AI note taker security questionnaire: questions to send before you buy
On this page
An AI note taker security questionnaire asks the vendor to state, in writing, where interview audio, transcripts and summaries go, who can read them, how long each is kept, whether any of it trains a model, which subprocessors touch it, and how you delete one candidate's data. Send it before a trial uses real candidates, and compare the answers with the vendor's contract, subprocessor list and assurance reports. Interviews raise the bar because the person whose words are captured, the candidate, never agreed to the vendor's terms.
Below: what to fill in before you send it, the questionnaire itself in eleven numbered sections, what good and bad answers look like, a twelve-question short form for small teams, and how to read what comes back.
Before you send it: write down your own context
Vendors answer better when they know what you need. Put this at the top of the email or form, filled in:
OUR CONTEXT
Who we are: [agency / in-house team / hiring managers], [size]
What we will capture: [phone screens / video interviews / panels / in person]
What we need kept: [transcript / summary / scorecard / audio / video]
Where we hire: [US states / UK / EU countries / other]
Our retention policy: [e.g. interview records 2 years after decision]
Our ATS: [name]
Contact for answers: [name, email] Answers needed by: [date]If you do not yet have a retention policy, set one first; the candidate data retention policy template has a schedule you can adapt. Otherwise you will not know whether the vendor's retention answer is acceptable.
The questionnaire
1. Data flow and capture
- Describe, step by step, what happens to the audio from the moment an interview starts until the notes are finished. Which steps happen on our device, and which on your servers or a third party's?
- Does anything join the meeting as a participant? What name does it show, and what do candidates see or receive?
- Is audio or video stored as a file? If so, where and for how long? If not, is it processed in transit only?
- Is any message, summary or link sent automatically to meeting attendees, including candidates? Can we switch it off?
- Does the product create voiceprints, face templates or any other biometric data, for example to recognize speakers?
- Does any feature infer emotion, sentiment, honesty or personality from voice, face or text?
2. Storage and location
- In which countries, and with which hosting providers, are audio, transcripts, summaries, scores and backups stored?
- Can we choose the storage region? Is that included in our plan?
- Are search indexes, embeddings or caches derived from our transcripts stored separately, and do they follow the same deletion rules?
- Is any interview data stored on interviewers' devices? If so, how is it protected?
3. Retention and deletion
- What is the default retention period for each data type (audio, transcript, summary, scores, logs, backups)?
- Can we set our own retention periods by data type? Is deletion automatic when they expire?
- Can we delete all data about one named candidate across every interview, without deleting other records? How long until it is gone from backups?
- Can we place a legal hold that suspends deletion for specific candidates or roles?
- What happens to our data when an interviewer leaves, and when our contract ends? In what format can we export it first?
4. AI models and training
- Which AI models process our data (speech-to-text, summarization, scoring)? Are they yours or a third party's?
- Is our data, including audio, transcripts and edits, used to train or improve any model, yours or a provider's? Is that off by default, and is it in the contract?
- Do third-party model providers retain our inputs or outputs? For how long, and for what purposes?
- Do your staff or contractors review or annotate customer transcripts for quality or training? Under what controls and with whose consent?
- Does the product produce scores, rankings or recommendations about candidates? Can those features be switched off while keeping transcription?
5. Subprocessors
- Please send your current subprocessor list, with each subprocessor's purpose, location and the date of the list.
- How will we be notified of new subprocessors, how much notice will we get, and can we object?
- For subprocessors outside our region, what transfer mechanism do you rely on?
6. Access control
- Which of your employees can access customer interview data, under what circumstances, and how is access approved and logged?
- Within our account, can we restrict who sees which interviews, for example so interviewers see only their own and recruiters see only their desks?
- Do you support single sign-on and multi-factor authentication? Can we enforce them?
- Can we see an audit log of who viewed, exported or deleted interview records?
7. Encryption and infrastructure
- How is data encrypted in transit and at rest? Who manages the keys?
- How are production environments separated from development and testing? Is customer data ever used in testing?
- How often are penetration tests run, by whom, and can we see a summary of the latest results?
- For desktop or mobile apps: how are updates signed and delivered, and what data is stored locally?
8. Incidents and continuity
- How quickly will you notify us of a security incident affecting our data, and through which channel?
- Describe any security incident in the past three years that affected customer data, and what changed afterwards.
- What is your backup and recovery approach, and how long could the service be unavailable in a serious outage?
9. Assurance and documents
- Which independent reports or certifications do you hold (for example SOC 2 Type 2, ISO/IEC 27001)? Please send the report and say which products and features are in scope.
