What Is a Good AI Visibility Score? Grade Bands Explained

On our 0–100 scale, 75+ is Strong, 50–74 is Visible, 25–49 is At risk, and below 25 is Invisible. Most local businesses start in the bottom two bands, so a low first score is normal. The number that matters is not your first score — it’s the direction of the score across monthly re-audits.

What the score actually measures

An AI visibility score answers one question: when real customers ask AI systems the questions that lead to businesses like yours, how often — and how well — do you show up?

Ours is built from a fixed measurement, not an estimate. Every audit and every monthly re-audit runs the same battery: 20 localized customer questions — four intents (discovery, recommendation, comparison, task-specific) — against six AI systems: GPT-, Claude-, Gemini-, DeepSeek-, and Qwen-class models plus Google’s answer layer. That produces 100+ timestamped answers per audit. Each response is stored raw. Reports quote answers verbatim; we never edit, paraphrase, or fabricate a model output.

The score is the roll-up of those answers. It is not a prediction, a proprietary black box, or a survey. It’s a count of what the systems actually said, weighted by what matters.

The four components, and why mention rate dominates

Component	Weight	What it measures
Mention rate	50%	How often you’re named at all, across the battery
Prominence	20%	Where you appear — first business named, or an afterthought
Sentiment / accuracy	15%	Whether what’s said about you is positive and factually right
Source presence	15%	Whether you exist on the domains the answers cite

Mention rate carries half the score because it’s the threshold everything else depends on. A business that isn’t named has no prominence to improve, no sentiment to manage, no position to defend. Being in the answer at all is the difference between existing and not existing in this channel — so it gets the most weight.

Prominence matters because composed answers have an order, and readers weight the first name the way they weighted the first result. Sentiment and accuracy matter because an AI answer that names you but gets your services wrong, or summarizes your reviews unkindly, can be worse than silence. Source presence is the leading indicator: it measures whether you exist on the domains the answers cite, which in our audits is the pattern that predicts future mentions best. Every audit includes a citation-source inventory — the exact domains the AI answers drew on in your market — so this component comes with its own to-do list.

The grade bands

• 75+ — Strong. You’re named in most runs, usually early in the answer, described accurately, and present on the cited sources. The job is defense: keep reviews flowing, keep data consistent, watch for slippage.
• 50–74 — Visible. You show up regularly but not reliably — named in some phrasings and systems, absent in others. Typically a few specific gaps: one weak system, missing presence on two or three cited sources.
• 25–49 — At risk. You appear occasionally, often late in answers or in some systems only. Competitors are being named where you aren’t. The gaps are usually identifiable and fixable.
• Below 25 — Invisible. AI systems rarely or never name you for the questions that should find you. Common for otherwise healthy businesses; it means the sources being retrieved don’t know you exist.

The bands describe today’s answers, not your business. A 20 is not an insult — it’s a measurement of a channel you probably never optimized for.

Why most local businesses start low

Because absence is the default state. Three reasons:

First, AI answers are composed from a handful of retrieved sources, and those sources skew away from business websites. In one consumer niche our weekly tracking lab follows — 90 answers a week — the most-cited domains were YouTube, Reddit, and major review and news sites, not the businesses’ own sites. If your presence lives mostly on your own website, the systems composing answers may never encounter you.

Second, most local businesses have inconsistent or incomplete data on the sources that do get retrieved: a half-filled Google Business Profile, a phone number that differs between directories, review profiles never claimed. Retrieval systems corroborate across sources, and inconsistency reads as uncertainty.

Third, nobody was optimizing for this channel until recently, so the starting line is low across the board. That cuts both ways: your competitors likely score low too, which means specific, boring fixes — the kind listed in our answer engine optimization guide — move you past them faster than they would in mature channels.

The channel itself is worth the effort: 37% of consumers now start searches with AI instead of Google. And honestly, about 84% of US consumers still use Google daily for local searches — but Google’s own answers are now AI-composed, so both paths lead through the same measurement.

Why the trend beats the absolute number

A single score is a snapshot of a probabilistic system. AI answers vary between runs — same question, same day, different output. Models update. Competitors act. A snapshot can flatter or insult you by luck.

The trend is where the signal lives. A business moving 18 → 27 → 41 over three monthly audits is winning, even though 41 is still “At risk.” A business sitting at 62 for six months while a competitor climbs is losing, even though “Visible” sounds fine. Direction and slope tell you whether the work is landing; the absolute number mostly tells you where you started.

This is also why we resist score-shopping between tools. Different question batteries and different weights produce different numbers. Any score is only comparable to itself, measured the same way, over time.

Sampling honesty

A score is only as trustworthy as its sampling, so here is ours, plainly.

Same battery every time: the monthly re-audit runs the identical 20 questions against the identical systems, so month-over-month movement reflects the world changing, not the questionnaire changing. Timestamped raw storage: every answer is stored as received, with a UTC timestamp, and reports quote verbatim. Known limits: 20 questions × 6 systems is a sample, not a census. It won’t capture every phrasing every customer might use. What it will do is detect real change reliably — and it can’t be gamed by cherry-picking a lucky run, because the battery doesn’t move. The method is documented end to end on how it works.

And the caveat we attach to everything: source updates take days to weeks to propagate into AI answers. We measure monthly, and no one can guarantee a score, a mention, or a timeline. What improves the odds is fixing the inputs AI systems observably rely on — then measuring again.

What to do this week

1. Get a baseline. Run a scan or do it manually: ask 5 customer questions in 2–3 AI systems, log who’s named. If you haven’t checked, start with whether your business shows up in ChatGPT.
2. Read the citations, not just the answers. List the domains cited; mark which ones you’re missing from.
3. Fix the mention-rate inputs first — complete Google Business Profile, consistent name/address/phone, claimed review profiles. Half the score lives here.
4. Note prominence and accuracy. If you’re named, is anything wrong or stale in the description? Fix it at the source.
5. Schedule the re-measurement. Same questions, same systems, one month out. Write the date down.
6. Judge in three months, not three days. One cycle for propagation, two more for trend.

If you want your baseline score and the citation-source list for your market without doing the sampling by hand, run the free AI visibility scan.