| Quick facts | |
|---|---|
| Do buyers want tickets? | Yes — Polyshares uses a misdiagnosed-then-resolved ticket as its example of value[1] |
| Is your help desk already training on them? | Zendesk trains non-generative models on sanitized data; HubSpot may train unless you opt out[2][3] |
| Export format to ask for | JSON, to keep full comment threads[4] |
| The hard part | Free-text bodies written by customers |
Section 1
The short answer
Frequently yes, once the customers are taken out. Polyshares makes the point in its own pitch: “A support ticket that shows a problem being misdiagnosed, argued over, and finally resolved is worth more than a clean table of finished numbers.”[1] Support threads are a record of diagnosis under uncertainty — which is what customer-service agents are being trained to do.
The problem is authorship. Half of every ticket was written by a customer, under your privacy policy, and often contains order numbers, addresses, and sometimes health or payment detail.
Section 2
Check what your help desk already does with them
Before you license tickets, find out whether your vendor is already training on them — a buyer will ask whether the same data has been used elsewhere.
- Zendesk says its proprietary machine-learning models, which are non-generative, may be trained on customer data after identifier fields are excluded and PII is removed from free text; its generative features use pre-trained third-party models that are not trained on customer data.[2]
- HubSpot says it may use customer data to train and improve its own AI models, and that Super Admins can opt out in the account’s AI settings.[3]
Neither changes who owns the tickets. Both belong in your disclosure to a buyer.
Section 3
Getting a usable export
Zendesk exports tickets, users and organizations as JSON, CSV or XML.[4] Ask for JSON: CSV flattens comment threads and loses the back-and-forth that carries the value. Export users and organizations separately so you can pseudonymize requesters consistently across tickets.
Section 4
A de-identification checklist for tickets
- Replace requester and agent IDs with stable pseudonyms.
- Remove email signatures, phone numbers, addresses, order and account numbers from bodies.
- Drop attachments, or review them separately.
- Exclude ticket categories that routinely carry health, payment or legal detail.
- Sample and hand-review before delivery; keep the review record.
NIST’s guidance is the reference for why step two matters more than step one: identity leaks through combinations of quasi-identifiers in free text.[5] California’s statute sets the bar for a “deidentified” consumer record[6], and your own privacy policy sets what you promised. FTC staff have warned against quietly rewriting that promise after the fact.[7]
Section 5
What ticket data is worth
No buyer publishes a price for support tickets on its own. The published figures are for a company’s licensed data as a whole: micro1 advertises bands from $100K–$500K for a qualified partnership up to $3M+ for highly unique data[8]; Polyshares says its deals typically land between $100K and $2M[1]; Mercor publishes no price list and says compensation depends on data volume, the tools you connect and the depth of the data.[9]
Frequently asked questions
Can I sell customer support tickets to AI companies?
Does Zendesk train AI on my tickets?
Can I stop HubSpot training on my data?
References
Sourced from primary industry reports, government data, SEC filings, and peer-reviewed valuation research. Retrieved at time of publication.
- 1.Polyshares — Sell your company’s operational data — program page — A buyer’s own published deal size, terms, payment timing and anonymization statements; marketing claims, read 16 November 2026 (read November 2026)
- 2.Zendesk, Inc. — Zendesk AI Data Use Information — Edited 7 July 2025; proprietary non-generative models may be trained on sanitized customer data, generative features are not (2025)
- 3.HubSpot, Inc. — Understand HubSpot’s AI model training — Last updated 8 September 2026; HubSpot may use customer data to train its own models and Super Admins can opt out (2026)
- 4.Zendesk, Inc. — Exporting ticket, user, or organization data from your account — JSON, CSV or XML exports of tickets, users and organizations
- 5.U.S. National Institute of Standards and Technology — NISTIR 8053 — De-Identification of Personal Information — Reference treatment of de-identification techniques and re-identification risk (2015)
- 6.State of California — California Consumer Privacy Act, Civil Code § 1798.140 — definitions of "deidentified" and "sale" — Statutory standard a de-identified operational record must meet, and the breadth of "sale"
- 7.U.S. Federal Trade Commission — AI (and other) Companies: Quietly Changing Your Terms of Service Could Be Unfair or Deceptive — FTC staff position that retroactively broadening data-use terms can be an unfair or deceptive practice (2024)
- 8.micro1 — Monetize your company’s operational data — Company Data Partnerships program page — A buyer’s own published payout bands, eligibility criteria, payment terms and data-handling commitments; marketing claims, not audited outcomes (read November 2026)
- 9.Mercor — Mercor Data — Monetize your data — A buyer’s own description of its connector-based extraction, compensation basis and payment timing; read 16 November 2026 (read November 2026)
- 10.Business Insider — Micro1 is making a $1 billion bet on AI’s latest data craze — Seller interviews, the range micro1 says it pays, and a Gartner estimate of AI data spending, 26 October 2026 (2026)