Observatory of brand visibility in AI
How often does a brand actually appear when AIs answer its customers? Real, dated figures aggregated from our platform. Reusable with attribution.
Scope of this edition : 78 audited brands · 129 complete audits. Updated : July 2026.
The key figures
Model reliability: invented-URL rate
Share of URLs cited as sources that do not exist, per model. Of 26 435 verified URLs, 70 were invented (0.26 % on average).
| Model | Invented-URL rate |
|---|---|
| Perplexity Sonar Pro | 1.05 % |
| Anthropic Claude Sonnet 4.6 | 0.62 % |
| OpenAI GPT-5 Mini | 0.44 % |
| Anthropic Claude Haiku 4.5 | 0.19 % |
| OpenAI GPT-5 | 0.16 % |
| Google Gemini 2.5 Flash | 0.12 % |
| Perplexity Sonar | 0.04 % |
Trends by sector
Indicative sample — per-sector counts are still small (n shown). To be consolidated in future editions.
| Sector | Brands (n) | Visibility (/100) |
|---|---|---|
| Retail | 4 | 78 |
| Services | 4 | 77 |
| Construction | 3 | 75 |
| IT / Tech | 7 | 69 |
| Restaurants | 3 | 55 |
| Health | 3 | 51 |
Method & limits
These figures aggregate real audits of brands that chose to measure themselves — a sample not representative of all companies. Spontaneous visibility comes from an entity-extraction sub-sample. Sector trends rest on small counts and are indicative. Full methodology, confidence intervals included, on our methodology page.
Cite the Observatory
This data is freely reusable with attribution to “AI Labs Audit — AI Visibility Observatory”. Two in-depth publications complete this edition: the AI visibility barometer, the study on URLs invented by models.