How to automate candidate sourcing with AI
200 profiles read against the rubric you wrote down, 20 shortlisted with the deciding signal named, logged in your ATS with a first message drafted for each.
Profile 140 gets a harsher read than profile 12, for no reason but fatigue. A companion reads all 200 the same way and shows its reasoning per candidate, so you audit the calls instead of trusting them.
It works in your own logged-in LinkedIn session, at human pace and inside LinkedIn’s daily limits, and opens the GitHub of anyone who scores well. The prompts below are the whole pass, from rubric to queued outreach in Strawberry.
Write the rubric down first
Every bad sourcing run traces back to criteria that lived in someone’s head.
Make them explicit, disqualifiers included, and consistency comes free.
Prompt: “Here’s the role: senior backend engineer, Berlin or remote-in-EU, Go or Rust in production, has worked somewhere between 20 and 300 people. Hard yes signals: shipped and owned a service handling real load, open-source contributions I can read. Hard no: no production Go or Rust, or three jobs under 12 months in the last five years. Nice-to-have: fintech or payments. Score each candidate 1–5 against this and always tell me which signal drove the score.”
Where are the candidates?
In the LinkedIn you already pay for, plus GitHub and wherever else the evidence sits. The companion searches from your own session.
Prompt: “Search LinkedIn for profiles matching that role and screen the first 200. For anyone scoring 4 or 5, also open their GitHub if it’s linked and tell me what they actually build. Put results in a new sheet: name, profile URL, current role and tenure, score, the one-line reason, and the GitHub read where you have it.”
It browses at human pace and respects LinkedIn’s daily limits, deliberately.
Audit the 3s before you trust the 5s
Reading ten 3s tells you more about whether the screen works than reading the top ten.
Prompt: “Show me the ten candidates you scored 3 and why. I want to see where the rubric is too tight.”
Correct it in plain language, “Kotlin at scale should count as adjacent, bump those to 4”, and have it rescore. The correction sticks for the next role too.
Who logs the shortlist and drafts the outreach?
The companion does both.
Candidates go into Ashby with their score and reason, and each gets a first message built on something specific it read.
Prompt: “Add every 4 and 5 to Ashby under the ‘Senior Backend (Berlin)’ job with their score and reason in the notes. Then draft a LinkedIn message for each: two sentences, mention the specific project or repo that made them a fit, no ‘I came across your profile’, no adjectives about how impressive they are. Queue them all for my review. Send nothing.”
Same shape as our recruiting workflow: the companion finds and files, you own the first human contact.
How do you keep the pipeline warm?
Put the same search on a weekly routine and new profiles arrive already scored.
Prompt: “Every Monday at 8am, re-run this search for profiles that are new or newly open to work since the last run, screen them the same way, and add anyone scoring 4 or 5 to Ashby. Send me the list with reasons, no drafts unless I ask.”
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents