A single number went around X this month: Meta's Muse app keeps 73.3% of new users coming back the next day, far above ChatGPT. It is a real figure in circulation, but it is an estimate from a third party, relayed by posts and outlets, and not confirmed by Meta. This post says who published it, what each retention number covers, what is missing, and how to read it next to the rest of the curve. For the wider Muse numbers, see our earlier claimed-vs-verified post on Muse ad spend, downloads and retention.
TL;DR: claim by claim
| Claim | Who says it | Status |
|---|---|---|
| Muse day-1 retention 73.3% | Sensor Tower data, shared on X and repeated by outlets | Third-party estimate, not confirmed by Meta |
| ChatGPT day-1 retention 41.3% | Same data, per a Webull-hosted report | Third-party estimate |
| Muse day-7 retention above 30% | Investor Deedy Das on X, citing Sensor Tower | Matches US figure of 30% reported by 9to5Mac |
| Muse day-14 22.5% vs ChatGPT 26.8% | Same data set | Third-party estimate; one version rounds to 22% |
| Meta statement on retention | None found | Unconfirmed |
Where the 73.3% comes from
The earliest trace we could open is a post on X by Deedy Das, a venture partner, on October 12, 2026: "Meta Muse's retention is out, and it's almost as good as ChatGPT". He wrote that Muse is higher than ChatGPT at day 7 (above 30 percent) and just lower at day 14 (22 percent), and in a reply gave the source as Sensor Tower. The 73.3% day-1 number was then repeated by another X account and by write-ups, among them eWEEK, which reports Sensor Tower tracking new installs across 81 countries and a day-1 return of about 73%.
A Webull-hosted report adds the comparison some posts leave out: ChatGPT at 41.3% on day 1, a gap of about 32 percentage points, flipping by day 14 to Muse 22.5% against ChatGPT 26.8%. That report itself notes a decimal-point mismatch between 22% in the original text and 22.5% in the detailed data.
Note the headline problem: some aggregator pages say Muse "beats ChatGPT" without giving the ChatGPT day-1 number in the text, so the comparison only holds if you find the underlying chart. Always look for both sides of a comparison before repeating it.
What day-1, day-7 and day-14 retention mean
Retention measures what share of a cohort of new installs opens the app again on a given later day. A day-1 number of 73.3% would mean about three in four new users returned within a day. That is high for any app, which is exactly why it deserves scrutiny.
| Metric | What it tells you | Muse (reported) | ChatGPT (reported) |
|---|---|---|---|
| Day 1 | Was the first session good enough to return | 73.3% | 41.3% |
| Day 7 | Is a weekly habit forming | Above 30% | Lower, per the same chart |
| Day 14 | Is it sticking beyond the novelty | 22.5% | 26.8% |
Day-1 numbers are the easiest to inflate. A launch campaign that sends existing Meta users into the app, push notifications from Meta's other apps, or a first session that asks the user to set up something that needs a return visit can all lift the first return. The later numbers, day 14 and beyond, say more about habit.
Why the first figure should be read with care
Four reasons not to treat 73.3% as a clean comparison with ChatGPT:
- Estimates, not Meta's data. Sensor Tower says its estimates "may not accurately represent metrics for any app or advertiser, and should be used for directional trend analysis," as quoted in the coverage we found.
- Advertising. Sensor Tower also says Muse is potentially experiencing outsized gains because of Meta's advertising push. We covered the scale in the ad budget analysis, where Sensor Tower said Muse took 39% of Meta's ad budget.
- Undisclosed cohort. The reports do not publish the cohort, country mix, platforms or retention method. The 81-country figure is global; the TechCrunch and 9to5Mac numbers (30% at day 7 and 23% at day 14, versus an average of 19% for ChatGPT, Claude and Grok at day 7) are US-only. Mixing global and US cuts is a common way retention claims drift.
- Different definitions. "Retention" can mean opened the app, or returned on exactly that day, or any time on or after that day. Without the definition, a 73% and a 41% may not be measuring the same thing.
One more data point: a Morgan Stanley chart was cited by one account at about 23% on day 14, while another account said about 20%. The writer could not confirm which data was used, which shows how quickly figures blur.
What Meta has said
We found no Meta statement confirming the day-1 retention number or the ChatGPT comparison. Meta has published usage statements elsewhere, and outside reporting has put daily users lower than the headline downloads suggest. Our post on the daily-user plateau at 2.5 to 3 million shows how the company-side and third-party pictures can diverge. If Meta gives first-party retention, that should replace everything above.
Read the whole curve, not the first point
Put the numbers together and a clear shape emerges: Muse starts very high, matches ChatGPT around the end of week one, and falls below it by week two. For a consumer AI product that is a recognizable pattern, strong curiosity and a good first session, then a weaker reason to come back daily. It is consistent with the personal-agent pitch behind the product, which depends on recurring tasks, a theme in our Muse Wonder launch coverage and the Muse safety verdict.
It also sits next to other Muse stories that affect trust and return use, such as the paused share links and exposed chats and questions about what agent apps promise on privacy. Retention does not measure any of that, but a product that handles personal data has more to lose from a bad week-two experience.
What this means for people choosing an AI app
If you are deciding where to put your daily assistant habit, retention is a weak guide on its own. High day-1 return tells you the first session was inviting or the install was prompted. It does not tell you the answers are better, the privacy terms are safer, or the agent finishes tasks reliably. Those questions are answered by testing the product on your own work for a week.
For builders, the more useful reading is competitive. ChatGPT's curve is flatter: lower at the start, higher by day 14. That is the signature of a tool people reach for when they have a job to do. Muse's curve is steeper: more people try it and return once, fewer settle in. If the reported figures hold, Meta's challenge is not getting people in the door, which its ad budget and distribution through its other apps handle, but giving them a recurring reason to open it in week two. Features that turn one-off questions into standing tasks, such as inbox triage, reminders and shopping lists, are the lever, and they are the same levers every assistant is now racing to add.
It is also worth remembering what is not in any of these charts: revenue, paid conversion, and time spent per session. A product can retain a smaller share and still be worth more per user. The only way to settle which curve matters is to see Meta's own numbers, which is why the missing company statement is the biggest gap in this story.
How to check a retention claim yourself
- Find the original chart. Ask for the Sensor Tower, Apptopia or company source, not a screenshot of a post.
- Demand the cohort. Install months, countries, iOS or Android, and whether paid installs are included.
- Compare like with like. Same definition, same geography, same install window for both apps.
- Look past day 1. Day 7 and day 14 are harder to game. Day 30 is better still.
- Check for a second source. If two independent data providers agree, confidence goes up; if only one number circulates, label it as an estimate.
- Look for the company's own data. Earnings calls and filings often include first-party usage that overrides panel estimates.
If you build a consumer AI product, the same lesson applies to your own dashboards: track a cohort curve, not a single day.
Verdict
- Verified: the figure exists and is attributed to Sensor Tower data; day 14 numbers show ChatGPT slightly ahead.
- Not verified: that 73.3% is a like-for-like win over ChatGPT, that it is representative outside the sampled markets, and that it is free of advertising effects.
- Missing: a Meta statement, cohort details, and a method note.
We will update this post if Meta or Sensor Tower publish more.
Related reading
- Muse ad budget, downloads and retention: claimed vs verified
- Muse daily users plateau
- Muse Wonder launch, pricing and benchmarks
- Is Meta Muse safe?
- Muse paused share links
- AI agent privacy promises
- Sources: Deedy Das on X, eWEEK, 9to5Mac on Sensor Tower
Figures are third-party estimates as reported through October 27, 2026.