How AI search visibility is measured.
The scorecard below is a clearly labelled illustrative example, not a Broadcastwell or client result. It shows the structure of the paid AI Visibility Diagnostic: 35 buyer questions, three scheduled runs per question and engine, 525 scheduled observed answers, named and cited outcomes, competitive context and run-to-run spread.
Per engine
| Engine | Answers | Named example brand | Cited example.com | Calls failed |
|---|---|---|---|---|
| Claude (Anthropic, web search) | 105 | 32 | 61 | 0 |
| ChatGPT (OpenAI, web search) | 105 | 25 | 52 | 0 |
| Perplexity | 105 | 28 | 79 | 0 |
| Google AI Overviews | 105 | 25 | 32 | 0 |
| Google AI Mode | 105 | 26 | 25 | 0 |
| Total | 525 | 136 | 249 | 0 |
This is not unusual. In The 2026 State of GEO, Broadcastwell measured 85 B2B software companies across 61 categories and found that 35% were absent from every scored answer in their category. That finding is a research benchmark: 30 of 85 companies. Interval not reported in the published source. The cited Volume I research is published open access under CC BY 4.0 and is not peer reviewed. The scorecard on this page is illustrative.
Competitive context
The illustrative scorecard compares the example brand with a category leader and peer set.
The 10 published self-audit questions
This set was published under method v1.0 and is not the 35 question client set described on the methodology page.
Illustrative re-score
| Stage | Named | Cited | Leader | Leader count |
|---|---|---|---|---|
| Illustrative baseline | 79 of 525 | 184 of 525 | Category leader | 267 of 525 |
| Illustrative re-score | 136 of 525 | 249 of 525 | Category leader | 280 of 525 |
Method
The same method the AI Visibility Diagnostic uses. 35 buyer questions generated for the category, each fired at five engines with live web search, then scored for whether the brand is named and whether the domain is cited. Brand matching is word-boundary. Domain matching compares hostnames. Failed calls are excluded from the denominator rather than counted as a zero.
The full method is published at github.com/Broadcastwell/audit-method
Limitations
- ·Illustrative values are examples of report structure, not Broadcastwell or client results.
- ·AI answers vary between runs. The paid protocol uses three scheduled runs per question and engine, plus further runs where verdicts disagreed, and reports the spread.
- ·Question wording defines the denominator. Open shortlist, comparison, alternatives and use-case questions are scored separately.
- ·Missing source lists cap a citation score but do not change whether a brand was named.
- ·Being named is not the same as being recommended well. The report records presence, citation and recommendation context separately.
Start with a measured baseline.
See which buyer questions leave you out, the sources visible in those answers, and the three changes worth testing first. The Category Audit covers ten questions; the Diagnostic extends to 35 questions and 525 scheduled observed answers. AI answers vary, so repeated observations and disagreements are reported.
Start with the $490 Category AuditCredits in full against the Diagnostic within 30 days.
Get the Diagnostic, $990Credits in full against the program.