Seven AI visibility reporting mistakes that cost agencies credibility

1. A share with no sample size

"Cited 34% of the time" is not a measurement until you say out of how many. Nine of ten and ninety of a hundred are both 90% and only one supports a decision.

Fix: every number carries n and an interval, in the same sentence.

2. A composite visibility score

Vendors love them because they always move, which makes the dashboard look alive. Clients tolerate them until a data-literate person asks how the weighting works, and then the whole report is suspect.

Fix: report citation share per engine. Boring, auditable, defensible.

3. Averaging across markets

A global figure blends a 60% share in the US with a 4% share in Germany into a number that describes neither.

Fix: market is a reporting dimension, not something to collapse.

4. Comparing numbers from different tools

Different prompt sets, different sample sizes, different parsers, different collection dates. The comparison is noise dressed as a benchmark.

Fix: competitor figures must come from the same runs as the client's.

5. Counting links and calling it presence

Brands are named in answer text roughly 2.7 times more often than they're linked. A link-only metric misses most of the visibility and, worse, cannot distinguish "absent" from "described but unlinked" — two problems with opposite remedies.

Fix: report citation share and mention share as separate lines.

6. Silent parser changes inside a trend line

When extraction logic changes, measured values can shift without anything changing in the world. Presenting that as a trend is an unforced error.

Fix: annotate parser versions on the chart, the way you'd annotate a tracking change in analytics.

7. The screenshot

Covered at length elsewhere: one non-deterministic observation, presented as steady state, disproved by anyone with a phone.

Fix: rates and intervals, with the raw answers available on request.