We made our AI forget everything every 30 days. Users started sharing more.
When we first built Murror, we stored every journal entry forever. More data meant better emotional pattern detection, smarter insights, richer reflections.
Our AI kept getting better. Our users kept getting quieter.
We dug into the data. Users who'd been on the app for 3+ months were writing shorter entries. Less personal. More surface-level. "Had a good day." "Feeling okay."
We ran user interviews. The pattern was clear: the longer people used Murror, the more aware they became that they were building a permanent record of their most vulnerable moments. One user said, "I want to be honest, but I also know this is all being saved somewhere."
So we tried something radical: a 30-day memory window. After 30 days, raw journal entries get summarized into anonymous emotional patterns, then the original text is deleted. Users can pin specific entries to keep them — but the default is forgetting.
The team thought I was crazy. "You're throwing away our best training data."
What happened:
Average entry length increased 3x
- Users mentioned specific people by name 5x more often
- 30-day retention went from 23% to 41%
The lesson: in emotional technology, the right to be forgotten isn't a privacy feature. It's a trust feature. People will only be truly honest with technology that promises not to remember everything.
Has anyone else experienced this tension between data retention and user trust?
Replies
the pin feature is the part I'd want to understand better - if a user can pin an entry to keep it forever, doesn't that just recreate the same "this is a permanent record" anxiety for whatever they choose to pin? seems like the honesty boost you're seeing comes specifically from knowing the default is deletion, so I'd guess people pin way less than they'd expect to going in. curious if that held true, or if pinning became its own habit once people got comfortable with the 30 day window
@omri_ben_shoham1 I like the idea of having the option, but I think I'd use it sparingly. Once something becomes permanent, I'd probably write with a different mindset, which seems to work against the honesty the temporary model encourages. I'm curious whether users treated pinning as an exception for truly valuable moments or whether it slowly became part of their normal workflow.
@omri_ben_shoham1 @clara_roux This was a fascinating read. The increase in honesty after introducing automatic forgetting makes a lot of sense. Sometimes people don't need AI with a perfect memory. They need a space where they feel safe enough to be completly honest
Murror
@omri_ben_shoham1 Great question. You're right that pinning could theoretically recreate that anxiety, but what we found is that only about 12% of users pin anything — and they tend to pin moments of breakthrough, not everyday entries. The psychology shifts because pinning is an active choice ("this matters enough to keep") rather than passive accumulation. Most users told us they feel relief knowing the default is forgetting. Clara and Omri — to your point about whether it changes how people write pinned entries: we haven't seen that yet, but it's something we're watching closely.
This is super cool. I have a digital health app and think this could be beneficial. It also helps with shortening the prompt fed back into the model!
Murror
@nora_rodden thank you! and yes, the prompt-shortening angle is an underrated benefit. when you're not carrying months of context, inference is faster and cheaper, and the AI actually responds more precisely because it's working with a focused window instead of a sprawling history. would love to hear how the forgetting concept might apply in your digital health app — the trust dynamics around health data are even higher stakes.
@monatruong_murror It's a Catch-22 situation; you need the data but also the trust of the user.
We faced this issue whenever there was a policy update and we needed users to accept/deny. Many users had the tendency to flatly deny and uninstall the extension. Metrics show they didn't even open the policy.
Murror
@roopesh_donde Exactly the Catch-22 we wrestled with. Your experience with policy updates resonates — we saw similar patterns early on. What changed things for us was making the forgetting automatic and visible, not buried in a settings page. When users see "this entry will be summarized in 18 days" right on their journal entry, it builds trust in a way that a privacy policy never could. The transparency of the countdown made the difference.
@monatruong_murror Makes sense, but does that clutter the UI? How did you address the tradeoff. For every generative statement or artifact, we have to add an info tip saying "AI Generated." User experience is compromised.
Murror
@roopesh_donde That's a real tension we think about a lot. Our approach has been to avoid labeling individual pieces of content and instead make the system behavior itself transparent. So rather than stamping "AI Generated" on every insight, we show users the mechanism: "your entries from before June 17 have been summarized" with one line explaining what that means. The clutter problem usually comes from trying to disclaim at the content level. We moved the transparency up to the system level — users understand how the AI works once, and then they trust (or don't trust) the outputs based on that understanding. It's fewer labels, more context.
i think giving users the choice to pin important entries while automatically forgetting the rest is a thoughtful balance. It respects both long-term reflection and personal privacy without forcing everyone into the same experience.
Murror
@bryan_williamson3 appreciate this framing. "not forcing everyone into the same experience" is something we think about a lot. the pin is really about giving users agency without undermining the default. it's a subtle distinction but it matters — the system assumes forgetting, and you opt into remembering, rather than the other way around.
