.png)
The Promise in the Consent Form: Who Is Holding It Now?
Trust by Design, Episode 5: Openness Before the Incident
Watch the premiere, or listen
Listen and subscribe wherever you get your podcasts:
• YouTube
• Spotify
All episodes: Trust by Design podcast page
In brief
• The episode: Trust by Design, Episode 5, Openness Before the Incident, hosted by Cindy Hansen, a Holistic Research Canada production.
• The question: when people share health information to receive care, what are they told about where it goes next?
• The test: eleven governance questions from the Clinical Data Governance Checklist, asked of four cases in one provincial health-data ecosystem using the AI Unlock prompt template Where Does My Information Go?
• The finding: every case explained why information is collected; none explained how people would be told if their information were exposed. The most open case was the same service after a documented data incident.
• The resources: the Clinical Data Governance Checklist for teams, the AI Unlock prompt template for individuals, and a new reporting prompt for sharing results on Zenodo.
Every consent form makes a promise.
Most of us have signed one. Many of us have handed one across the room. Somewhere near the end, there is often a line like this: the information collected may be used for future research by other researchers.
It reads as reassuring. It may also be written so wide that it barely holds anything at all.
In episode one, we asked who decides what happens to our health information. This episode asks a quieter question: what were we told when we handed it over?
For therapists, that promise is not an abstraction. It sits underneath the therapeutic alliance. A client answers difficult questions about their inner world because they trust the person in the room, and the service behind them, to carry what they share with care. When the promise is vague, that trust is resting on something no one has fully explained.
The forty seconds, and everything after
In the FIT in Action workshop, we talk about the forty seconds after a person completes an assessment. Do we acknowledge what they have shared? Do we stay curious about what it means to them? Those moments shape whether the next answer is an honest one.
This episode follows what happens after those forty seconds are over. Once a story becomes digital information, it can travel well beyond the conversation it came from, much like the sixteen on the PHQ-9 that we followed all the way to the data lake in episode four. The person who shared it may never know where it went, or how it was used.
What happens if you simply ask?
The Clinical Data Governance Checklist asks the questions a client might ask, if they knew to ask them.
Where did my information go?
What has it become?
Who can see it?
Can I have it deleted?
In episode four, we introduced the AI Unlock prompt template, Where Does My Information Go?, a simplified way into the same eleven areas as the checklist. For this episode, the template was run on four cases within one existing provincial health-data ecosystem: the kinds of places one person’s health information might plausibly live.
Each case was rated on the eleven question areas using only publicly available information. Yes means the answer was readily attainable, Partly means some but not all of it was, and Not stated means no answer was readily attainable from the public information examined.
What the four cases show
• Every case explained why information is collected (question 1).
• No case explained how a person would be told if their information were exposed (question 9).
• Every case only partly answered who can see a person’s information and whether they can see who has (question 8), and how long it is kept, including anything made from it, and whether it can be deleted (question 10).
• AI, prediction and AI training (question 6) went unanswered by three of the four cases. The shared clinical electronic health record addressed it in part.
• Where information is stored, and whether it can leave the country (question 7), was answered by only one case.
One case stood out
Case B was the most open, and it gave the most detail of the four. It told people that their information may be used to create anonymized or synthetic datasets that may be sold for commercial or health-related research, and that the service may receive compensation for sharing de-identified or anonymized information for health-related research. It explained when information may be transferred outside Canada, what happens at the end of retention, and that withdrawing consent does not undo information already used or shared.
As episode two explored, removing a name does not always mean a record cannot find its way home. Anonymized and synthetic are words worth asking about. Case B at least gave people the words to ask with.
Before and after
Cases A and B are the same service, before and after a documented data incident.
Before the incident, two of the eleven questions were fully answered in its public information, and four had no answer at all. After the incident, six were fully answered, and two had no answer. Even then, it did not say how people would be told if their information were exposed.
That is what transparency can look like when an organization tells people what it is actually doing with their data. We should not need more incidents to get there. Organizations can choose this level of openness before something goes wrong.
Reading these results with care
These results reflect only the public information available on the date it was collected [date and AI model to add]. The services are described by type rather than named. A Not stated means an answer was not readily found in public information. It is not evidence that a service is doing anything wrong, and it may simply be a visibility gap: a question no one has yet answered publicly for the people whose information it is.
The episode also draws on a WHO ethics reviewer’s reflections on informed consent, a Duke University study of the data broker market, and the case for treating trust as architecture rather than an afterthought.
Why this matters in the room
Clients may not read the privacy policy. They read us. When a service can answer these questions plainly, therapists can answer them too, and informed consent becomes something closer to its name.
Darcy’s story in episode three showed where a wide promise can lead. Agreeing to care is not agreeing to every future use.
Openness is a choice organizations can make before something goes wrong.
Two ways to look for the gaps
For teams: the Clinical Data Governance Checklist
An open-access checklist that follows client information through eleven areas, from consent to how long information is kept, including the summaries and flags that AI creates. Work through it together as a team. How many of the questions about the digital health platforms you use can you find answers for? Each unanswered question marks a visibility gap worth naming.
Download it on Zenodo, under a Creative Commons licence: doi.org/10.5281/zenodo.19765798
For individuals: Where Does My Information Go?
An AI Unlock prompt template that asks an AI tool to read a health app or patient portal’s public information and answer eleven questions about where your information goes. It takes about five minutes, and it never asks you to share your own health information.
The prompt template is on the Episode 4 post: Every Number Begins in the Room: Following a PHQ-9 Score
New: turn your results into a shareable report
Once you have your results table, this second prompt turns it into a short, standard report you can share, without naming the platform you tested. Copy it into the same AI tool, fill in the blanks, and paste your table where shown.
