Every week someone publishes a new AI leaderboard, and every week it measures the same thing: how models perform on tests. Nobody measures how they perform on Tuesdays — in the hands of a student, a founder, a retiree, someone in a parked car asking about a rash. So we built that leaderboard. We exported the 10,000 most recent App Store and Google Play reviews for ChatGPT, Claude and Gemini — 60,000 reviews — and scored ten dimensions the way users score them: with stars and profanity.
How the scoring works
For each dimension we take every English review that mentions it — same keywords for all three apps, no cherry-picking — and compute the average star rating of those reviews. It reads like a review score: above 4 means people mostly praise it, near 3 means contested, below 2.5 means an open wound. Sample sizes and negative shares shown on every chart; full methodology at the bottom.
Ease of use
Winner: GeminiPhoto finish. All three clear 4.5★ — the last thing anyone complains about is the chat box. Gemini takes it by a nose, and it is Gemini’s only round of the night.
Speed
Winner: ChatGPTChatGPT and Gemini tie in practice; Claude falls almost a full star behind — 38% of its speed mentions are 1–2★, mostly describing waits and mid-response stalls.
Writing
Winner: ChatGPTChatGPT wins raw. But adjust for fanbase mood — compare each app’s writing reviews to its own average — and Claude is the only app whose writing reviews score meaningfully ABOVE its baseline (+0.32★). Writers who find Claude, stay.
Coding
Winner: ChatGPTThe upset round. Raw score says ChatGPT. But Claude gets MORE coding mentions than ChatGPT (1,400 vs 1,248) from the same-size sample — and four times Gemini’s. Claude’s problem isn’t coder love, it’s that its coders hit usage limits and rage-rate the app. Gemini, meanwhile, posts the worst coding sentiment of the three.
Accuracy
Winner: ChatGPTNobody wins this round; ChatGPT just loses it most slowly. Every app scores below 3.5 when users discuss correctness, and ChatGPT generates the most accuracy complaints in absolute terms — 1,332 reviews.
Image generation
Winner: ChatGPTChatGPT wins comfortably. Gemini — the app from the company that owns half the world’s images — posts a 2.99, with 46% of its image mentions negative, many of them "it says it can, then it can’t."
Voice mode
Winner: ChatGPTChatGPT’s clearest feature win — people use voice while driving, cooking, walking. Claude’s voice reviews run 48% negative, and its users say the quiet part out loud: "Chatgpt always works better."
Reliability
Winner: ChatGPTChatGPT wins by being merely mediocre. Claude and Gemini’s stability reviews are roughly two-thirds negative — crashes, freezes, vanishing chats.
Subscription value
Winner: ChatGPTEveryone is underwater the moment money comes up. ChatGPT is least-hated at 3.10; Claude’s billing reviews run 68% negative — overwhelmingly people who’d happily pay MORE if the limits would just go away.
Customer support
Winner: ChatGPTThe round where the sport itself loses. ChatGPT "wins" at 2.73. Claude’s support reviews average 1.61. Gemini’s average 1.00 — literally every review that mentions Google’s customer support is negative. The AI industry’s true shared benchmark.
Final scoreboard
About that zero. Claude sweeping zero rounds is the most misleading true fact in this article. Its baseline average (3.39 among English reviews) is dragged down by one thing users scream about — usage limits — and that anger bleeds into every topic its heaviest users touch. Adjust each score against the app’s own baseline and the picture flips: Claude’s writing reviews run +0.32 above its average (the only positive writing delta on the card), its coding mention count beats ChatGPT’s outright, 1,400 to 1,248, from the same-size sample. Claude doesn’t have a quality problem. It has a quantity problem — and its users’ one-star reviews are, functionally, demand letters.
And a word on Round 10. The one benchmark all three labs ace is ignoring their customers. ChatGPT "won" customer support with a 2.73 — a score that would kill a food-delivery app. Across 60,000 reviews, OpenAI replied to 0.0% of them, Anthropic to 0.1%. Google, to its credit, answers 13.8% on iOS — and still scored a perfect 1.00 from every user who mentioned its support. The most advanced listening machines ever built, and nobody’s listening back.
Run this benchmark on any app
Every number here came from a CSV. Rivioo exports any app’s reviews, yours or a competitor’s, with no signup. Pay once per export, from $1 for up to 100 reviews to $25 for up to 10,000.
Export an app’s reviews →Methodology & honest caveats
10,000 newest reviews per app per store, exported via Rivioo July 16–17, 2026; 60,000 unique after dedup; scores computed over 32,594 English reviews. "Newest 10k" spans different time windows per app (Gemini’s Android sample covers ~3 days; Claude’s ~7 weeks) because review velocity differs ~17x. Keyword matching undercounts paraphrase — treat within-app percentages as floors and cross-app comparisons (identical keywords, identical method) as the reliable signal. Small samples flagged on charts; Gemini’s support n=12 means "1.00" is directional, not precise. Reproduce it: export the same three apps and recompute — we’d love someone to check us.