Last updated: June 5, 2026
A lot of AI products talk about themselves in marketing language and then describe the actual mechanics in three buried sentences inside the privacy policy. We do not want to do that. This document explains, in plain English: what AI is doing inside HireKey, what data the AI sees, who the AI providers are, whether your content is used to train models (it is not), whether humans can see your sessions (rarely, and only in defined cases), what the AI cannot reliably do, and how we make decisions about AI features.
We built HireKey to help candidates prepare more confidently and realistically. The AI is a coach and an editor, not a ghostwriter for your life. We pick the model that fits the task, not the most expensive one. We do not pretend the AI is more capable than it is. We do not feed your private content into a training loop. AI output is presented as a draft to be edited, not as a verdict. Nothing is auto-submitted on your behalf. Every AI output goes to you before it goes anywhere else.
Features that use AI: Resume tailoring and ATS scoring, voice interview coaching, written interview practice, cover letter generation, company briefings, salary benchmarks, and resume builder.
Features that do NOT use AI: Account management, calendar syncing, application tracking, the job board, and search.
Anthropic (primary). All reasoning, writing, and coaching features use Anthropic's Claude models, processed on Anthropic's infrastructure in the United States.
| Feature | Model |
|---|---|
| Voice interview coaching | Claude Sonnet (claude-sonnet-4-6) |
| Resume tailoring | Claude Sonnet (claude-sonnet-4-6) |
| Cover letters, written interview prep, STAR evaluation, resume builder, company briefings, salary benchmarks | Claude Haiku (claude-haiku-4-5-20251001) |
OpenAI (text-to-speech only). The voice interview coaching feature uses OpenAI's text-to-speech API to convert coaching responses into audio. OpenAI does not receive your voice input or your resume; it receives only the text it needs to speak aloud.
| Feature | What we send | What we do NOT send |
|---|---|---|
| Resume tailoring | Your resume text, the job description, any tailoring instructions | Your email, payment info, account ID |
| ATS scoring | Your resume text, the job description | Same, plus your other resumes |
| Voice interview coaching (Anthropic) | Your voice (transcribed to text), the job context, our coaching prompt | Other users' data, your unrelated resumes |
| Voice interview coaching (OpenAI TTS) | The text of the AI's coaching response, to convert to audio | Your voice input, your resume, your account info |
| Cover letter | Your resume, the job description, your tone preference | Other letters you have generated |
| Company briefing | The company name; may also include recent web search results | Your resume or interview history |
| Salary benchmark | Role title, location, years of experience, optional industry | Your specific compensation history unless you choose to include it |
We do not send your account email or payment information to any AI provider.
Under our contracts with Anthropic and OpenAI, your content sent through HireKey is not used to train their public models. Our full commitments: we do not use your uploaded resumes, interview responses, voice recordings, or transcripts to train public AI models. We do not sell your AI inputs or outputs to third parties. When we change AI providers or significantly change what data we send them, we will update this policy and notify users for material changes.
Almost no one, almost never. Engineers can technically access production systems, but routine review of user content is not part of any engineering workflow; access is logged. Support staff can read sessions only when you have asked for help. An engineer may look at a specific session to investigate a bug you reported. An authorized team member may review session data during abuse investigations. We do not have a team doing routine quality assurance by reading interview transcripts.
When the AI is wrong, it is sometimes confidently wrong. Treat AI output the way you would treat advice from a smart friend who has read a lot but does not work in your industry: take what is useful, edit what is not, and apply your own judgment.
We use AI providers that publish model cards and have demonstrated investment in reducing bias. We design prompts to focus on skills and impact, not on demographic or biographical signals. We test new prompts on a curated set of resumes from a range of backgrounds and career types before shipping. Users can flag biased or unfair output from inside the product. We do not claim to have eliminated bias. We claim to be paying attention to it.
| Data type | Stored by HireKey? | Sent to AI provider? | Used to train public models? |
|---|---|---|---|
| Resume uploads | Yes | Yes (Anthropic, relevant features) | No |
| Interview practice responses (text) | Yes, while account active | Yes (Anthropic) | No |
| Interview practice responses (voice) | Audio retained briefly unless saved. Transcript: while account active. | Audio transcribed; text sent to Anthropic. TTS response sent to OpenAI. | No |
| Job descriptions | Yes | Yes (Anthropic) | No |
| Account info (email, name) | Yes | No | No |
| Payment information | No (handled by Stripe) | No | No |
| Usage analytics | Aggregated (PostHog) | No | No |
Questions about AI in HireKey: privacy@hirekey.io