How AI-generated content is labelled
Where Protocol tells you (and your client) that something came from AI, what gets recorded even when no label is shown, and why a report you reviewed goes out under your name.
Basic AI touches a few specific places in Protocol: an assistant you connect yourself, a progress report an automation drafts for you, and the meal scanner your client uses in the app. Everywhere it does, you can tell.
This page is the map: what carries a visible label, what is recorded behind the scenes, and what Protocol deliberately does not put in front of your client.
Where you see a label
The four places a label appears today:
| Where | What it says | Who sees it |
|---|---|---|
| Connecting an AI agent | “You are connecting an AI system. It acts on your Protocol account and its actions are logged; content it creates is recorded as AI-generated.” | You |
| The approval screen when an assistant asks for access | “This authorizes an AI system to act on your account. Its actions are logged.” | You |
| A progress report drafted by an automation | AI draft, beside the title in the editor | You |
| A meal logged with the scanner | AI estimate, on the food entry | You and your client |
The AI estimate label is the only one your client sees, and it is on the food entry in their app as well as in your foods panel. It means the portion and macros were read from a photo rather than entered by hand, so treat the numbers as an approximation and correct them when they look off.
What is recorded even when nothing is shown
Whether or not a label is on screen, Protocol records provenance at the moment content is created: what produced it, which automation or agent connection it came from, and when. That record is attached to the content itself, so months later it is still answerable which reports began as a draft and which you wrote from scratch.
Two details worth knowing:
- Only content an AI actually created is marked. If you write a report yourself, nothing marks it, even if an assistant was connected at the time. Editing an AI draft does not erase the record either, so a draft you rewrote end to end is still recorded as having started as one.
- Your assistant’s actions are logged. Everything an agent key does goes to the same usage log you can see on the AI Agent page, alongside what it touched. See Safety, scopes & privacy.
What your client does not see
A progress report you reviewed and sent carries no AI label for the client. That is deliberate. Once you have read a draft, edited it and chosen to send it, the report is your professional communication, and the draft that preceded it is a detail of how you work rather than something your client needs to weigh. The record of the draft still exists on our side.
The same reasoning is why the meal scanner is labelled: nobody reviewed that number before your client saw it, so the app says where it came from.
No AI talks to your clients
There is no AI that converses with your client inside Protocol. When you use an assistant, the conversation happens in claude.ai or ChatGPT, on your side, and Protocol is the account it acts on. Nothing in your client’s app is an AI chatbot, and no message from you to a client is written and sent by an agent without you asking for it.
Related
- Connecting an assistant — the connect flow the banner belongs to
- Safety, scopes & privacy — access levels, hard limits, the usage log
- Analysing & reports — what the agent can draft for you