- Please send your data processing agreement and your security documentation.
- Have you assessed your product against AI rules that apply to hiring, such as New York City's Local Law 144 or the EU AI Act? What did you conclude, and on what basis?
10. Candidates and consent
- What does the product do to help us tell candidates and get their agreement before capture?
- How do we stop capture instantly if a candidate declines, and what is kept from the seconds before?
- How do we find and export all data about one candidate to answer an access request?
- Can candidates or other non-customers ever access recordings or transcripts through your product?
11. Exit
- On termination, how do we export everything, and in what formats?
- When is our data deleted after termination, including backups, and will you confirm deletion in writing?
What good answers look like
| Area | A good answer | A red flag |
|---|---|---|
| Data flow | A diagram or numbered steps naming each system and location | "Your data is secure in the cloud" |
| Storage | Countries and providers named, matching the subprocessor list | No country named, or a list that contradicts the answer |
| Retention | Defaults per data type, configurable, with per-candidate deletion | "We keep data as long as your account is active" |
| Training | A contract clause saying customer data is not used to train models, covering subprocessors | "We may use anonymized data to improve our services" with no definition |
| Subprocessors | A dated public list with purposes and locations, and a notice period for changes | "Available on request" and then never sent |
| Access | Named roles, approval, logging, and customer-side permissions | "Only authorized staff" with no detail |
| Assurance | A current report whose scope includes the product you are buying | A certificate for a different product or entity |
| Scoring and emotion | Scoring optional and documented; no emotion inference | "Sentiment" or "confidence" scores on candidates |
How to read the answers
- Check claims against documents. The storage answer should match the subprocessor list. The training answer should match the contract. Where they differ, the contract wins.
- Read the assurance report, not the badge. Scope, period and exceptions are what matter; SOC 2 for recruiting software shows how to read one quickly.
- Put the commitments in the DPA. Retention, deletion, subprocessor notice and incident timelines belong in the contract; see data processing agreements for recruiting software.
- Map where the data goes. Each answer adds a location to your data map; where AI note takers store interview data has a template.
- Test what you can. Deletion of a test candidate, export formats and access permissions can all be checked during a trial, as in the AI note taker evaluation checklist.
A short form for small teams
If a 43-question form is more than a two-person agency can process, send these twelve and keep the replies on file.
Subject: Security questions before we use [product] for interviews
1. Where (countries, hosting providers) are audio, transcripts and
summaries stored?
2. Is audio kept as a file? For how long?
3. What are the default retention periods, and can we change them?
4. Can we delete all data about one candidate? How long until it is
gone, including backups?
5. Is our data used to train or improve any model, yours or a
third party's? Where is that stated in the contract?
6. Which subprocessors handle our data, for what, and where?
7. Who at your company can access our interviews, and is it logged?
8. Do you support SSO and enforce MFA?
9. How are data in transit and at rest encrypted?
10. How fast will you tell us about an incident affecting our data?
11. Which independent report (e.g. SOC 2 Type 2) can we read, and is
this product in scope?
12. Does any feature score candidates or infer emotion, and can it be
switched off?The answers also help with candidate questions later. When someone asks what happens to their interview, you can tell them accurately, and when they ask for a copy, you will know where to look; see candidate data access requests.
Questions people ask
What security questions should I ask an AI note taker vendor?
Ask where interview audio, transcripts and summaries are stored and in which country, how long each is kept and how you delete one candidate's data, whether your data trains any model, which subprocessors handle it, who at the vendor can access it, how it is encrypted, how incidents are reported, and which independent assurance reports you can read.
Do AI note takers use interview recordings to train their models?
It depends on the vendor, and sometimes on your plan or settings. Policies differ, and interview data may also pass to AI model providers that have their own terms. Get the answer in writing, for the vendor and for each AI subprocessor, and make sure the contract says the same.
Is a SOC 2 report enough to approve an AI note taker?
No. A SOC 2 report tells you whether the vendor's controls were designed and, for Type 2, operated as described over a period, within the scope it covers. It does not tell you where your candidates' data is stored, how long it is kept, whether it trains models, or whether the scope includes the AI features you will use.
Why are interviews different from other meetings for security reviews?
The person whose data is captured, the candidate, is not your customer or employee and never agreed to the vendor's terms. Interviews also contain sensitive details people mention in passing, and they are hiring records with retention and disclosure duties. That raises the bar on consent, retention control and per-candidate deletion.