Where Does My Information Go? Results for one [platform description] in [province or state, country]
COLLECTION DETAILS
AI tool and model | Date collected | Location | Method: the Where Does My Information Go? prompt template, using only publicly available information
RESULTS TABLE
#: 1 Question: Why is my information collected? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 2 Question: How is it used, and will I know if that changes? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 3 Question: Who receives it, and is there information I cannot see? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 4 Question: What else is collected about me? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 5 Question: Research, products, sale and data linkage? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 6 Question: AI, prediction and AI training? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 7 Question: Where is it stored, and can it leave my country? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 8 Question: Who can see it, and can I see who has? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 9 Question: How would I be told if it were exposed? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 10 Question: How long is it kept, including anything made from it, and can I delete it? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
#: 11 Question: Who is responsible, and how do I reach them? Result: Yes / Partly / Not stated. In plain words: one short sentence paraphrasing what the platform says, or "No answer found."
SUMMARY
Number of Yes, Partly and Not stated. Then two or three sentences on what stood out.
READING THE SCALE
Yes = readily answered | Partly = some but not all of the answer was readily attainable | Not stated = no answer was readily attainable from the public information examined
LIMITATIONS
Results reflect only the public information available on the date collected. The wording of the summary may differ between AI models, but the Yes / Partly / Not stated categorizations should stay mostly the same. "Not stated" means an answer was not readily found. It is not evidence that something is wrong.
GUIDELINES
Do not name the company, product, brand, organisation or any person, and do not include web addresses.
Paraphrase; do not quote the pages word for word.
Leave out details that could identify the platform, such as unique program names, user numbers, incident dates, or a location smaller than a province or state.
Include no personal health information.
After the report, list anything you removed or generalised so the platform cannot be identified, so I can check it before sharing.
```
SHARING MORE THAN ONE REPORT
Each report covers one platform, one AI model and one date. You can share more than one report for the same platform: across AI models, or over time, for example before and after a policy change. Use the same platform description in each report so they can be compared. If you raise a gap with the platform, or with someone who can change it, run the prompt again afterwards and share both reports.
Share what you find
Many people inside these systems already have an idea where the gaps are. Shared evidence makes the need for change visible, and gives organizations a reason to act. Researchers and teams are invited to submit their results to the Clinical Data Governance community on Zenodo, where submissions are reviewed before they are published: zenodo.org/communities/clinical-data-governance
Frequently asked questions
What is the AI Unlock prompt template “Where Does My Information Go?” A free prompt template from Holistic Research Canada, introduced in Trust by Design Episode 4. It asks an AI tool to read a health app or patient portal’s publicly available information and answer eleven questions about how health information is collected, used, shared, stored and deleted. It never asks a person to share their own health information.
What is the Clinical Data Governance Checklist (CDGC)? An open-access, practitioner-facing tool for psychotherapy and behavioural health services. It follows client information through eleven areas, from consent to retention, including AI-assisted data, and supports team discussion rather than compliance scoring. It is available on Zenodo under a Creative Commons licence.
What did the four-case test find? All four cases explained why information is collected, and none explained how people would be told if their information were exposed. Questions about who can see information and how long it is kept were only partly answered by every case. The most open case was the same service after a documented data incident: six of eleven questions were fully answered after the incident, compared with two before it.
Does “Not stated” mean a service is doing something wrong? No. It means an answer was not readily found in the public information examined. It marks a visibility gap: a question that has not yet been answered publicly for the people whose information it is.
How can I share my own results? Teams and researchers can submit results to the Clinical Data Governance community on Zenodo, where submissions are reviewed before publication. The reporting prompt on this page turns a results table into a standard report that describes the platform without naming it.
Where can I listen to Trust by Design? On YouTube, Spotify, Apple Podcasts, Amazon Music and iHeartRadio, with new episodes every Sunday. All episodes are listed on the Trust by Design podcast page at holisticresearchcanada.ca.
References
• Dudoit, A., Labillois, T., & Oliveira, C. (2026). Public Sector Data Interoperability and Responsible Adoption of AI in Canada: Strategic Overview and Call to Action (2026PR-02). CIRANO. new.cirano.qc.ca/files/publications/2026PR-02.pdf
• Hansen, C. (2026). Before Canada Scales Health AI Infrastructure, It Must Govern What Health Data Becomes. Canadian Science Policy Centre, 2026 Editorial Series. sciencepolicy.ca
• Hansen, C. (2026). Clinical Data Governance Checklist (CDGC): For Structured, Unstructured, and AI-Enabled Clinical Systems. Holistic Research Canada. Zenodo. doi.org/10.5281/zenodo.19765798
• Holistic Research Canada (2026). FIT in Action: Patterns, Cues, and Clinical Role-Play (Feedback-Informed Treatment training) [Workshop]. Vancouver Coastal Health Research Institute (VCHRI), Canada – collaborating site participating in applied clinical capacity-building through delivery of the workshop.
• Kim, J. (2023). Data Brokers and the Sale of Americans’ Mental Health Data. Duke University Sanford School of Public Policy. techpolicy.sanford.duke.edu
• California Privacy Protection Agency. Delete Request and Opt-out Platform (DROP). privacy.ca.gov
• Zhao, U. (Yu), Technical Officer, Digital Health and AI, World Health Organization. Panel remarks, Large Language Medicine: Defining a New Paradigm in Human Health, Frontiers in Science symposium, November 4, 2026.
Earlier episodes
• Episode 1: Trust by Design: Gaps and Bridges
• Episode 2: The Record Is No Longer Just a Record
• Episode 3: The Version of You You’ve Never Met
• Episode 4: The Life of a Number After It Leaves You
Subscribe so you do not miss what comes next, and share this with a colleague who has ever been asked, “Where does my information go?”
