Frequently asked questions

When you connect Gmail during setup, Threshold scans your inbox for cold asks you’ve actually been receiving, then tailors a 3-tier paywall to them. After that, every new email is screened in two steps:

1. Skip anyone you’ve emailed before. On setup we walk your sent folder and remember every address you’ve ever written to. Those people skip the paywall automatically and never hit the AI.
2. AI reads the message. For anyone we haven’t seen before, Claude reads the sender, subject, and the message body (truncated to the first ~4 KB) and decides whether it’s a cold ask — a pitch, intro request, “pick your brain.” Below a confidence threshold, the email is left alone.

When a sender actually pays your paywall, that’s the strongest possible signal we classified correctly. Those messages get anonymized — every reference to you, your name, your domain, your employer — and added to a shared training set that tunes the classifier for every Threshold user.

Yes — reading inbound mail is what makes the product work. Threshold uses two Gmail permissions: read (to classify, detect your replies, and improve the AI) and send (to deliver paywall replies on your behalf). We can’t modify, label, archive, or delete anything — we never asked Google for that permission.

What we actually read, and why:

• To generate the paywall. For new cold senders, Claude reads sender + subject + the message body (the first ~4 KB; long pleasantries before the actual ask get missed otherwise) to decide whether to fire your paywall.
• To detect your reply. After a sender pays, we periodically check the Gmail thread for a message you sent. That’s the signal we use to move the booking to “in flight” and (after the 48-hour grace window) capture the held payment.
• To train the AI. When a sender successfully pays a paywall, we anonymize the message body (regex pass for your name/email/domain, then a second Claude pass for nicknames, family or employer references, distinctive titles) and add it to a shared training set used to improve the classifier.

You can read and reply in Gmail like normal — every booking row has an Open in Gmail link that opens the thread directly. Full message bodies are never stored in our database; we re-fetch from Gmail on demand each time and the body lives in memory only.

Per-email metadata only — never full bodies. Specifically:

• Sender address and name
• Subject line
• The short snippet Gmail auto-generates (~200 chars)
• The classifier’s confidence + one-sentence reasoning
• Gmail thread + message IDs (pointers, so we can re-fetch from Gmail later)
• Payment state and Stripe IDs (no card data)

Per-user we store your tier list, charity choice, paywall wording, and your Gmail OAuth refresh token — encrypted at rest with AES-256-GCM, so a database leak alone wouldn’t expose it.

For training data we store the anonymized body + subject of paidpaywalls, with every reference to you scrubbed by a two-pass redactor (regex first, then a Claude pass to catch nicknames, family references, employer hints, and location clues).

You choose, anywhere from 0% to 100%. Set it on the Paywall page with the charity slider. Slide it all the way to 100% and the entire payment goes to charity (Threshold takes no fee). Slide it back and you keep a share, with at least the rest going to your selected charity.

A 16% fee on each captured paywall payment when your charity slider is below 85%. No fee when your charity slider is 85% or higher — Threshold absorbs Stripe processing as marketing.

Each paywall payment splits roughly as: Stripe processing (~2.9% + 30¢), your charity share (you choose), Threshold fee (16% when charity slider is below 85%, 0% when 85% or higher), and the remainder to you. Stripe handles the routing.

Never until you’ve engaged. Stripe holds the payment as an authorization (the card is reserved, not charged) when the sender pays. The charge only completes when Threshold detects your reply in Gmail and the 48-hour grace window elapses — or, for call-tier bookings, when you mark the call done. Until that capture, the sender’s card is never charged and refunding is one click.

The authorization sits in escrow for 21 days. If you don’t engage in that window — no reply in Gmail, no manual action — the authorization auto-cancels and the sender pays nothing. You’re never on the hook for engagement you didn’t want.

Yes. Every paid request has a “Refund (won’t engage)” option in the expanded row on your Home page. The Stripe authorization cancels and the sender’s card is never charged. You can also refund after the grace window started but before capture by clicking “Cancel & refund” on an in-flight row.

Use the Report a mistake page from inside the dashboard. We add the sender to your trusted list so it won’t happen again from that address, dismiss the misclassified request if it’s still pending, and feed the report into the training set so the classifier gets better for everyone.

Threshold treats every email address independently. If someone paid from foo@example.com and later writes from bar@example.com, the new address starts as a fresh cold sender. Reporting the new address (or replying naturally so the not-solicitation path picks them up) adds them to your trusted list.

Three layers, depending on what you do:

• Walk away (no action). If you don’t open the dashboard for 30 days, Threshold automatically stops classifying new mail and stops sweeping for replies on your behalf. Existing data sits inert. The moment you load any page again, processing resumes.
• Revoke Gmail access. If you revoke our OAuth grant in your Google account, we detect it on the next Gmail call and purge your stored OAuth tokens automatically.
• Delete your account. Email hello@thresholdmail.com and we’ll honor the lifecycle of any in-flight paid request (capture or refund as you choose) and then purge your user record, requests, and known-senders cache. Anonymized training examples remain — by the time they exist, every reference to you has been scrubbed.