Local AI for Recruiters: Writing About Candidates Without Leaking Their Data

Recruiting is a writing job disguised as a people job. Outreach messages, candidate summaries for hiring managers, interview feedback, offer emails, rejections. AI writing tools are an obvious help, and many recruiters already use them daily.

The problem is what goes in. A typical recruiter prompt contains a candidate’s name, current employer, salary, notice period, and frank interview notes. That is personal data under every privacy law that matters, and some of it, such as health information disclosed during a reasonable adjustments conversation, is special category data.

Pasting it into a consumer chatbot is a data transfer to a third party that the candidate never agreed to. Doing the same rewrite with a model on your own laptop is not.

The Messages That Benefit Most

Outreach. Turning a generic template into a short, specific message that reads like a person wrote it.

Hiring manager summaries. Condensing a long interview write-up into five lines a busy manager will read.

Candidate feedback. Turning blunt internal notes into constructive, fair feedback for the candidate.

Rejections. Making them kind, clear and quick. See how to write a candidate rejection email.

Offer and negotiation emails, where tone and precision both matter.

Where Care Is Needed

AI in hiring is regulated in a growing number of places, and the rules focus on tools that screen, rank or make decisions about candidates. Using a model to rephrase an email you wrote is a different activity from using one to score CVs, but the line is worth keeping clear:

• Do not ask a model to assess or rank candidates. That is the use that attracts discrimination rules and audit requirements. See US state AI employment rules in 2026.
• Do not let a rewrite change the substance of feedback. “Lacked depth in SQL” becoming “was not a fit” changes what you are telling the candidate.
• Keep protected characteristics out of notes. If they are not in the text, no tool can misuse them.

In the US, the EEOC has made clear that existing anti-discrimination law applies when employers use software in hiring. A rewrite tool does not make a decision, but notes that reveal bias are still evidence.

A Before And After

Turning internal notes into candidate feedback.

Before:

Honestly not senior enough, rambled on the system design question and couldn’t explain tradeoffs, but nice guy and good culture fit, maybe in 2 years

After:

Thank you for the time you put into the interviews. The panel was impressed by your communication and collaboration, and enjoyed meeting you. For this senior role, we were looking for more depth in system design, particularly in explaining the tradeoffs between approaches. We would be glad to stay in touch as your experience grows, and we encourage you to apply for future roles with us.

The second version gives the candidate something specific and usable, drops the “culture fit” phrasing that invites questions, and stays honest.

A Private Workflow

1. Use a local model for anything with a candidate’s details. In Wrivio, switch to the Local engine and download the model once. Rewrites then run on your machine without network calls.
2. Use separate Contexts for outreach, feedback, rejections and manager summaries, each with its own instructions.
3. Strip names when using cloud tools for templates or general phrasing help.
4. Check your ATS and email client for AI features that read candidate data by default, and know where that data goes.
5. Keep notes factual and job-related, which makes every downstream message easier.

A Wrivio Context for candidate feedback could say:

Rewrite these interview notes as feedback to the candidate. Warm, professional, specific and honest. Mention one or two genuine strengths and the main area that decided the outcome. Keep every factual point exactly as written. Do not add praise or criticism that is not in the notes, and do not mention personal characteristics, age, background or culture fit.

Press Ctrl+Shift+Space, paste the notes, and read the result. Watch for feedback that has become vaguer than the notes, which defeats the purpose.

Agency Recruiters And Client Data

Agency recruiters hold two kinds of confidential data: candidates’ and clients’. A role brief that names an unannounced restructure is commercially sensitive. The same rule applies: if it would embarrass your client to see it in someone else’s system, keep the rewrite local. See freelancer client data in AI tools for the same problem from a contractor’s side.

Common Questions

Is it legal to use ChatGPT to write candidate emails?

Rewriting your own email is generally fine, but pasting candidate personal data into a third-party service is a data transfer that your privacy notice and contracts should cover. A local model avoids that transfer entirely.

Does using AI for rejection emails count as automated decision-making?

No, if a person made the decision and the tool only phrases the message. Automated decision-making rules apply when software makes or materially shapes the decision itself.

Can I use AI to summarise CVs?

You can, but be careful: a summary that emphasises some details and omits others shapes how the hiring manager sees the candidate. Use consistent instructions for every CV, and keep the original attached.

Will candidates know I used AI?

They may suspect it if the message is generic. The fix is specificity: mention the role, one real detail, and a clear next step. See why AI written emails get ignored.

Download Wrivio for Windows to write outreach, feedback and offers faster, with candidate data kept on your own machine.