On this page
Your brand appeared in ten recorded answers last week and ten this week. Yet the visibility percentage rose from 50% to 83%.
Before calling that progress, check how many answers were successfully collected. If the second collection has fewer usable responses, the percentage can rise even though the number of appearances stays the same.
That does not automatically mean the calculation is wrong. It means the two percentages may describe different samples. To understand the change, look at completion coverage and compare the same question/provider positions.
Start with the count beneath the percentage
For a response-based visibility metric, the calculation is:
Successful answers mentioning or recommending the target brand
÷ all successful answers in the defined scope
× 100A successful answer contributes once to this appearance count, even if it names the brand several times. A failed request contributes neither a positive appearance nor a successful negative answer.
Prerender Buddy's AI Visibility overview uses successful responses in the selected collection for its Visibility percentage. That makes completion coverage essential context: “10 of 20 successful answers” and “10 of 12 successful answers” describe different amounts of evidence.
This article concerns changes within a monitoring comparison. If the question is why a dashboard disagrees with a manual chat, first compare the exact question and context using the brand-mention checking workflow.
A higher percentage with no change in the comparable answers
Illustrative example only. The providers, counts and outcomes below are fictional teaching data, not customer results or a comparison of named AI services.
A team tracks ten unchanged questions across two provider profiles. Each collection plans twenty responses: ten from Provider A and ten from Provider B.
In the baseline, all twenty requests succeed. In the later collection, eight Provider B requests fail. The twelve positions that succeed in both collections retain the same appearance outcomes.
| Provider profile | Baseline successful / planned | Baseline brand appearances | Later successful / planned | Later brand appearances |
|---|---|---|---|---|
| Provider A | 10 / 10 | 8 | 10 / 10 | 8 |
| Provider B | 10 / 10 | 2 | 2 / 10 | 2 |
| Total | 20 / 20 | 10 | 12 / 20 | 10 |
The full-collection percentages are 10/20 = 50% and 10/12 ≈ 83.3%. A display rounded to whole percentages would show 50% and 83%.
Now compare only the same twelve question/provider positions that succeeded in both collections. In this constructed example, Provider B's two later successful positions are also its two baseline appearances. Both periods therefore have ten appearances across those twelve positions: 83.3% in each.
The apparent increase in the overall percentage comes from which responses are available. The comparable answers show no change in brand appearance.
We still do not know what the eight failed requests would have answered later. They must remain unavailable evidence. Their earlier negative answers do not establish that they would have remained negative.
Equal sample sizes do not guarantee a fair comparison
Two collections can each contain twelve successful responses while covering different questions or providers. Their totals look comparable, but their composition may have changed.
Check the exact question, provider profile, language, market and intended repetition where applicable. If a model or search setting changed, record that too. A question about a broad category should not quietly replace a narrow use-case question in an otherwise “unchanged” baseline.
Branded and non-branded questions also answer different business questions. Adding more questions that explicitly name your product can change the overall appearance rate without showing better discovery among people who have not named it.
Use the prompt-selection guide to define a stable reference set. The coverage check then asks whether you actually obtained usable evidence for that set.
Compare the shared successful positions
For a manual comparison, list the planned question/provider positions and attach each period's result. Keep succeeded, failed, pending and missing records separate.
Identify the positions with successful evidence in both periods. Apply the same brand definition and counting rule to that shared set, then calculate the counts and rates for each period.
Show this comparison alongside the full collection coverage. Do not silently discard the other records or replace the headline with a more flattering subset. State how many positions were excluded and why.
The shared subset answers a narrower question: what changed where comparable evidence exists? It does not remove bias from missing responses or recover the answers you failed to collect. If one provider is largely missing, the useful conclusion may be that the comparison remains incomplete.
This worksheet is a manual reporting method. It does not imply that PB automatically creates a matched-response report.
Investigate coverage before choosing a content action
A change in coverage may call for a collection investigation before a website rewrite. Inspect the recorded failure, the affected provider and the timing. Avoid repeatedly requesting new answers simply to obtain a preferred result.
If a retry is appropriate and available, preserve its actual time and relation to the earlier attempt. A successful retry is a later observation; it is not evidence that the original request succeeded.
Once you have comparable answers, inspect the changes within them. A source difference, a clearer product description or an additional recommendation may support a specific next question. The aggregate percentage alone does not tell you which page to edit.
Explain the change in one paragraph
For the fictional example, a client-facing explanation could read:
“The appearance percentage rose from 50% to approximately 83%, while the appearance count remained ten. Successful coverage fell from 20 of 20 planned responses to 12 of 20 because eight requests failed. Across the twelve positions with successful evidence in both periods, appearance outcomes were unchanged. We will investigate the missing responses before interpreting the overall percentage as an improvement.”
That explanation gives the reader the number, its cause within the observed calculation, and the remaining uncertainty. It does not suggest that a collection problem improved the brand's real-world visibility.
When reporting an absolute rate change, use percentage points: 50% to 83.3% is approximately 33.3 percentage points. Retain the counts and underlying precision if the dashboard rounds its display.
What to check in Prerender Buddy
Open the intended site's AI Visibility results and confirm the selected collection and date. Inspect successful coverage and the underlying provider answers before interpreting a change. If a cell looks like a miss, check whether its run actually succeeded.
Keep the shared-position comparison in your reporting notes, with the original records available for review. PB's recorded answers provide the evidence; your explanation should identify which population each percentage describes.
The client-reporting framework shows where that coverage note belongs in a wider report. Start the next review with “Which answers are we comparing?” before deciding what a higher score means.