Which products do AI engines actually recommend?

Why you can trust this

• Pre-registered — the prompt set + brand universe are frozen (hashed) before every run; no post-hoc cherry-picking
• Confidence intervals — every share ships with a 95% CI, and per-engine columns expose where the engines disagree
• Conflict-of-interest rule — our own category is not ranked here, and Clear Cited clients are excluded from the public rankings (disclosed, not hidden)
• Open data — every leaderboard downloads as CSV/JSON, CC BY 4.0; products cannot pay to appear or rank

What these numbers mean

Citation rate counts sources named in the answer, not sources retrieved into the model's context. The two are frequently conflated in published figures.

Share of model is a ratio over a pooled mention denominator whose observations are clustered — one answer may name several products, and repeated runs of a prompt are correlated. The paid audit already resamples at the answer level with a percentile bootstrap. The public AI Visibility Index still reports a Wilson interval, and Wilson assumes an independence that a clustered denominator does not have, so the Index's published share intervals are NARROWER than a cluster-corrected estimate would give. Bringing the Index onto the same clustered bootstrap the audit already uses is planned for a future measurement cycle; we have not restated the released figures, because a DOI points at them. We state this because it works against us: wider intervals would make MORE of our published rankings statistically indistinguishable, not fewer. The full method →

What changed in the July 1, 2026 edition

• API platforms has a new #1: Kong overtook AWS API Gateway (vs the June 23, 2026 run).
• Feature flag platforms has a new #1: LaunchDarkly overtook Flagsmith (vs the June 23, 2026 run).
• Vector databases: Redis moved up 2.1 pts share-of-model since the June 23, 2026 run (the two runs’ 95% CIs do not overlap).

Coverage note: the earlier run measured 3 of the five engines; the July 1, 2026 edition measures 5 — part of these movements can reflect the added engine coverage. Each category page shows its measured coverage.

Every change compares this edition against the previous stored measurement of the same category. A leader change is a rank fact; a share move is called out only when the two runs’ 95% confidence intervals don’t overlap. With no earlier snapshot, nothing is claimed.

The Google AI Surfaces Index — measured separately

Google AI Overviews and AI Mode are a distinct search surface, not one of the five assistants in the leaderboards above. Google's AI surfaces are measured separately and never summed with the five engines — a different surface with different mechanics. We measure how often a buyer prompt triggers an AI answer here and which vendors it cites — on their own track, with their own sample size, 95% CI and date, and never add them into the five-engine share-of-model. Why we keep them separate →

Honest-empty by design: a category reads "not yet measured" until its Google AI-surfaces edition is measured (owner-armed via DataForSEO, spend-capped) — never a fabricated number, and never merged into the leaderboard. Join the Index Pulse to hear when new editions land.

Where AI citations come from

Beyond who gets named, we measure where the engines’ citations come from — owned (a brand’s own domain), earned (press, listicles, review sites) and community (Reddit, forums, Q&A). The large majority are off-site, so this is the most actionable map of how to influence what AI cites. It is a separate measurement from share of model and is never summed into it. How we measure →

How AI frames brands

Presence is the headline; quality is the second layer — whether a mention is an endorsement, neutral, cautious or negative, and where in the answer it appears. This is a heuristic read of the answer text, labelled as such, and it never inflates the share of model. Client-specific false-claim flags are private to each report and are never published here. How we measure →