Your data and AI
What Sprout is given, and what stays here.
Sprout drafts your minutes, your pages and your emails — and it does that without your members' contact details. Email addresses, phone numbers, postal addresses, dates of birth, government identifiers and the values in your own custom fields stop at our systems: a redaction step removes them, and a validator refuses any request still carrying one. Sprout writes the pattern; our software runs it against your records.
What a request carries
- What you asked for, in your words
- Your organization's public-facing writing — page copy, event descriptions, the way your past posts sound
- The names of the fields you keep, so a draft can be shaped around them
- A member's name, where you are drafting something about that person
What stays on our side
- Email addresses
- Phone numbers
- Postal addresses
- Dates of birth
- Government identifiers — tax ids and social security numbers
- The values typed into your own custom fields
A member's name on its own is allowed across, so a draft can address someone by name. It is the rest of their record that is stripped.
How the rule holds, and not just on paper.
Three steps, in this order, on every request SorrelSites makes.
- 1
Stripped before the request exists
Member data is passed through a redaction step first, which replaces contact and identifier values with placeholders. By the time a request is composed, they're already gone.
- 2
Checked again at the door
Every request leaves through one place in our code, and that place scans it a second time. Anything holding a real-looking email, phone number, tax id or date of birth is refused before it reaches a provider.
- 3
The pattern comes back; we run it
What the model returns is a draft or an instruction — the shape of the email, the fields to create, the minutes to review. SorrelSites applies it to your real records itself.
What happens to what comes back.
Output is a draft until a person says otherwise
Anything an AI feature produces — a page, an email template, a set of member fields, a meeting's minutes — is saved as a draft. It reaches your members or your public site when somebody in your organization reads it and publishes it.
Every request is logged for your audit trail
Each call is recorded against your organization with what it produced, so there's a record of what was asked and what came back. It's your log, and it's one of the reasons the policy above is checkable instead of promised.
Your information is not training data
What your organization holds in SorrelSites is not used to train AI models — ours or a provider's. The models draft; they do not learn your roster.
Documents are opt-in, one at a time
An uploaded document is readable by AI features only when an admin switches AI reading on for that specific document. Switch it back off and the content is withdrawn again.
The Notetaker joins as a guest everyone can see.
When your organization records a board meeting, the Notetaker appears in the call as a named participant — "SorrelSites Notetaker" — in the same list as everybody else. There is no silent recorder.
Recording is yours to control at two levels: your organization can switch it off entirely, and any single meeting can be held back from recording on its own. SorrelSites shows a reminder to tell attendees in advance, on the meeting itself and in the confirmation email — worth taking seriously, because consent law varies by state and by country, and that call belongs to your board.
The recording is transcribed so the minutes can be drafted from it, and the transcript stays with that meeting in your organization's records. Once your board approves the minutes, the approved version is frozen as the official record.
Who else touches it.
Two model providers generate the drafts — Anthropic and Google — under the limits above. Recall.ai handles recording and transcription, and only for organizations that use the meeting Notetaker. Your payments run through your own Stripe account, and your dues land there directly.
Every processor SorrelSites relies on is listed by name, with what it handles and why, in the Privacy Policy — which is also where the full legal version of everything on this page lives. This page is the plain-English version; the policy is the one that binds us.
Data and AI, answered
- Does the AI see our members' personal information?
- No. Member email addresses, phone numbers, postal addresses, dates of birth, government identifiers and the values typed into your own custom fields are held back from AI models by a rule enforced in code. A redaction step strips them before a request is built, and a validator re-checks every request at the single point where it would leave our systems and refuses the call if one is still present. A member's name on its own is allowed, so a draft can address someone by name.
- How does it draft an email to members if it doesn't have their details?
- Sprout writes the pattern, and our own software runs it. It drafts the email with placeholders for the things that vary — a name, a renewal date, a chapter — and SorrelSites fills those in from your records at the moment of sending, on our side of the boundary. The result is the same personalized message; the model producing it works from a template, not from your member list.
- Which AI providers do you use?
- Anthropic and Google provide the models that generate drafts, under the data limits described above. If your organization uses the meeting Notetaker, Recall.ai handles the recording and transcription for that meeting. Every processor SorrelSites uses is listed by name and purpose in our Privacy Policy.
- What about our board transcripts?
- A transcript you upload or record does go to the model, because it is what the minutes are drafted from — and it keeps names and speakers, since attendance and who moved what are the parts that matter. Contact details found in the transcript are stripped before it is sent. Recording is something your organization controls per meeting, and the Notetaker joins the call as a visible, named participant.
- Can we turn AI features off?
- You can leave them unused — nothing in SorrelSites acts on its own, so an organization that ignores Sprout gets a membership system that behaves like one. Documents stay unreadable by AI features until somebody opts a document in, and meeting recording can be switched off for your whole organization or for any single meeting.
- Do you sell our data?
- No. Your members, your content and your data belong to your organization. SorrelSites processes them on your behalf and at your direction to run the service, and you can export them at any time without asking us and without an export fee.
Bring the questions your board will ask.
Half an hour with the person who builds SorrelSites — including the parts your secretary, your treasurer or your counsel want pinned down.