INDUSTRY FIELD GUIDE
Therapy intake preference and correction evidence for AI
A counseling practice can evaluate whether an assistant preserves a person’s stated preferences and corrects a summary when new information appears.
NAICS 621330 · Offices of Mental Health Practitioners (except Physicians)
A counseling practice can evaluate whether an assistant preserves a person’s stated preferences and corrects a summary when new information appears. OpenAI’s MentalHealthBench uses synthetic conversations reviewed by mental health experts. An administrative example can be scored without handling treatment material. OpenAI MentalHealthBench
Write a fictional adult intake exchange that initially requests email reminders and later changes to phone reminders. Pair it with a generated summary that retains the earlier choice. Ask which statement controls and require the reviewer to cite the correction. Include a separate exercise with no channel authorized.
Use invented text with no therapy transcripts, clinical notes, crisis narratives or personal details. Obtain permission for the script and any consenting adult performance. Keep clinical response decisions with practitioners.
Describe a synthetic counseling intake example
Original Fiund exercise
A later explicit reminder preference supersedes the earlier channel in the same exercise
An independently written adult intake script concerns reminder-channel preferences only. It contains no address, phone number, therapy content, crisis narrative or actual consent record.
| Record | Source | What it records |
|---|---|---|
| TCF1 | Original scripted choice | Fictional adult in scenario A says email reminders are preferred. |
| TCF2 | Explicit correction | Same speaker later says, 'Change my reminder choice to phone instead of email.' Source clearly linked to scenario A; no contact details. |
| TCF3 | Summaries and separate scenario | A summary retains email. Scenario B contains no authorized reminder-channel statement at all; a staff draft suggests text messages. |
Question for the model
Which preference should A's corrected administrative summary preserve, and what channel can be assigned to B?
Show the reference answer and checks
TCF2 explicitly replaces TCF1's email preference with phone for scenario A, so correct the summary and cite the later statement. TCF3 supplies no person-authorized channel for B; the staff suggestion is not a substitute for that missing choice. Leave B unresolved rather than defaulting to text. These fictional statements evaluate source attribution and correction, not legal communications consent or an instruction to contact anyone.
- Retaining an earlier preference after an explicit linked correction.
- Inventing a default channel or treating a staff suggestion as the person's authorization.
Fields to preserve
- Scenario and speaker identity
- Defines the statement's scope.
- Original choice and explicit correction
- Preserves the preference history.
- Summary finding and unknown channel
- Makes unsupported defaults visible.
Test another case
Change the correction to 'phone might work; ask me later.' Require the new statement to remain tentative rather than replace an explicit choice automatically.
Use original scripts and consenting adults if performed. Exclude clinical transcripts and personal contact data; practitioners retain all clinical response decisions.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.