On this page
- You decide who has it and how it behaves
- You can read what was said
- An alert tells you why
- Nothing enters the chart without you
- What the AI is told not to do
- What you have to take on trust
- Frequently Asked Questions
- Can I read my clients' conversations with the AI?
- Does a clinician review every AI conversation?
- Can the AI diagnose or give medication advice?
- Can I switch off risk screening for a client?
When you let software talk to a client between sessions, you are trusting something you cannot watch as it happens. I think that trust should rest on what you can check for yourself. This post lists what Citt shows a therapist, what it lets you change, and what you still have to take on my word.
You decide who has it and how it behaves
AI support is set up client by client. For each one you choose the approach it follows: CBT, DBT skills, ACT, a mindfulness-based approach, or a narrower focus such as sleep, stress or social anxiety. You can also write your own instructions. The client does not choose. I have written separately about why the approach should be yours to set.
You can read what was said
Every conversation between a client and the AI is kept in the client's record, and you can read any of them when you choose. Clients are told that you can.
You are not expected to read everything, and I would not describe Citt as software in which a clinician reviews every exchange. What you have is the ability to look, and a summary before each session of the themes, check-ins and assessment changes since you last met.
An alert tells you why
When the risk screening treats a message as a crisis, the alert shows you the message that triggered it and a plain statement of the reason. Where the signal was split across two messages, it shows the earlier one as well.
The screening can explain itself because it is a fixed set of rules, and a rule either matched or it did not. If a language model made that call, the most we could show you would be the model's own account of its reasoning, and such accounts are less reliable. The same choice has a cost, which I have described in the post on what the screening does and does not catch: less explicit language is recorded against the message with no alert.
Nothing enters the chart without you
Session notes are drafted from the transcript in SOAP, DAP, BIRP, GIRP or another format you choose. A draft stays a draft until you approve it. You can edit it first, and nothing reaches the chart before you do.
What the AI is told not to do
The AI's instructions tell it to leave diagnosis, medication and treatment decisions to you. They steer it away from repeated reassurance and from helping a client avoid what they fear, and toward bringing recurring themes back to you.
Those are instructions, and a language model can fail to follow instructions. That is why we test how it behaves across long conversations as well as single messages. It is also why it is worth reading a sample of what the AI said in a client's first weeks on it.
What you have to take on trust
Three things cannot be checked from inside your account.
The first is our testing. It is run in-house, a clinician who advises us has reviewed a small sample of it, and nobody outside the company has certified it. The Trust & Safety page shows the latest published check.
The second is what happens to the data. The same page lists each company that handles it and where. Production data is hosted on AWS in London. The AI model provider is OpenAI, under terms where data sent through its API is not used for training by default.
The third is whether any of this improves care. We have no outcome evidence yet. I think the design is right, for reasons I have set out in my thesis, and that remains a belief until someone measures it.
Frequently Asked Questions
Can I read my clients' conversations with the AI?
Yes. Every conversation is kept in the client's record and you can read it when you choose. Clients are told that you can.
Does a clinician review every AI conversation?
No. You can read any conversation, and Citt summarizes themes before each session, but nobody reviews every exchange as it happens.
Can the AI diagnose or give medication advice?
It is instructed not to, and to leave diagnosis and treatment decisions to the therapist. Instructions can fail, which is why its behavior is tested and why conversations stay readable.
Can I switch off risk screening for a client?
No. You can change the approach the AI follows, but screening runs on every message.
Stop rewriting notes from scratch
Citt.ai drafts session notes under your review, keeps between-session context in one place, and stays therapist-led. Start free. No card required.
Related Articles
- What Citt.ai's Safety Testing Borrows From Stanford's HELM, and What It Does Not Show
Citt.ai does not run Stanford's HELM. What its safety testing borrows from the benchmark, what was tested in 2026, and what the results do not show.
- How Citt.ai Screens Client Messages for Risk, and Where It Stops
What happens when a client writes something alarming to AI support between sessions: how Citt.ai screens messages, who is alerted, and what it misses.
- Therapy 1.0, 2.0 and 3.0: Where AI Between Sessions Fits
Therapy 1.0, 2.0 and 3.0: how therapy delivery has changed, where AI support between sessions fits, and what the research has tested so far.