AI Apps Are Taking Over Both App Stores. Their Users Are Getting Kinder — Until the Free Trial Ends

The outside numbers

Generative AI is now the fastest-moving part of both app stores. Sensor Tower's State of Mobile 2026 report counted more than $5 billion of in-app purchase revenue from generative AI apps in 2025, nearly triple the year before, on 3.8 billion downloads (Sensor Tower). Its State of AI 2026 report, published on 16 June 2026, projects 36 billion hours spent in generative AI apps in the first half of 2026, up from 17.2 billion a year earlier, and more than $4 billion of in-app revenue, 36% above the second half of 2025 (Sensor Tower). The same report found ChatGPT's share of the AI-assistant audience dipped below 50% for the first time in March 2026, as Gemini and Claude gained ground.

What sets off a download wave has changed too. Appfigures measured that image-model launches bring 6.5 times more downloads than other model updates; Google's Nano Banana image model alone added more than 22 million Gemini downloads in 28 days (TechCrunch).

36B hoursProjected time in generative AI apps, H1 2026 (Sensor Tower)

Those are counts of downloads and dollars. They say nothing about whether the people downloading are happy — which is the part Appeye measures, review by review, on both stores.

What 1,571 AI apps' reviews say

As of 8 October 2026, 1,571 of the 84,901 apps Appeye tracks carry AI in their store name — "AI", "GPT", "chatbot", or an assistant brand such as ChatGPT, Claude, Perplexity or DeepSeek. From March to August 2026 their share of negative reviews fell every single month, from 24.0% to 21.3%. Over the same months the rest of the catalogue barely moved, from 23.6% to 23.2%.

It happened while AI apps were being reviewed far more. Monthly review volume in the group rose from 24,767 in March to 45,873 in August, an 85% rise, against 64% for every other app. In March, AI apps drew slightly more criticism than the average app; by August they drew 1.9 points less.

+85%Monthly review volume in AI apps, March to August 2026

Not all AI is equal

Split by what the app actually does, the June-August picture is far from uniform. Learning and tutoring apps drew the least criticism: 13.6% negative across 154 apps. Photo, video and art generators drew the most: 27.4% across 381 apps, the largest and most criticised group. Assistants and productivity apps (21.1%, 340 apps), companions and chat characters (22.2%, 183 apps) and health, fitness and wellbeing apps (22.3%, 154 apps) sit in between.

There is an irony in that. Image generation is the feature that brings the downloads, and it is also the kind of AI app users criticise most once they have it.

The assistants people know

Among the best-known assistants, the June-August reviews Appeye analysed — a few hundred per app, so read these as a snapshot rather than a verdict — put Perplexity lowest at 15.5% negative (213 reviews), then DeepSeek at 21.4% (491), Gemini at 21.8% (216), Claude at 22.8% (461) and ChatGPT at 24.5% (462). The companion app Replika stood apart at 35.3% (343). Headline store ratings tell a sunnier story: ChatGPT holds 4.83 on the App Store and 4.77 on Google Play.

Five stars, mostly

The store-wide star split on Google Play points the same way. Across 1,155 AI apps, 77.2% of ratings are five stars and 8.7% are one star; across 60,140 other apps it is 74.8% and 9.2%.

Same app, two verdicts

For the 189 AI apps Appeye tracks on both stores, the App Store rating averages 0.27 stars above Google Play (0.23 for other apps), and 47 of them sit half a star or more apart. A few are not the same app at all in users' eyes: Character AI holds 4.10 on the App Store and 2.15 on Google Play; AI Emojis 4.73 and 1.80; Goodnotes: AI Notes 4.72 and 3.09.

The one complaint AI apps own

Appeye extracts the two-word phrases that recur in each app's negative reviews. Of 121 AI apps with enough negative reviews to have them, 53 — 44% — have "free trial" on the list, 624 mentions in all. Among other apps it is 278 of 4,011, or 7%. "Don't waste" appears for 31 AI apps (26%) against 191 other apps (5%). Next to them sit "cancel subscription", "credit card", "without paying" and "pro version" — and one phrase only an AI app can earn: "wrong answers", 131 mentions across 7 apps.

The pattern is consistent: users are increasingly happy with what AI apps do, and increasingly unhappy with how they are made to pay for it — a trial that converts without warning, a paywall that arrives mid-task.

Users are increasingly happy with what AI apps do, and increasingly unhappy with how they are made to pay for it.

What this means

For app owners, the trial is now the single biggest reputational risk in an AI app, bigger than model quality. For marketers, image features win installs and also the harshest reviews, so expectations set in the store listing matter. For anyone comparing AI apps, a 4.8-star headline rating and the last three months of reviews can tell very different stories.

How we measured

AI apps are the 1,571 apps in Appeye's published catalogue whose App Store or Google Play name contains "AI" (or "A.I.") as a word, "GPT", "chatbot", or one of the assistant brands Copilot, Perplexity, DeepSeek, Grok, Claude, Character.AI, Replika or Midjourney; apps that mention AI only in their description are not included. Negative review share is the share of reviews Appeye analysed in each month that are negative, weighted by review count; June-August figures cover 1,422 AI apps and 126,049 reviews. Star splits and ratings are the stores' own published figures. All Appeye figures were measured on 8 October 2026; external figures are from the sources linked above.