How an AI visibility score is calculated

A number is only useful if you can reproduce it. This is the entire Citewise scoring formula — the same one the engine runs — worked through a real example by hand. No weighting you can't see, no editorial judgment. Two rules, one division.

The two things we count

Every AI answer to a prospect question is scored on two mechanical checks, both yes/no:

Mentioned — your firm's name appears anywhere in the answer text. Matching is whitespace- and punctuation-insensitive, so "Bluff Creek" matches "BluffCreek" and "bluffcreek.com." No fuzzy guessing; the normalized string is either present or it isn't.
Cited — your domain appears in the answer as a source (a link or a named reference), not just your firm's name.

These are separate on purpose. Being mentioned means the model recalled you. Being cited means it pointed a reader at you. A firm can be mentioned without being cited (named from memory, no link) — and citation is the stronger signal, because it's what actually sends a prospect to your door.

The formula

score = round( (mentions + citations) ÷ (answers × 2) × 100 )

The denominator is answers × 2 because each answer offers two points of possible visibility: one for a mention, one for a citation. If you were mentioned and cited in every single answer, you'd hit both points every time and score 100. Answers that errored out (an engine that failed to respond) are dropped from both the numerator and the denominator — a failed API call never counts against you.

Worked example: a 12-answer scan

Say we run 3 prospect questions across all 4 engines — 12 answers. Here's a plausible result for a mid-visibility firm:

Question	Engine	Mentioned	Cited
Q1	ChatGPT	✓	✗
	Claude	✗	✗
	Gemini	✓	✓
	Perplexity	✗	✗
Q2	ChatGPT	✓	✗
	Claude	✓	✗
	Gemini	✗	✗
	Perplexity	✓	✓
Q3	ChatGPT	✗	✗
	Claude	✗	✗
	Gemini	✗	✗
	Perplexity	✗	✗

Count the checks: 5 mentions and 2 citations across 12 answers, none errored. Plug in:

score = round( (5 + 2) ÷ (12 × 2) × 100 )
= round( 7 ÷ 24 × 100 )
= round( 29.17 )
= 29 / 100

A 29 lands in the "Invisible" band. Read the table and you can see why the number is fair: the firm shows up for Q1 and Q2 but is completely absent from Q3, and it's cited (not just mentioned) in only two of twelve answers. The score isn't a mood — it's arithmetic you can redo.

Letter grades, and why one check swings them

The percentage maps to a letter grade so it survives a screenshot (nobody remembers "29%"; everyone remembers "F"). The bands:

Grade	Score	Verdict band
A+	90–100	Strong — you win most answers
A	80–89	Strong — you win most answers
B	65–79	Strong
C	50–64	Partial — known, but losing shortlists
D	35–49	Partial — known, but losing shortlists
F	0–34	Invisible

Because the denominator on a small scan is small, individual checks move the grade a lot. In a 12-answer scan, each mention or citation is worth about 4.2 points (100 ÷ 24). So a single competitor's directory listing that displaces you in two answers — costing two mentions — is roughly an 8-point swing, enough to drop a low B to a C. This is the honest reason the paid diagnostic measures 48 answers, not 12: a larger denominator makes each grade more stable and less sensitive to the luck of a single run. The free scan runs just 2 questions, so it doesn't produce a graded 0–100 score — it produces the raw answers themselves. It's still enough to see the obvious case: if the engines already name you in both, there's likely no gap worth $490 to close, and we'll tell you not to buy.

What the score deliberately does not do

It doesn't weight engines. A mention in Perplexity counts the same as one in ChatGPT. We don't claim to know which engine your specific prospects use, so we don't pretend to.
It doesn't reward brand-name questions. Every question is category-level ("best X for Y"), never "is [your firm] good." Asking an engine about you by name tests recall, not whether you make the shortlist. See the methodology.
It doesn't grade sentiment. A mention is a mention. If an answer names you critically, that's a content problem the report flags separately — the score measures presence, and presence is the prerequisite for everything else.
It doesn't guarantee tomorrow's number. AI answers drift with model updates; the score is a measurement window, which is the entire reason re-measuring matters.

See the raw answers — free.

Two of your firm's questions, run across the major engines, with the exact answers emailed to you — free. It's the same measurement the score is built from (just 2 questions, so no graded number yet). If the engines already name you consistently, we'll tell you not to buy.

Get my free scan See a full sample report