Twenty-five candidates, twenty-five different first sentences

What is the exact prompt?

It takes a candidate list as input, an ATS view or a sheet, and produces staged drafts:

• "Here are 25 candidates for the [role] role: [list / sheet / ATS view]. For each, spend a few minutes on their public work in my logged-in session (profile, portfolio, GitHub, talks, writing) before drafting anything."
• "Then write a first message per candidate, three sentences: (1) one specific observation about something they made or wrote (named, not vague); (2) why this role connects to that specific thing; (3) a plain ask for 15 minutes, with two concrete time windows."
• "Banned: 'I came across your profile', 'impressed by your background', any compliment that could apply to a stranger, and any claim about our company that is not on our careers page. If you cannot find one specific thing about a candidate, put them on a skip list with a note instead of writing generic copy."
• "Stage all drafts for my review in the ATS or sheet. Send nothing, connect with no one, and stop at 25 today regardless of how fast it goes."

Why is the skip list the most important output?

It is the anti-slop mechanism.

A drafting system without a skip list degrades to templates exactly when a candidate is hard to research, which is when a generic message does the most damage. "I could not find anything specific, so I did not pretend to" is what separates this from the mail-merge it replaces.

The skip list is also honest feedback on your sourcing. If 15 of 25 are unresearchable, the list was built on title-matching, which candidate sourcing fixes upstream.

Review the drafts, then follow up on day three

You read 25 drafts in about ten minutes, checking one thing per message: is sentence one actually specific? Kill anything that survives a name-swap. Expect to send roughly twenty and edit three. Email goes through the native Gmail or Outlook integrations, where sends pause on an approval card, and LinkedIn messages go out by your hand, one at a time.

Follow-ups are their own small prompt three business days later: "for each candidate in [sequence] who has not replied, draft one follow-up that adds a new piece of information (a team blog post, a relevant product change) never a naked bump." One follow-up, then stop. Greenhouse users keep the whole sequence state in the ATS through the Greenhouse integration.

Check these before running it at volume

Reply rate against your own baseline after 50 sends, not vibes after five.

If researched three-sentence messages are not beating your previous sequences, the role pitch or the list is off. Spot-check the factual claims weekly too, since a companion summarizing a candidate's repo can misread what the repo does.

And keep the cap. Raising 25 to 250 re-creates the spam machine with better prose, and on LinkedIn it leaves session-pace use entirely. This playbook wins on hit rate, the same trade the recruiters using agents page describes across the hiring funnel.

Frequently asked questions