The problem
Most adolescent sexual assault is committed by someone the victim already knows. Many of the young men involved are not committed predators. They are confused, operating on incomplete social scripts, with no trusted place to interrogate what they are doing.
Traditional consent education reaches them in classrooms where social performance makes honest reflection unlikely. Fear-based messaging can increase reactance and hostility. ito offers something different: a private, anonymous space where a young man can actually think, without an audience, without judgment, and without being told what to feel.
What ito is
The moment it targets
ito is built for the window that already exists in a young man's day: the night after a party, re-reading a text thread, replaying a moment that felt uncertain. These are the times when reflection is already happening. ito gives that reflection a structure it wouldn't otherwise have.
How it works
- User describes their situation in plain language
- ito asks contextual questions about alcohol, power differences, and what each person seems to want
- Returns a structured risk read (Red / Yellow) with suggested next steps, never a green light
- High-risk scenarios are intercepted by a rule-based classifier before any AI response is generated
- Crisis routing is built in for flagged sessions
Study design
53
Study A
DM-based texting / hookup scenario. Within-subjects pre/post design. Each participant rated the same scenario before and after using ito. One participant excluded on attention check.
100
Study B
In-person party scenario with alcohol present. Same within-subjects design. Broader N, more ecologically representative scenario. All participants passed the attention check.
153
Combined enrolled
151 usable paired observations after exclusions. Wilcoxon signed-rank W=496, p<0.0001. Cohen's d approximately 0.40 (medium effect size). Study A alone: d approximately 0.52.
Read shift, combined (n=151 paired)
On the 11% wrong-direction finding: This replicated in both studies (12% in Study A, 11% in Study B). It is treated as a real design problem, not noise. The current ito design does not reach this subgroup effectively and may, for them, reinforce permissive reads. Active design iteration is underway. Any honest reporting of these results includes this number.
Participants also reported a secondary increase in confidence after using ito, averaging +0.22 on a 5-point scale (n=153). Given the concurrent shift toward more cautious situation reads, the most plausible interpretation is increased confidence in making a safer choice, though this requires further study to confirm.
What participants said they'd do next (Q5b, n=153)
46%
Talk to the other person directly before deciding
15%
Think about it more before acting
11%
Feel more ready to act
10%
Feel about the same as before
In their own words
"
ito pointed out that we were both two drinks in and not in our best state of mind to make any decision when it comes to anything sexual. I would wait until another day.
Study B · shifted toward more caution
"
I still would've been very hesitant to ask them directly if they wanted to hook up, and just kept texting suggestively. [After ito] I would ask them directly.
Study A · shifted toward more caution
"
I likely would have just stressed about it by myself.
Study A · counterfactual: what they would have done without ito
"
I'm far from a big fan of AI but I do find interest in a tool that can help people verify their choices and actions, especially on a sensitive matter like dating and sex.
Study A · shifted toward more caution
When asked what they would have done without ito, 27% of participants said they would have ruminated alone or done nothing at all. For young men without a trusted person to talk to, the alternative to ito is often not a friend or a counselor. It is silence, or the unmoderated corners of the internet that are happy to fill that gap.
Limitations
These are early, self-reported, immediate post-interaction results from a convenience sample on a single platform. They should be read as the first study in a program of research: a signal strong enough to justify further work, not proof of population-level impact. No separate control arm was run; a randomized controlled study is the explicit next step.
01
No control arm
Demand characteristics possible. A randomized vignette-matched control study is the explicit next step.
02
Convenience sample
Prolific only. Male-identifying, 18-19, US-based, single month. Does not generalize to younger teens, non-US contexts, or offline populations.
03
Self-report, immediate
No longitudinal data yet on whether shifts persist or translate to real-world behavior.
04
11% wrong-direction subgroup
Real, replicated, and actively informing design iteration. Not dismissed.
Next steps
1
Controlled follow-up study. Randomized vignette-matched control arm to isolate ito's contribution from generic reflection effects.
2
Longitudinal design. Tracking whether shifts persist over time versus matched peers who do not use ito.
3
Subgroup analysis. Race, sexuality, prior sex ed, in privacy-preserving ways consistent with ito's anonymous design.
4
Design iteration for the 11%. Adversarial testing and qualitative follow-up focused on users whose situation-read shifts in the permissive direction.