How to compare AI visibility over time
To compare AI visibility over time, start with the same questions, model IDs, and recorded search settings. Then compare completed answer analysis within that shared scope. A missing response gives you no evidence that a brand disappeared.
Match the measurement conditions before reading the trend
Braign's recurring reports retain earlier runs. The comparison uses unique question-and-model pairs with equal recorded search settings. Missing responses, excluded answers, unfinished analysis, and ambiguous pairs do not become visibility losses.
This matters when a report was intended to collect 60 answers but only 54 are ready to compare. Read the completion notices and the number of matched pairs. Legacy results without recorded job settings carry a qualification because those conditions cannot be fully confirmed.
Current classification corrections apply to both periods. If you correct a brand variant or source ownership, the comparison uses that corrected context for the earlier run too. Record the correction when it changes how a stakeholder would interpret previous reporting.
Work through a comparison with missing answers
Imagine two runs planned for 20 questions across three models. The first has 60 completed answers. The second has 54 completed answers, and 50 pairs meet all the matching conditions. The numbers below are illustrative.
| Result within the 50 matched pairs | Answer count |
|---|---|
| Brand present in the earlier run | 20 |
| Brand present in the later run | 25 |
| Newly present in the later run | 8 |
| No longer present in the later run | 3 |
The brand's presence in the matched set rose from 40% to 50%, a gain of 10 percentage points. Eight gains and three losses give a net increase of five answers. Inspect both groups; they can involve different buyer questions.
The other pairs remain outside this comparison. Treating them as brand absences would confuse missing evidence with changed visibility.
Braign hides headline metric deltas when the complete answer sets differ. The change brief can still compare their valid intersection. Read the scope attached to each view before combining its figures with an overall result.
Use percentage points when reporting a rate change
A move from 40% to 50% is a gain of 10 percentage points. It is also a 25% relative increase because 10 divided by the original 40 equals 0.25. Writing “up 10%” leaves the reader to guess which calculation you mean.
Report the two rates, their counts, and the point difference: “Presence increased from 20 of 50 to 25 of 50 matched answers, or 40% to 50% (+10 percentage points).”
For share of voice, keep the included brand set and counting rule consistent. A rival receiving fewer mention occurrences can increase your share even if your own mention count stays flat. The metric comparison explains why presence is useful alongside share.
Inspect changed answers before explaining why they moved
Open questions where the brand was gained or lost. Compare the recommendation language and sources in both answers. A model may still mention your brand but now recommend another product first; that change can matter even when presence stays flat.
AI answers can vary between runs. Model behavior, retrieval, and stochastic sampling can change without any action by your team. A repeated pattern across several comparable runs gives you more context, but it still needs examination at the answer level.
Marketing activity is an annotation. Logging a page update on September 1 and observing a new citation on September 15 does not establish causation. Describe the sequence accurately and use the cited evidence to decide what to investigate next.
Choose a cadence your team can use
Braign offers one-time reports and monthly or quarterly recurring reports. Choose the interval around the decisions your team makes and the amount of source review it can complete.
A recurring handoff should identify the comparable scope, show the changes worth reviewing, and attach the underlying answers. Include the content work completed since the previous run as context. Keep any claim about business outcomes tied to separate evidence from your own analytics or sales records.
Use AI visibility monitoring to review the recurring-report workflow, and the citation guide to plan page reviews between runs.
See the evidence in a report
Explore the public sample, or choose your brand, buyer questions, and models to price your own report.