Your content and how we use AI

1. How we use AI

We do use AI, but only to help build what we deliver — never to train anything.

It writes first drafts. A person at Bridger reviews every one and releases it — nothing reaches your workspace until they do. Nothing about your content makes any AI model smarter: we do not use your material to train, fine-tune, develop or improve any model.

We would rather you checked than took our word for it, so we name every service that touches your content and link its terms below.

AI drafts two things in the Portal. Each one is started by a person at Bridger.

• A short description of a presentation, for when its link is shared. It is drafted from that presentation’s title, goal, audience, key message and call to action, and your workspace’s name. A Bridger team member saves it; until then it exists only on their screen.
• A sample of Practice content, so you can see the shape of it before you send us anything. It is drafted from your workspace’s name, the text of up to five pages of your public website, and the model’s general knowledge of your market. The model is told not to name real customers or state their results, and customer stories are removed from the draft automatically. The draft stays hidden from your workspace until a Bridger team member releases it.

Before you’re a client, we draft that sample from your public website. Once it has been released to you, it is never drafted from your website again. Drafting from your own material — the decks and documents you send us — is not built yet. When it is, your material will be the only source, and this page will say so before it is used.

Each request carries material from one workspace only. Nothing from another client’s workspace ever goes into a draft for yours. The draft that comes back is stored in your workspace and nowhere else, and we do not keep the requests we send.

The drafting is done by Anthropic’s Claude models, through Anthropic’s API.

Anthropic does not train its models on the text we send. Its commercial terms (anthropic.com/legal/commercial-terms) say so directly: “Anthropic may not train models on Customer Content from Services.” We send that text through Anthropic’s API, and Anthropic deletes API inputs and outputs within 30 days. The exceptions are Anthropic’s, not ours: if its systems flag a request as a possible violation of its usage policy, it may keep that request for up to two years, and it keeps anything the law requires it to keep. We do not have a zero-data-retention arrangement with Anthropic, so these standard terms apply to what we send.

2. Who can see your workspace

Your workspace is visible to your own team and to Bridger’s Portal team, all of whom are under written confidentiality agreements. Bridger’s access is granted at the team level rather than per client workspace.

On your side, each person sees what their role allows; your workspace owner sets the roles. On Bridger’s side, changing a workspace is limited to Bridger’s owners and administrators and to team members assigned to it as editors. No other client can see anything in your workspace: database row-level security and a workspace check on every request keep each one separate.

3. Where it is stored

The Portal’s database, application and files are all in the United States.

• The database and sign-in: Supabase, US East (North Virginia).
• The application: Vercel, Washington, D.C. Pages are delivered through Vercel’s global network.
• Files: Cloudflare R2, Eastern North America. That covers each presentation’s published files, the images and media uploaded to your workspace, files submitted through forms in a presentation, and screenshots of presentations.

A presentation’s published files are public at their address; that is how a shared link works.

Files submitted through a form are stored at a unique, unguessable address. They are not listed or indexed anywhere, and the address is generated per file. They are not individually access-controlled today.

The source of your presentations is kept in a private GitHub repository that Bridger creates for you.

4. The companies that touch your content (our subprocessors)

These are the companies that store or process your content for us, what each one does, and its terms.

• Supabase — the database and sign-in. Terms
• Vercel — hosting the application. Vercel KV holds rate-limit counters: IP addresses, no content. Terms
• Cloudflare — R2 file storage, and Browser Run, which opens a published presentation to take its thumbnail. Terms
• GitHub — the private repository that holds your presentations’ source. Terms
• Anthropic — AI drafting (section 1). Terms
• Resend — sending email: sign-in codes, invitations, notifications and digests. Terms
• Sentry — error monitoring. It is set not to send personal data by default. Terms
• Google — Bridger staff calendars, for Bridger’s own time tracking. None of your workspace’s content goes to Google. Terms

We will tell your workspace owner before we add a company to this list that would receive your content.

Your workspace can also send notifications to Slack, Zapier, email or a webhook. Those go wherever your workspace’s settings send them. You choose them, or we set them up at your request, so they are your destinations rather than our subprocessors.

5. How long we keep it, and deleting it

Your content is kept while your workspace exists. A record that someone ran a Practice play is deleted after 24 months.

Deleting a workspace takes two steps by a Bridger owner or administrator: archive it, then delete it. Deleting it removes the workspace’s database records — people, projects, content, Practice records and settings — its uploaded images and media, and its presentation screenshots.

Three things are removed on request rather than by the delete: your GitHub repository, your presentations’ published files, and the files submitted through your forms. Ask, and we remove them.

6. What Practice records about people

What we don’t record:

• No score, no rating, no pass or fail. Practice is not assessed.
• The criteria a person ticks during a run are not stored at all.
• Self-recordings stay on the person’s own device. We never receive them.
• Nobody but the person assesses the person.

What we do record:

• That a person ran a play, and when. Kept 24 months.
• That the step is complete, in any learning path that includes the play. Kept while the workspace exists.
• Their own answer to “What was hardest?” on that run. Their managers see it, and so does Bridger’s Portal team.
• Their feedback on the play itself. The play’s owner sees it without their name.

7. Questions

Questions about this page, or a request to remove something: legal@bridgercreative.com.