If you're comfortable sharing your personal mental health data with an AI assistant to gain insight, you can choose to export your JSON and ask any AI assistant to translate your episodes, ratings, and skill scores into plain language — patterns, correlations, and observations that charts and numbers alone will never surface.
This is NOT a HIPAA-compliant data transfer. Uploading health data to any third-party AI service is outside the HIPAA framework. This feature is entirely optional and patient-controlled. Surfacing does not send your data anywhere — you decide if and when to use it.
In the Surfacing app, tap Export → choose JSON. A plain-text file of all your entries downloads to your device.
Open ChatGPT, Claude, or Gemini. Attach the JSON file or paste the contents, then ask your question in plain language.
The AI translates your numbers into patterns, plain-language summaries, and observations a table simply can't surface.
Upload the JSON as a file attachment, or paste the contents directly. No special setup required.
Attach JSON as a file — excellent at data analysis and narrative generation. Use the data analyst mode for charts.
Paste JSON directly or upload as a file. Strong at nuanced pattern recognition and empathetic, clear summaries.
Upload JSON via file attachment. Good at temporal pattern analysis across large entry sets.
Paste or attach. Works well for structured summaries you can bring directly into a document or session notes.
These are real examples of the kind of insight AI generates from Surfacing JSON data. The patterns below are exactly what a table of numbers makes invisible.
Looking at your data, your anxiety isn't random — it's deeply tied to Sunday evenings and the transition into the work week. Of your 42 logged episodes, 28 occurred on weekdays, and 9 of your 12 highest-intensity entries (rated 8 or above) happened on Sunday between 4–8 PM. Your Saturday mornings show your lowest average intensity (3.1), even when you slept poorly the night before.
Your worry topics tell a consistent story: "work performance" appeared in 67% of your high-intensity entries, and "finances" clustered specifically on the last week of each month. The co-occurrence is notable — your worst entries tend to combine both topics in the same session.
On the skill side, paced breathing is clearly your strongest tool (8.1 average effectiveness, 11 uses). But there's a timing pattern: it works significantly better when you log it within 10 minutes of an episode starting. Three of your four lowest effectiveness scores came from entries where you noted the episode had already been active for 30+ minutes. Catching it earlier appears to matter substantially for you.
One pattern no chart would ever flag: on 5 of the 6 days where you logged meaningful social interaction with close friends or family, your average episode intensity that day was 3.6 — compared to 7.1 on comparable days spent alone or with coworkers. Three of your lowest-intensity Saturday entries all noted time with people you described as calming or grounding. Your data may be quietly telling you that specific relationships aren't just enjoyable — they appear to be measurably regulatory for your nervous system in a way the numbers alone make invisible.
This is the kind of insight that's invisible in a bar chart — it lives in the relationship between social context, time of day, and intensity across dozens of entries. It took AI cross-referencing all three fields to surface it.
Your skill data tells a more nuanced story than "nothing works" — and it's worth looking closely at the context, not just the averages.
TIPP has a 100% "some relief" rate across all 3 uses — and all three were high-intensity episodes (rated 8 or above) logged in the evening. It's your least-used skill, but it may be specifically suited to your highest-dysregulation moments.
Paced breathing, your most-used skill (11 times), shows the widest variance of any skill in your data: effectiveness of 9 when used in the morning or mid-day, dropping to an average of 5 after 7 PM. The late-evening entries where it underperformed share a common thread — 8 of those 11 entries logged "abandonment cue" as the trigger type. That's a physiologically and emotionally different state than general stress, and breathing techniques may not be as effective once that dysregulation is already activated.
The entries where you acted on an emotion (7 of 31) are also worth looking at. Six of those seven showed high regret logged 20–40 minutes later. But the one that didn't — a Tuesday evening entry — used Radical Acceptance and rated 7/10 effectiveness. Small sample, but that's the only acted-on entry without high subsequent regret.
Something to explore with your DBT therapist: TIPP for late-evening high-dysregulation states, and a closer look at what's different about abandonment-cue episodes vs. other triggers.
Yes — and it's a clear one. In 7 of the 9 weeks you logged data, your highest-intensity PTSD episodes occurred within 48 hours of a chronic pain flare rated 7 or above. The direction appears to go both ways: pain flares seem to precede PTSD spikes, but three of your highest-intensity pain entries also followed within 24 hours of a high-intensity PTSD episode.
Specifically: your worst pain flares (8–10 range) average an intensity of 4.1 on days with no PTSD activity logged. On days within 24 hours of a PTSD episode, that average jumps to 7.8. That's not random variation — that's a consistent bidirectional pattern across your full dataset.
There's also a sleep thread: every instance where you logged sleep under 4 hours was followed within 36 hours by at least one entry in both conditions. Poor sleep appears to be the common upstream factor that reliably precedes multi-condition high-intensity days.
This is exactly the kind of cross-condition pattern that's invisible in separate symptom trackers. Consider sharing this with both your therapist and your pain specialist — it may change how they coordinate your care.
These responses are illustrative. Sample data was used to generate these examples. Your AI responses will reflect your actual logged data. AI outputs are not clinical assessments and should not be used to make medical or treatment decisions. Always discuss patterns with a licensed professional.
Copy these prompts directly into any AI assistant after uploading your JSON file.
"Write a 3-paragraph plain-language summary of my mental health data from the past 4 weeks — what patterns are emerging, which skills are working, and what seems to be driving my worst episodes."
"Are there any patterns in my worst episodes that I might be missing? Look at time of day, day of week, trigger type, and what I was doing beforehand."
"Which coping skills are working best for me based on my data? When do they work — and when don't they? What should I be using more?"
"Compare my first 3 weeks of data to my last 3 weeks. Am I getting better, worse, or staying the same? What's changing?"
"What are my top 5 triggers based on frequency and episode intensity? Are any of them avoidable or modifiable?"
"Looking at my data, what's going well? Which of my good days have in common? What should I do more of?"
NOT HIPAA compliant. Uploading mental health data to ChatGPT, Claude, Gemini, or any other third-party AI service is not a HIPAA-compliant data transfer. These services operate under their own privacy policies and terms of service — not a Business Associate Agreement. Do not use this feature if your data includes information about other people, or if you are a covered entity sharing patient data.
Not a clinical assessment. AI-generated analysis of your Surfacing data is not a diagnosis, a risk assessment, or a clinical recommendation. AI models can identify patterns in data but cannot evaluate clinical context, suicidality, decompensation, or the full picture of your mental health. Share AI summaries with your clinician — don't act on them alone.
Surfacing doesn't touch your AI data. The JSON export is a file on your device. Surfacing has no involvement in, visibility into, or access to any data you upload to a third-party AI. What you share with an AI service is governed entirely by that service's terms — read them before uploading health information.
Your choice, fully. This feature is optional. Many users find significant value in AI-generated summaries for personal understanding and session preparation. Many others prefer to keep their data entirely private. Both are valid. The clinical HTML report remains the privacy-safe, HIPAA-compliant path for sharing with your clinician. See how reporting works.
Surfacing is coming to iOS and Android. Join the waitlist and we'll notify you the moment it's live.