What AI data buyers redact from support tickets
Every ticket buyer we track says on its own page that it de-identifies or scrubs personal data. Do the removal yourself first anyway: customer and agent names, emails, phone numbers, addresses, order and account numbers, card digits and attachments, so nothing identifying leaves your systems.
Buyer pages last checked 7 November 2026
Redaction checklist for a ticket export
- Customer names in the requester field, the greeting, the sign-off and quoted replies.
- Agent names and internal handles, replaced with a role.
- Email addresses and phone numbers anywhere in the body or signature.
- Postal addresses, including partial ones in shipping complaints.
- Order, account and invoice numbers, plus tracking numbers.
- Card digits, even the last four, and any bank details.
- Health details customers volunteer, removed or the ticket dropped.
- Attachments and screenshots, left out by default.
- Custom fields holding identifiers your product passes in.
- Links and hostnames pointing into private systems or accounts.
- Credentials pasted into technical tickets: keys, tokens, passwords.
- A hand-checked sample of the redacted output before transfer.
What do buyers say they remove?
Paraphrased from the buyers' own sites as of 7 November 2026. Not one suggests you can skip redacting on your side.
| Buyer | What its page says about removing personal data |
|---|---|
| micro1 | micro1 runs the de-identification itself, describing processes for transforming sensitive information. Sellers can look over representative samples, and some datasets are rewritten synthetically. Where the work happens is not stated. |
| Scale AI | Anonymization happens on the seller's side, within its own systems, ahead of any extraction. PII is stripped and every data package needs the seller's sign-off before Scale receives it. |
| Labelbox Prism | Prism takes on ingestion and normalization and scrubs PII on the Labelbox side; the page does not say where. |
| Nyne | Nyne promises de-identification before any release without saying which party carries it out. |
Fee note: a buyer in this table may pay Sell Support Tickets if our introduction ends in a license with you. The quotes above come from their pages unchanged regardless.
Sources behind this page
- micro1, micro1.ai/data, checked by Sell Support Tickets on 7 November 2026
- Scale AI, scale.com/data-partnership, checked by Sell Support Tickets on 7 November 2026
- Labelbox Prism, prism.labelbox.com, checked by Sell Support Tickets on 7 November 2026
- Nyne, nyne.ai/sell-company-data, checked by Sell Support Tickets on 7 November 2026
Who decides what you can license?
Your own documents do. Your privacy policy and your customer terms set what you told customers you would do with their messages. If they do not cover licensing support conversations for AI training, a buyer's de-identification promise does not fix that. Read them before scoping, and have your own lawyer read any agreement.
You also need to own the data or hold the right to license it. Tickets you handled for a client under an outsourcing contract usually are not yours to sell. Customers' personal data stays out unless your contract and the law allow otherwise.
Questions about redaction
If the buyer de-identifies the data, why should we redact first?
Because until it is removed, the personal data is yours to protect. Removing it before transfer means a mistake on the buyer's side cannot expose your customers, and it keeps the conversation with your own privacy policy simple.
Are agent names personal data too?
Yes. Your agents are people, and their names, sign-offs and internal handles should come out with the customers'. Replace them with a role, such as "Agent".
Can we keep the customer's company name in B2B tickets?
Treat it as identifying. A company name plus a ticket about a billing dispute points at a person quickly. Replace it with an industry or size band if the buyer needs context.
What do we do with tickets that mention health or card details?
Remove the details, or drop those tickets entirely. Expect every buyer to raise them early, and keep them out of anything you deliver.
Is placeholder replacement better than deleting?
Usually. Swapping a name for a consistent token such as CUSTOMER_1 keeps the conversation readable, which is what makes it useful for training, while removing who it was.
Keep reading
Get a quote
Start with the head office ZIP, an inbox for our reply, then a handful of taps describing the helpdesk history. A person here reads it and writes back naming any buyer whose published entry rules your company clears, or telling you plainly that none does.
Describe the archive; never paste it. No tickets, exports, screenshots or sample files belong in this form, and this site has nowhere to upload them.
Said once, plainly: Sell Support Tickets makes introductions, and the buyer on the other end may pay us for one. More on the how this works page.
Received. Nothing has gone to any buyer.
- We set your answers beside every buyer's published entry rules (headcount, years, language, location).
- An email follows with any names that fit, each with a link to that buyer's terms so you can check our reading.
- Your company stays private unless you answer that email agreeing to an introduction.