This is such a counter-intuitive finding and honestly one of the most interesting product decisions I've read this year. Everyone in AI is racing to build persistent memory, and Murror deliberately going the other direction because it changes user behavior is really bold.
The insight that "permanent memory quietly kills honesty" is exactly why so many AI journaling and therapy tools plateau. People filter themselves the moment they think something will be remembered forever. Deliberate forgetting turning into a trust signal is one of those design lessons more AI builders need to sit with.
Murror
@nolan_vu thank you — this really captures what we've been feeling but couldn't articulate as well. you're right that most AI builders are racing in the opposite direction. we've had conversations with other founders who think we're leaving value on the table by deleting data. but the value isn't in the data — it's in what users are willing to share when they feel safe. that's the part that's hard to put on a pitch deck.
I like this approach because trust often matters more than storage. Have you tested whether users choose to pin meaningful entries later?
Murror
@advin_jadis yes — and the data surprised us. users do pin, but not the way we expected. early data shows pinning is mostly a one-time "this matters" gesture rather than bookmarking for later. about 70% of pinned entries never get revisited. it's closer to "I don't want this one to disappear" than "I'll come back to read this." which actually reinforces the core idea — the forgetting makes the present moment feel more meaningful, and pinning is how users honor that feeling. we're watching whether pinning habits shift as users get more comfortable with the 30-day cycle.
the part I'd push on is the "anonymous emotional pattern" summary that survives the 30 days. anonymous is doing a lot of work there - if the summary says something like "recurring conflict with a sibling around money" that's still identifiable to the user even with names stripped, it just moved the permanent record one layer down instead of removing it. did you test whether the summaries themselves feel safe to users, or is the trust boost purely about the raw text being gone regardless of what the summary retains
Murror
@galdayan You're pushing on exactly the right thing. We tested this — early summaries were too specific and users flagged them as "still feeling personal." We iterated toward emotional-pattern-level summaries: "tendency toward self-criticism when discussing work" rather than "conflict with sibling about money." The trust boost comes from both layers: knowing the raw text is gone AND feeling safe about what remains. We're still refining the abstraction level, and honestly it's one of the hardest design challenges we face.
@monatruong_murror That's a good line to have found - specific enough to be useful, abstract enough not to feel like surveillance. Curious about the compounding case: after several 30-day cycles, do you re-summarize the existing summary alongside the new period, or do old summaries just accumulate as a growing list? If they compound, does the abstraction drift even higher over time to stay non-identifiable, or does it eventually get too vague to be useful for the pattern detection that's the whole reason to keep it around.
Murror
@galdayan Really sharp question. Right now each 30-day cycle produces its own summary, and they accumulate as separate entries — not re-summarized into each other. We deliberately avoided compounding because of exactly the drift you're describing. A summary-of-summaries loses the texture that makes pattern detection useful. So a user who's been on the app for 6 months would have ~6 discrete emotional-pattern snapshots. The tradeoff is that over a long enough timeline, you're maintaining a list that itself becomes a record — but at a level of abstraction where users have told us it feels more like a "mood journal" than surveillance. We're watching closely to see if that holds at 12+ months.
@monatruong_murror makes sense, and "mood journal" is a much better mental model than a growing archive. one thing I'd watch for at 12+ months - even if each snapshot is non-identifying alone, six of them read together could start triangulating toward something specific, the way anonymized datasets get re-identified by combining several weak signals. probably a smaller risk at pattern-level abstraction than raw text, but worth a look before you call it settled
I appreciate that you measured behavior instead of assumptions. Have you explored whether clear reminders about automatic deletion encourage even more honest journaling from new users?
Murror
@hana_salazars we actually found the opposite surprised us. we tested showing new users a clear "your entries are automatically deleted after 30 days" banner during onboarding. the group that saw it wrote longer first entries on average, but the effect was strongest for users who had previously tried and abandoned other journaling apps. they told us the impermanence felt like permission to be messy. we're still running longer cohort comparisons though.
@Scarlett agreed, and it's a bit counterintuitive too - you'd think perfect memory would build more trust over time, but it sounds like the opposite happens, people start self-censoring once they know everything's being kept forever. forgetting as a feature instead of a limitation is a genuinely interesting design bet.
@monatruong 12% is lower than I expected honestly, I figured more people would hoard "important" stuff out of anxiety even with a forgetting default. makes sense that the framing changes the behavior though. one thing I'd be curious about if you ever track it: whether the pinned entries end up getting revisited later, or if pinning is more of a one-time "mark this as done" action and people rarely go back to it
Murror
@omri_ben_shoham1 great question. early data shows pinning is mostly a one-time "this matters" gesture — about 70% of pinned entries never get revisited. people seem to pin as a way of honoring a moment, not bookmarking it for later. it's closer to "I don't want this one to disappear" than "I'll come back to read this." we're still watching whether that changes as users accumulate more pins over time.