AI Visibility Benchmark Report 2026

Between 29 June and 2 September 2026 we measured our own brand six times. The measurement tool is Rankie.ai; the last four scans ran across ChatGPT, Gemini and Google AI Overview. Raw per-engine mention rates (totals across three deep scans): • Gemini — 1,075 queries, zero mentions (margin ±0.5 pp) • Google AI Overview — 1,052 queries, zero mentions (±0.5 pp) • ChatGPT — 14.3% on 31 August (n=300), 6.5% on 2 September (n=214), margin ±3.4 pp We also mapped 50 demand themes: we lead in 0, contest 11, and 39 are entirely unclaimed.

Each scan runs commercial-intent queries across three AI engines and records, for every response, whether the brand was mentioned, its position within the answer, sentiment, and citation share. The score is a weighted average of four dimensions: mention 40%, position 30%, recommendation 20%, citation 10%. Margin of error: for each engine, a 95% confidence interval computed from that engine's response count. In this series: ChatGPT ±3.4 pp, Gemini ±0.5 pp, Google AI Overview ±0.5 pp. Comparability rule: two scans are comparable only when run with the same engine set and the same scan type. A difference between two scans of different types cannot be read as improvement or decline.

This is the most important section of the report. We claim no trend on ChatGPT. Our two measurements (14.3% and 6.5%) come from scans of different types and are not comparable. We present the difference as neither improvement nor decline. We hold an older, higher Gemini figure; we do not use it, because it was measured with a different engine set. Broad market claims — 'brands doing GEO are N times more visible', 'AI traffic converts at X%' — do not appear in this report, because we cannot verify them with our own measurement. The zero findings, however, are solid: across three scans on two engines, 2,127 queries produced not one mention, and the margin of error on those engines is half a point.

Every measurement figure on this page comes from our own scans and is given above together with the engine, query count and margin of error. For academic findings on the relative effect of GEO techniques, see the Princeton study published at KDD 2024 — 'GEO: Generative Engine Optimization'. Its figures are peer-reviewed and reproducible; they must not be confused with our measurements. A summary with attribution is on our /methodology page. This report contains no figure without a stated source.