not the only one that matters. Here's a fuller framework for thinking about it — five separate questions, each pulling on different levers. A quick note up front: this is an analytical lens for agencies to apply manually, not a description of a score RankSignal calculates for you. Where that distinction matters, it's called out below.

1. Visibility — are they mentioned at all?

The baseline question. Ask a plain, customer-shaped question about the category and town, and see whether the business gets named. This is binary, easy to check, and the one most agencies haven't started tracking yet — which is exactly why it's worth starting with. This is also the one RankSignal measures directly today: whether a business is named, and how consistently, across the models and prompts you're tracking.

2. Sentiment — how are they described when they are mentioned?

Being named isn't automatically good. A model can mention a business alongside a caveat ("mixed reviews", "can be pricier") that undercuts the recommendation as much as being left out entirely. Worth reading the actual wording, not just counting the mention. RankSignal doesn't score sentiment today — its results are the literal answer text, which you read yourself, the same way you'd read a real customer's message rather than trust a summary of it.

3. Consistency — is it the same answer every time?

AI answers aren't deterministic. Ask the same question three times and you can get three different result sets — a business named once and dropped twice isn't the same situation as one named reliably every time. A single check tells you a moment; repeated checks over time tell you whether that moment was representative. This is closer to what RankSignal's scheduled tracking is for, even though it isn't packaged as a named "consistency score" — it's the visibility trend over repeated runs.

4. Authority — why is the model drawing on this source, not another one?

Models don't recommend businesses at random — they're drawing on patterns learned from citations, review volume, directory presence, and how well-structured a business's own web content is. Two businesses with identical Google rankings can have very different AI outcomes because one has a cleaner, more citation-friendly web presence. This is a real, useful lens for diagnosing *why* a gap exists — but it isn't something RankSignal computes as a number. Treat it as the question to ask once you've found a visibility gap, not a metric this tool reports back.

5. Risk — what's in the answer instead, when your client isn't?

When a business isn't mentioned, something else usually is. Is a direct competitor filling that space consistently? Is the model naming a lead-gen middleman instead of a real local business? Knowing what's *displacing* your client is often more actionable than the absence alone — it tells you who you're actually competing against for that recommendation. Again: a genuinely useful thing to look at, not a number RankSignal calculates for you today — the free scan and the full product both show you the literal answer, competitor names included, so you can make this call yourself.

Using this without over-claiming

None of the above requires a fancy tool to start applying — it's a way of reading AI answers more carefully, the same way an experienced SEO reads a SERP for more than just position. RankSignal's job in this is the part that's genuinely hard to do by hand: running the same checks repeatedly, across multiple models, on a schedule, and keeping an honest record of what changed. What it won't do is hand you a single invented number that pretends to collapse all five of these into one score — that's a harder, more subjective judgement than any one metric should claim to make for you.