How AI visibility scoring works

"What's our AI visibility score?" sounds like it should have a simple answer, the way a search ranking does. It doesn't, quite — there's no single AI to rank you, only a scattering of assistants giving different answers to different people at different times. Here's how Referenced turns that mess into a number you can actually track.

Start from real questions

Everything starts with prompts: the actual questions your customers might type into ChatGPT or ask Google. Not keywords — full questions, phrased the way a person would phrase them. "Best accountant for a small business in Leeds", not "accountant Leeds".

When you add a site, Referenced suggests a starting set of prompts based on your domain — what you do, who you compete with, the kind of questions someone shopping in your category would ask. You can edit them, delete the ones that don't fit, and add your own. The prompts you track are the whole basis for everything downstream, so it's worth taking a minute to get them right rather than accepting the defaults blind.

Ask every engine, on a schedule

A prompt on its own is just a question. It becomes useful when it's asked repeatedly, over time, across the engines your customers actually use. Referenced re-runs your tracked prompts automatically every day, sending each one to ChatGPT, Google AI Overviews, and Perplexity, and storing every answer it gets back.

The schedule matters as much as the coverage. A single check tells you what one engine said on one day; a scan that repeats tells you whether that's typical or a fluke. Neither you nor anyone on your team has to remember to run it — it happens on its own, and the record builds up whether or not you log in.

Parse the answers

Each answer that comes back is just text, and text on its own isn't a metric. Referenced reads every response and pulls out the pieces that matter: whether your brand is mentioned, whether any competitors are named alongside or instead of you, which URLs the engine cited as its sources, and the sentiment of what's said about you — favourable, neutral, or critical, with the actual sentence as evidence.

That parsing step is what turns a pile of raw AI answers into structured data you can query and trend, rather than a stack of transcripts you'd have to read by hand.

Roll it up

Two numbers come out of all that parsing, and they answer different questions.

Your visibility score is how often you appear at all — the share of your tracked prompts, across engines and scans, where the answer mentions your brand. It's the simplest read of the data: are you in the conversation or not.

Your share of voice is sharper. Across every brand mention Referenced finds — yours and your competitors' — it's the percentage that's yours. Two businesses can both get mentioned constantly and still be in very different positions, if one of them is consistently first and the other is squeezed in as an afterthought. Share of voice is what tells them apart.

Why one answer never matters

Ask the same question twice and you can get two different answers, sometimes minutes apart. Each engine samples fresh results and generates a fresh response, so no single run is definitive proof of anything — not that you're winning, and not that you're losing. Screenshotting one good answer and calling it evidence is the mistake this whole system is built to avoid.

What matters is the trend across scans: is your visibility score climbing or slipping over weeks, not what one lucky or unlucky answer said on a Tuesday. That's why the scans repeat, and why the score is always shown as a line, not a single figure frozen in time.

From score to action

A score on its own tells you where you stand, not what to do about it. That's the job of opportunity tracking: it looks at the gaps in your data — the prompts where competitors are named and you aren't, the questions you're consistently missing — and turns each one into something concrete to act on, rather than leaving you to stare at a chart and guess.