The short answer
Resumes, contact details, interview notes and recordings, references and assessments are personal information about candidates who never agreed to AI training: leave them out. What a recruiting agency might license is process with the people removed, such as pipeline stage timings, its own outreach templates and screening rubrics, checked against its candidate privacy notice and client terms.
On this page
Of all agency types, recruiters hold the archive that looks most valuable and is hardest to license. Years of searches, thousands of conversations, structured pipelines with dates at every stage. Nearly all of it is about people.
Whose data is it?
A candidate who sends a resume is sharing it to be considered for a job. The interview notes, salary expectations and reference calls that follow are about that person. The agency holds them, but licensing them to an AI buyer is a new purpose the candidate never agreed to, and in many cases never imagined.
That is the core problem, and no amount of de-identification fully solves it. A career history with employers, dates and a city can identify someone without a name. Interview notes are free text, the hardest kind of record to scrub reliably.
What the buyers say
The buyers have drawn their own lines. Scale AI turns away staff and customer identities, credentials and raw system access. micro1 promises it will not pass data on until identifying details are anonymized. Both statements describe what the buyer does once data is in hand; neither makes it safe for an agency to hand candidate files over in the first place.
Off limits
| Record | Why it stays out |
|---|---|
| Resumes and candidate profiles | Personal information; identifiable even without a name |
| Contact details | Personal information with no training value worth the risk |
| Interview notes and scorecards about a person | Opinions about an identifiable individual |
| Video and audio interviews | Faces and voices can be biometric data under Illinois’ BIPA, Texas’ CUBI and Washington’s biometric law |
| References and assessments | Third parties’ statements and test results about a person |
| Salary histories and offers | Sensitive, and often also the client’s confidential information |
Why “anonymized” is not a way out
It is tempting to think that removing names, emails and phone numbers turns candidate records into process data. It does not. A note that says a senior engineer left a well-known fintech after four years, wants a remote role and turned down an offer last spring identifies a person to anyone in that market. Free text resists clean redaction, and a buyer’s later anonymization cannot undo a disclosure you already made.
The client side
Recruiting agencies also carry client secrets: unannounced roles, reorganizations, pay bands. Retained search agreements commonly treat these as confidential, and some assign search materials such as shortlists to the client. A job specification written from a client’s confidential hiring plan is not yours to license just because you typed it.
What might be licensable
Process, with every person and client removed:
- Stage-by-stage pipeline timings: how long searches take from brief to offer, by role type, without names or employers.
- Outreach sequences and screening rubrics you wrote yourselves.
- Search playbooks, recruiter onboarding material and internal SOPs.
- Role specifications rewritten so that no employer, team or pay band can be traced.
Even these need checking against two documents. Your candidate privacy notice records what you told people you would do with their information; if pipeline data is derived from their records, the notice matters. Your client terms decide whether role details are yours.
Is it worth it?
For many recruiting agencies the honest answer is that the licensable part is smaller than it first looks. SOPs and templates are clean but thin. Pipeline timings can be valuable, but only if your tracking system holds them in structured form and you can strip identifiers with confidence. None of the four buyers whose terms we track names recruiting records specifically, though micro1 lists process documentation and operations records and Scale AI lists CRM and HR records.
Ownership, or a right to license, is the entry ticket. Take personal details out before a sample goes anywhere, and ask your lawyer to read the license with the candidate notice open beside it.
If the agency is winding down
Recruiting agencies that are closing face the same rules with more pressure to sell quickly. A closure does not release candidate data from what you told candidates, and client confidentiality obligations often survive the end of the contract. If anything, a wind-down calls for more care, because nobody will be around to answer a candidate’s question about where their records went. Delete what you are obliged to delete, and license only the process records you would have been comfortable licensing as a going concern.
A short checklist before any conversation
- Reread the candidate privacy notice for every year of records in scope.
- List which client agreements treat role details or shortlists as confidential or client-owned.
- Confirm your tracking system can export pipeline stages without names, employers or free-text notes.
- Decide in advance that resumes, notes, recordings and references are out, so nobody negotiates them back in.
Related reading
The full guide to what can and cannot go is on recruiting agency data. The wider picture, from ownership to published terms, is on the front page about how to sell agency data.
For a recruiter, the candidates’ trust is the asset; every licensing decision should leave it intact.