How the AI Visibility Score Works — And Why It Matters

Every publisher site in the AuthoritySource.ai marketplace can show an AI visibility figure. It answers one question, and only one: is this site actually being named as a source inside AI assistant answers?

Not "could it be." Not "does it look authoritative." Named, in real answers, right now.

Where the data comes from

We query DataForSEO's LLM Mentions index for each publisher domain. That index is built from recorded answers produced by ChatGPT, Gemini and Google's AI Overviews — the actual output users see, captured at scale.

Two things this is not:

• It is not a crawl of the publisher's site. Nothing about the page HTML, schema markup, or word count feeds this number.
• It is not a prediction. If a site scores above zero, real answers cited it.

When the index has no coverage for a domain — usually a small or very new site — we fall back to live prompt sampling: we generate questions from the site's own topics, run them against live models, and count how often the domain surfaces. That fallback is labelled differently in the UI, because a zero from sampling means "our prompts didn't surface it," while a zero from the index means "no recorded mentions anywhere."

The two inputs

The score combines exactly two measurements.

1. Recorded mentions. How many real AI answers name this site as a source. Raw volume.

2. Share of voice. In those same answers, how often this site is named compared with the single most-cited other domain that appears alongside it. It's a head-to-head ratio, so 50% means "tied with the leader," above 50% means the site is the most-cited source in its own conversation, and 10% means the leader is named nine times for every one mention of this site.

Share of voice is the part buyers usually underrate. A site with modest volume that dominates its niche conversation is often a better placement than a bigger site that's a footnote in someone else's.

The formula

score = 100 x (0.6 x volume + 0.4 x share of voice)

volume = log10(1 + mentions) / log10(1001)   (capped at 1)

Volume is log-scaled on purpose. Going from 1 to 10 recorded mentions is a real change in status; going from 900 to 1,000 is noise. A linear count would make a handful of huge publishers look infinitely better than every specialist site on the platform, which isn't how citations actually work.

The 60/40 split reflects what the two signals tell you. Volume says the site clears the bar at all. Share of voice says whether it's the source models prefer once it's in the room.

The displayed figure is a rolling average of the last few measurements, not a single snapshot, so one unusual crawl doesn't swing the number.

Worked example

A site with 103 recorded mentions and 56% share of voice:

• volume = log10(104) / log10(1001) ≈ 0.73
• score = 100 x (0.6 x 0.73 + 0.4 x 0.56) ≈ 63

That lands in the Cited band. In plain language: this site is named in real AI answers regularly, and within those answers it is cited slightly more often than the strongest competing source.

The bands

The gap between "—" and "Not yet cited" matters. One is an absence of data; the other is data showing an absence.

Why any of this matters

Search is splitting in two. Ten blue links still exist, but a growing share of queries are answered before the user ever clicks — and in that answer, a small set of sources get named.

That changes what a placement is worth:

• AI assistants are a referral channel. Sites are seeing real, measurable sessions arriving from AI assistants, and that traffic is pre-qualified — the user already read a summary and chose to go deeper.
• Citations compound. Models associate brands with topics based on where those brands appear. A mention inside a source the models already trust is more likely to be pulled into future answers than the same mention on a site they never quote.
• It's a signal backlink tools can't see. Domain Rank measures the link graph. AI visibility measures the answer layer. A site can be strong in one and invisible in the other, and only one of those is where your buyers are increasingly asking questions.

This is exactly why brand mentions — editorial references with or without a link — are a distinct product on AuthoritySource.ai. For generative engines, the sentence around your brand is the asset.

How to read it honestly

Three caveats we'd rather state than bury:

1. It's topic-scoped. A site can be well cited for its own subject and absent everywhere else. High AI visibility only helps you if the site's topics overlap yours — which is what the match score is for.
2. It moves. Models retrain, indexes refresh, and competitors publish. Treat the score as a current reading, not a permanent rating.
3. It's one input. Read it next to Site Score, Domain Rank, organic traffic and match score. A well-cited site with a weak topical fit is still the wrong placement.

Used that way, AI visibility answers a question no other metric on the page can: when someone asks an AI assistant about this topic, does this site get to be part of the answer?