An industry selling "visibility" owes buyers exact words. This is our measurement vocabulary, published so that any report we deliver can be checked against it — and so that when another vendor shows you a "visibility score," you can ask what's actually in it.

The unit: an observation

One prompt, run once, on one engine, at one recorded time. That precision matters because engines are non-deterministic — the same prompt produces different answers across runs. Any metric built on single runs inherits that noise. Ours are built on three runs per prompt per engine, which is why a Citation Ledger percentage comes with a range, not false precision.

The four metrics

• Citation share — observations where your domain is linked as a source. The strictest and rarest signal: only a small fraction of AI answers carry visible links at all. [1]
• Mention share — observations where your brand is named, linked or not. Most of your real exposure lives here.
• Mention sentiment — how you're characterized when named: recommended, neutral, or caveated. Being consistently named "the expensive option" is a finding, not a win.
• Consensus set — the three-to-five brands engines converge on when answering your category's prompts. This is the competitive frame: your movement is measured against theirs.

What good looks like right now

Benchmarks in this field are young, so hold them loosely — but the working numbers: most B2B brands appear in fewer than 10% of the buyer-intent prompts relevant to them, and above 20% presence indicates strong AI visibility. [2] The gap between those numbers is the opportunity, and it's still wide open because only about 14% of marketers measure any of this. [2]

Three ways vendors inflate these numbers

Questions to ask anyone showing you an AI visibility metric, including us:

• Blending. Merging linked citations and unlinked mentions into one number makes it bigger and less meaningful. Ask for them separately.
• Prompt gerrymandering. A prompt set full of branded queries ("tell me about Acme") produces spectacular presence. Only non-branded buyer-intent prompts measure discovery.
• Single-run snapshots. One lucky run is a screenshot, not a rate. Ask how many runs are behind the percentage.

Why we publish this

A metric definition you can hold a vendor to is worth more than a case study you can't audit. Every Citation Ledger report ships with the raw observation log — timestamp, engine, prompt, answer, brands named — so the percentages above are recomputable by anyone on your team. If a number can't survive that, it shouldn't be in a report.