AI Recruiting Academy › Module 2: Finding candidates with AI
The short answer
Candidate rediscovery means searching the people you already have, such as past applicants, runners up and candidates you sourced before, before you go out to the market. AI helps by reading old resumes and notes for meaning, matching them against a new role, flagging who has changed jobs and ranking who to call first.
It is often the fastest and cheapest source of hires, because these people already know you. It only works if your database is reasonably clean, and it needs care with old notes and with people who asked not to be contacted.
Why do recruiters overlook their own database?
Most applicant tracking systems were built to move people through one job, not to find them again for the next one. Search is often keyword only, profiles are frozen on the day someone applied, and the same person may sit in the system three times under different emails. So recruiters go back to LinkedIn, even when a strong candidate is already on file.
Many teams are still finding those people by accident. Gem’s data shows the share of sourced hires who were already in the company’s CRM or ATS rose from 29.1% in 2021 to 44.0% in 2024. Gem sells a recruiting CRM, so it has an interest in the finding, but it matches what many desks see: the database is a real source when you can search it properly.
How does AI rediscover candidates?
| Step | What the AI does | What you check |
|---|---|---|
| 1. Clean | Reads resumes and records, merges duplicates, fills missing fields | That merged records really are the same person |
| 2. Match | Compares every record to the new role by meaning, not just keywords, as in Lesson 2 | That hard requirements are enforced |
| 3. Refresh | Looks up each match’s current role and contact details, as in Lesson 3 | That the update is the right person |
| 4. Rank | Orders matches and explains the fit, including past interview stage | Why each person is on the list |
| 5. Reengage | Drafts a message that mentions how you know each other | Every message before it is sent |
Some ATS and CRM products now do this inside the system. Others work by connecting an AI assistant to your ATS through an API or MCP server, which Lesson 5 of Module 1 explains.
Which past candidates are worth contacting first?
| Who | Why they are worth a call | How to open |
|---|---|---|
| Runners up, often called silver medalists | Already passed your interviews for a similar role | Name the role they interviewed for and what has changed |
| Past applicants for a different role | Already interested in your company | Explain why this role fits them better |
| Sourced candidates who did not reply | Matched a past search; timing may be better now | A short note that does not pretend it is the first contact |
| Former employees | Known performance and quick to onboard | A personal note from someone they worked with |
| Referrals who were not hired | Someone on your team vouched for them | Mention the referrer, with their permission |
SHRM makes the same point in its guide to reengaging top candidates: runners up are one of the quickest routes to a strong hire, because much of the assessment is already done.
What should you fix before you turn AI loose on your database?
Duplicates. If the same person appears three times, they may get three messages. Merge first.
Statuses. Make sure “do not contact”, “hired elsewhere” and “withdrew” are recorded in fields the AI can read, not buried in free text notes.
Retention. Privacy laws such as the GDPR say personal data should not be kept longer than needed for its purpose. If your policy says records are deleted after a set period, rediscovery should not bring back people who should already be gone.
Old notes. Interview notes can contain comments that should never drive a decision, such as remarks about age, accent or family. If the AI reads free text notes, check what it is picking up. Module 4 covers how bias enters AI tools.
What can go wrong with AI rediscovery?
| Risk | Example | Guardrail |
|---|---|---|
| Contacting people who opted out | A candidate who asked to be removed gets a new message | Filter on consent and status fields before matching |
| Repeating an old rejection | A past “not a fit” note sinks someone who has since grown | Rank on the current profile, and read old notes yourself |
| Wrong person merged | Two people with the same name become one record | Spot check merges before any outreach |
| Awkward messages | A cheerful first contact to someone you rejected last month | Always name the history and approve every message |
How do you measure whether rediscovery is working?
Track three numbers: the share of hires that came from people already in your database, the reply rate on rediscovery messages compared with cold outreach, and time to fill for those hires. If the share is rising and time to fill is falling, keep going. If replies are low, the list or the message needs work, not the idea.
Quick answers
Does my ATS already do this?
Many ATS and CRM products now include some form of AI matching against past candidates. Test it on a role you recently filled: did it surface the people you would have chosen?
How old is too old for a record?
There is no single answer. Follow your retention policy first. After that, the older the record, the more you rely on a fresh profile check before reaching out.
Should I tell candidates how I found them?
Yes. Mention when and how you were in touch before. It is more honest, it lands better, and in the EU and UK it helps meet the duty to tell people where their data came from.
Is rediscovery only for large companies?
No. An agency recruiter with a few thousand records in a CRM often gets more from rediscovery than a large employer, because the relationships are more personal.
Explore in the directory
Module 2: Finding candidates with AI
- What is AI sourcing?
- Semantic search vs Boolean search
- What is contact data enrichment?
- How does AI rediscover candidates in your database?
- How do you personalize outreach with AI without landing in spam?