Who this covers
This policy covers Groundwork coaching clients and their plus-ones. If you're a consulting client, your contract covers how AI is used on your engagement, and new consulting contracts include an AI rider that spells it out. The retention schedule below applies unless your contract sets different terms. Our Privacy Policy covers everything else we collect and lists the services we use.
What AI does with what you send in Groundwork
Voice messages. When you send a voice message in your channel, it's transcribed on Matthew's own computer so he can read it with everything else you've sent. It isn't sent to a transcription service.
A brief before each reply. Before Matthew sits down to reply, AI writes him a short brief for your channel: what you asked, what's changed since last time, one to three resources from our library that might help, and the full text of your messages underneath. The brief doesn't draft a reply. Matthew reads your messages and writes his own.
A check for messages that can't wait. An automated check reads each message for signs that you might need Matthew before his next reply, and alerts him if it finds one. It's a convenience that may get your message read sooner, but it can miss things, so please don't rely on it to get help. It isn't safety monitoring, nothing replies to you on its own, and it doesn't promise any response time. If you're in danger, call 911, or call or text 988. How Groundwork coaching works says more.
The librarian bot. Your channel has a bot you can ask about Fieldway's library and courses. You can ask it follow-up questions and it'll tell you which part of a resource fits what you asked, and every answer points to Fieldway material. It doesn't give advice, interpret your situation or search the web, because that's Matthew's and April's job. When the library has nothing on your question, it says so and Matthew sees the question. He can read the log of every question and answer, and your questions also show up in your next brief.
Answers come from a Haiku or Sonnet model made by Anthropic; the exact model is named in your channel, and we'll tell you there if it changes. A model can be wrong, so check anything that matters. The bot asks you to accept a short notice, and we keep a record of your answer, whether you accept or decline, as the Privacy Policy describes. If you decline, the bot won't answer in your channel, and Matthew and April reply to you as usual.
Calls. We use Granola, with its model training turned off, to transcribe discovery calls and quick calls. We'll tell you when a call is being transcribed, and if you'd rather it not be, tell us and we won't transcribe it.
Notes on how you like to communicate. Our contact records include notes on how each person prefers to communicate. They're drafted with AI from your messages, used only to shape what we write for you, and never shown to anyone else.
Web research. The text of your messages never goes into a web search. When a brief needs outside material, the search is written without anything that identifies you.
What AI never does
AI never writes or drafts a coaching reply in your channel. Apart from the librarian bot's answers about our library, every reply there is written by Matthew or April, and the judgment in it is theirs.
No AI model sends email for us. AI may help Matthew draft an email, and he reviews every draft before it goes. He writes or approves every email we send you one to one, and scheduled emails, such as a welcome series, are written ahead of time and sent on a schedule.
Who sees what you send
Only Matthew and April read your channel, along with your plus-one if you have one. Your briefs are Matthew's, and April sees one when Matthew brings her in on a message.
These services handle your messages so the tools above can run. We've set up every one of them so it doesn't use your messages to train AI, and Matthew checks those settings each time he reviews this policy.
| Service | What it handles |
|---|---|
| Slack | Your private channel. We've opted out of Slack's model training. |
| Anthropic | The briefs, the check for messages that can't wait, the librarian bot, help drafting email, and the notes on how you like to communicate |
| Vercel and Neon | Receiving and storing your channel's messages, voice messages included, for the bot and the briefs |
| Granola | Call notes and transcripts, with model training turned off |
| Matthew's own computer | Transcribing voice messages, and working copies of AI sessions, with encrypted backups through Backblaze |
The Privacy Policy lists the other services we use and what each one handles.
How long we keep it
Coaching clients. We keep your coaching data for six months after your term ends, so you can pick back up where you left off if you come back, and then we delete it within 30 days. Matthew checks in with you before we delete it on that schedule, and you can ask us to delete it sooner or to keep it longer because you're returning. At the end of your term your channel is archived rather than deleted, so it can reopen. We delete the archived channel at six months too, unless you've asked us to keep your data.
Consulting clients. We keep engagement data for 90 days after the engagement ends, then delete it within 30 days, unless your contract sets different terms.
Call notes and transcripts. Granola keeps call notes and transcripts until Matthew deletes them, and it doesn't delete them on its own. He keeps each client's calls in a folder of their own, and when he deletes the rest of your data, on the schedule above, he deletes the notes and transcripts in that folder, along with the copies that sync into his own notes. If we talk on a discovery call and you don't go on to sign, he deletes that call's notes and transcript six months after the call, when we delete your application answers.
What the law makes us keep. Your signed agreement, invoices and payment records are business records, so we keep them for as long as the law requires, separately from the rest of your data.
At any time. Anyone can email privacy@fieldway.org and ask us to delete their Slack data. We do it within 14 business days, apart from any record we keep of how you answered the bot's notice, which the Privacy Policy describes. Copies in our encrypted backups are kept for up to a year after that, then removed.
What the services keep. Anthropic keeps copies of the bot's requests for up to 30 days, or longer if a request is flagged for misuse. AI sessions on Matthew's own computer leave working transcripts there, and they're deleted within 60 days: they move to the Trash after 30 days, and the Trash empties within another 30. Some services keep their own copies under their own terms, which we can't shorten.
What stays after a deletion. Our contact record keeps your name, contact details, key dates, Matthew's own notes and the notes on how you like to communicate, with any quoted message text removed. You can ask us to delete those notes too. Our Data Deletion page has the full schedule.
Why there's no opt-out
Matthew can't run Groundwork efficiently or affordably without these tools, so there isn't a version of it without them. They're the same tools he uses with multi-billion-dollar companies that have strict security requirements. Because you can't opt out, this policy is linked from the Groundwork page, you acknowledge it in the client agreement before you sign, and it's linked from your invoice. It covers your plus-one too, and when your plus-one is invited to your channel, they're asked to accept the same bot notice you were.
The library and courses
When you open a library item or a course, or finish a course module, we record which one it was and when, against your account. We don't record anything you type or read beyond that. Only Matthew sees it. We use it to improve the material and to pay the partners whose work is included: our Conflict Campaign partner, for example, receives a royalty for each client who uses the Conflict Campaign scenario library, which comes with Groundwork as a course. It's kept on the same schedule as the rest of your data.
Changes to this policy
Matthew owns this policy and reviews it every six months. When we change how your messages are handled, we update this page and the date at the top.
Questions about this policy go to privacy@fieldway.org.