The August 2026 ranking
ChatGPT and Google Gemini share first place. We compared eleven general-purpose AI assistants: ChatGPT, Google Gemini, Claude, Grok, Meta AI, Qwen, Perplexity, Microsoft Copilot, DeepSeek, Mistral's Vibe (formerly Le Chat) and Kimi. Six of them met the eligibility rules of the Net App Satisfaction Score (NASS) on the iOS App Store worldwide in August 2026. NASS is the share of 5-star ratings minus the share of 1-to-3-star ratings among those posted on the App Store in the month, on a scale from −100 to +100, calculated half from ratings left without a review and half from written reviews. This compares how users rate the iOS apps, not the performance of the AI models behind them.
| # | App | NASS | Company |
|---|---|---|---|
| 1 | ChatGPT | +69.8 | OpenAI |
| 1 | Google Gemini | +69.5 | |
| 3 | Qwen (千问) | +68.1 | Shanghai Zhixin Puhui Technology |
| 3 | Meta AI | +66.5 | Meta |
| 5 | Grok | +55.1 | X Corp. |
| 6 | Claude | +41.2 | Anthropic |
Scores are per app, not per company: Gemini's score is Gemini's alone, not Google's. Each is precise to ±2 points or better at 90% confidence. When the gap between two apps is smaller than their combined margin of error, the square root of the sum of their squared margins, the data cannot separate them and they share a rank: ChatGPT and Gemini (0.3 points apart, combined margin 0.5), then Qwen and Meta AI (1.6 points apart, combined margin 2.0). Gemini is ahead of Qwen by 1.4 points, more than their combined margin of 1.2. Combined margins are calculated from the unrounded margins, so they can differ by 0.1 from a calculation on the rounded values shown. Each score reflects the app's own users: Qwen, Alibaba's assistant, is rated mostly in China, so its score compares less directly with apps rated mainly elsewhere.
Disclosure: mwm.ai is operated by MWM, which publishes none of the apps in this comparison.
ChatGPT and Gemini: too close to separate, with a caveat
From September 2025 to July 2026, ChatGPT led Gemini in ten months out of eleven, by 2 to 19 points. The exception was March 2026, the month of the QuitGPT backlash. Gemini stood at +49 in July 2026 and reached +69.5 in August. The whole move came in the second half of the month: Gemini's weekly 5-star reviews went from about 1,100 in July to about 6,000 in the weeks of 16 and 23 August, in all its largest markets at the same time. Most of these reviews have one- or two-word titles ("Good", "Rating", "App review").
That pattern is typical of an in-app prompt asking users to rate the app. We cannot confirm it from outside: Google has not said so, and the reviews themselves do not say how they were requested. What the data does show is where the gain sits. Gemini's written reviews rose from +23 in July to +62 in August; its ratings without a review, the larger channel, barely moved (+75 to +77). September will tell whether the new level holds.
ChatGPT's March: the QuitGPT backlash in the numbers
ChatGPT's lowest point of the year was March 2026, at +46.5, down from +64.7 in February. On 28 February OpenAI announced that the US Department of Defense could use its models in classified operations, and users began deleting the app and cancelling subscriptions under a "QuitGPT" banner (Tom's Guide, Yahoo Finance). In the United States the drop was sharper, from +63 to +40.
The recovery was fast: +61.1 in April, then a new high of +69.8 in August. In March, Claude received 79,000 ratings without a review, more than three times February's 23,000, and its written reviews went from about 1,100 to 3,900. March is the first month in which Claude's score was precise enough to be ranked at all.
Twelve months of NASS
Worldwide, iOS App Store. A dot marks a month in which the app did not meet the eligibility rules.
| App | Sep 25 | Oct | Nov | Dec | Jan 26 | Feb | Mar | Apr | May | Jun | Jul | Aug | 12-month change |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ChatGPT | 58 | 54 | 61 | 64 | 66 | 65 | 46 | 61 | 65 | 68 | 68 | 70 | +11 |
| Google Gemini | 48 | 52 | 54 | 53 | 57 | 57 | 59 | 57 | 49 | 50 | 49 | 70 | +21 |
| Grok | 75 | 68 | 74 | 64 | 65 | 72 | 51 | 52 | 45 | 56 | 43 | 55 | −20 |
| Qwen (千问) | · | · | · | · | · | −38 | · | 39 | 75 | 75 | 72 | 68 | n/a |
| Meta AI | · | · | · | · | · | · | · | · | 57 | · | 58 | 66 | n/a |
| Claude | · | · | · | · | · | · | 46 | 36 | 43 | 30 | 38 | 41 | n/a |
Grok: the only major assistant down over the year
Grok started the period as the best-rated assistant, at +75, and ended it at +55. The trend is downward, with large swings: +72 in February 2026, +45 in May, +56 in June, +43 in July. It comes almost entirely from written reviews, which fell from +63 to +26; Grok users who rate without writing remain the most satisfied of the group (+83.8 in August, against +87.6 a year earlier).
Claude: satisfied raters, divided reviewers
Claude has the widest gap between its two channels. Its ratings without a review (+81.7) match ChatGPT's (+81.9); its written reviews are evenly split, at +0.7. Half of the NASS comes from each channel, so Claude ranks last. The written channel is where users with a problem go first, which makes it the one to watch for Claude. Its overall NASS has moved between +30 and +46 since March, without a clear trend.
Not ranked this month
Perplexity (+47.0, ±3.3), the two Microsoft Copilot apps, DeepSeek, Mistral's Vibe and Kimi did not have enough volume in August 2026 to give a score precise to ±2 points. DeepSeek received fewer than 2,000 ratings without a review on iOS worldwide that month. Their scores are calculated every month and will enter the comparison once they meet the rules.
Detailed scores
August 2026, worldwide. Ratings score: ratings left without a review. Reviews score: written reviews. Each is the share of 5-star ratings minus the share of 1-to-3-star ratings in that channel, and NASS is their average. Ratings (n) and Reviews (n): counts received during the month.
| App | NASS | ±90% | Ratings score | Reviews score | Ratings (n) | Reviews (n) |
|---|---|---|---|---|---|---|
| ChatGPT | +69.8 | ±0.3 | +81.9 | +57.6 | 2,990,373 | 41,393 |
| Google Gemini | +69.5 | ±0.4 | +76.7 | +62.3 | 671,646 | 24,300 |
| Qwen (千问) | +68.1 | ±1.2 | +67.8 | +68.4 | 28,612 | 2,790 |
| Meta AI | +66.5 | ±1.6 | +79.9 | +53.0 | 131,127 | 1,759 |
| Grok | +55.1 | ±1.3 | +83.8 | +26.3 | 117,812 | 3,743 |
| Claude | +41.2 | ±1.8 | +81.7 | +0.7 | 100,498 | 1,998 |
The margin of error covers statistical precision only; it does not cover who chooses to rate, rating requests made by apps or reviews missing from the collection.
How NASS is calculated
For each app and each month, NASS takes the ratings received without a review and the written reviews separately. In each channel, the score is the share of 5-star ratings minus the share of 1-to-3-star ratings, 4 stars being neutral. NASS is the average of the two channels, and every score comes with its 90% margin of error. The full method, the eligibility rules and the limits are on the NASS methodology page.
Source: rating counts published on each App Store listing and written reviews, recorded every day by MWM Intelligence. iOS App Store, worldwide, September 2025 to August 2026.