Sample workspace · yoursite.com · fictional
What the assistants say when someone asks about you
24 questions · 6 engines · 90 days · every figure ships with its readings
The headline
AI Visibility Score
Visibility
54/ 100
n = 1,320 readings · 708 named the brand · 53.6% rounded to a whole score · 95% CI 50.9%–56.3%
Change over the window
+12.3
Against the first reading of the 90-day window. Part of this is a model change, annotated on the chart.
Sample size
1,320
24 prompts × 6 engines ≈ 55 readings per prompt. Thin engines are shown thin rather than hidden.
Share of voice · yoursite.com
31.4%
Of every brand named across the same readings. This brand leads the set.
Window: 90 days ending 1 Sep 2026. Interval is a 95% Wilson score interval on the pooled proportion — it stays inside 0–100 where the textbook interval would not, which matters for the thinner engines.
The share of readings in which an assistant named this brand unprompted. Nothing is weighted and nothing is smoothed — add up the table below and you get this number back.
By engine
Six engines, six different answers
| Engine | Mode | Readings | Named | Mention rate | 95% CI | Avg. position | Sentiment |
|---|---|---|---|---|---|---|---|
| ChatGPT | GroundedSearches the web before answering | 240 | 150 | 62.5% | 56.2%–68.4% | 2.1 | Positive +0.44 |
| Claude | From memoryAnswers from model knowledge | 240 | 136 | 56.7% | 50.3%–62.8% | 2.5 | Mildly positive +0.34 |
| Perplexity | GroundedSearches the web before answering | 240 | 168 | 70.0% | 63.9%–75.4% | 1.6 | Mildly positive +0.28 |
| Google Gemini | From memoryAnswers from model knowledge | 240 | 117 | 48.8% | 42.5%–55.0% | 3.2 | Mildly positive +0.12 |
| Microsoft Copilot | GroundedSearches the web before answering | 168 | 74 | 44.0% | 36.8%–51.6% | 3.4 | Neutral +0.06 |
| Google AI Overviews | Read from the pageRead from the live results page | 192 | 63 | 32.8% | 26.6%–39.7% | 4.1 | Neutral -0.04 |
| Pooled | All six | 1,320 | 708 | 53.6% | 50.9%–56.3% | — | — |
Average position is the mean rank among the brands named in an answer, over the readings that named this one. Copilot and Google AI Overviews ran fewer readings because their surfaces were unavailable on some days — the count says so rather than the gap being filled in.
An engine that searches before answering is telling you what a buyer sees today. One answering from memory is telling you what it learned months ago. Reporting them as one number is the claim a competitor would attack, so the mode is on every row.
90-day trend
The line, with the model change marked on it
Model change · 5 Aug 2026 · Gemini 3.1
Gemini shipped a retrieval change on this date. The step is in the data, not in your site — the annotation is there so nobody reads it as a win.
A step in the line is usually an engine changing, not a brand improving. When a model ships, the annotation goes on the chart so a jump is never read as a win you earned.