Pipeline
What to Put in a Cold Message So It Gets Answered
A cold message works when three sentences carry something only this person or team could hear, and it fails the moment it reads as a template, because a recruiter who gets the same paragraph a dozen times a week spots it before finishing the first line. Know what a referral does that a cold message can't, which is move you out of the application queue rather than up it, ask for something small enough to answer in one line, and skip the message when you have nothing specific to say.
The take"Stop applying, start reaching out" has been repeated so often it stopped meaning anything specific, and a chatbot happily fills in the blanks of whatever template that advice implies. The advice was never wrong. What made it work was always the part a template can't supply: a real, specific reason you're writing to this particular person. Skip that part, and a model-assisted message is just a faster way to be ignored by more people at once.
Where Olive fits
Open a role and see what the work shows
A cold message can move you up a queue at best. Olive sits elsewhere in the same funnel, as six evidenced findings written about one candidate, and any report it produces is shown to you as well as to the employer, free of charge.
Rank your shortlistWhy does a template read as a mass send before the first line ends?
A recruiter who has seen the same opening line a dozen times this week recognizes it instantly, and the recognition happens before your actual point arrives. A model returns the median instance of whatever genre you ask it for, and the entire value of a cold message was always in being non-median. Ask it for a generic outreach message and it hands you the same generic outreach message everyone else is also sending.
- The tell isn't AI. It's the shape: no name, no specific project, an ask that could go to anyone in the industry.
- What survives. One line that could only be about this team or this person, because you actually looked.
The volume behind this is real, and employers are told nearly the same thing from their side of it: why applications per opening tripled this year, and how many of them are real walks through the arithmetic a candidate is competing inside. Candidates submitted 356 million applications on Workday's recruiting platform in 2024, up 26 percent, while the jobs its customers posted grew 7 percent 1. Reaching out was supposed to be the escape hatch from that math, and in a Greenhouse survey of 2,900 job seekers, nearly half said they were sending more applications than a year earlier 2.
That is the exact pressure that turns a genuinely good idea, writing a real person a real message, into another mass-produced format. The idea still works. It just stopped working as a shortcut you could run at scale, because scale is what broke the resume pile in the first place, and a recruiter's inbox is downstream of the same math.
What does a message actually need to say?
Three sentences: who you are in one line, the specific thing about this team or role that made you write, and what you're asking for. Nothing in that structure requires length, and length is usually where a template hides its own emptiness. A recruiter reading a hundred of these a week rewards the one that gets to the specific line fastest.
- Line one: who you are, in a phrase, not a paragraph.
- Line two: the specific thing you noticed, named precisely enough that it couldn't be about a different team.
- Line three: the ask, small enough to say yes to in one reply.
The most common uses of a model in a job search are already this shape. In a Greenhouse survey of more than 2,200 workers and job seekers, interview prep and analyzing a posting for skills to highlight were the two most common, ahead of having an agent apply on a candidate's behalf 3. The honest use of a tool here matches that pattern: sharpen a draft you already wrote, don't outsource the noticing that made it worth writing.
A useful check before sending: read the message once as if you were the recipient, and ask whether it tells you anything you couldn't have guessed about the sender. If the answer is no, the specific line hasn't landed yet, no matter how polished the sentence around it has become.
Ask for something small enough to answer in a line
The size of the ask decides whether answering costs the recipient anything. "Can you tell me one thing about how the team actually uses AI day to day" can be answered in one line from a phone. "Can we hop on a call this week" asks a stranger for a block of their calendar on a first message. Match the ask to the relationship you actually have, which for a cold message is none yet.
- Too big. A call, a coffee chat, a referral, on message one.
- Right-sized. One specific question a person can answer in two sentences from their phone.
A referral and a cold message are not doing the same job, and it helps to be honest with yourself about which one you're sending. A referral moves you out of the queue entirely; a well-written cold message, at best, moves you up within it. Neither replaces the application itself, and neither one is owed a reply just because you sent it.
Knowing which one you're actually attempting changes what a good outcome looks like. A cold message that gets a two-line answer and nothing more has still done its job: it put a name to a face before an application landed in the same queue as everyone else's. Treat that as a win, not a stall, and you'll send fewer messages out of disappointment and more out of an actual reason.
When should you not send a message at all?
Skip it when you don't have a real, specific reason for writing to this person, because a message built entirely on hope reads exactly like the templates it's trying not to sound like. Skip it too when the role is closed, when you've already messaged this same person about a different role in the last month, or when your only reason for writing is that you found their email address.
- No specific reason. Don't send it. Apply through the normal path instead.
- Recently messaged the same person. Wait. Don't repeat the ask.
- Nothing to ask for. A compliment with no question isn't a message. It's noise.
The same test that catches a generic cover letter catches a generic cold message: would this line survive being sent to someone else at a different company? Why an AI-written resume ends up sounding like everyone else's is the same convergence problem wearing a different format. Fix it the same way, with one fact only this message could contain.
Send a true message, to someone you have an actual reason to write to, asking for something they can answer without much cost to their day.
Common questions
Is it weird to message someone I've never met?
No. A cold message to a recruiter or a hiring manager is a normal thing to send. What reads badly isn't the fact of it, it's a cold message with nothing specific in it. One real, particular reason for writing carries the whole thing on its own.
Should I ask AI to write the entire message for me?
Write the specific line yourself first, the thing you actually noticed about this team or role, since that's the one part a model can't supply. Then let it tighten the sentences around it. A message built the other way around, generated first with specifics inserted after, reads exactly like the templates that don't get answered.
How many people should I message for one application?
One, maybe two, chosen because you have an actual reason to write to them, not because you found five emails on a company page. Messaging everyone at a company about the same opening reads as an obvious mass send, whatever the wording says.
What if I don't get a reply?
Most cold messages don't get one, including good ones; that's the nature of asking a busy stranger for their attention. Send it because it costs little and occasionally works, not because a reply is owed to you. One follow-up after a week or two is reasonable. More than that reads as pressure.
Does a bulk-messaging tool speed this up safely?
It speeds up the sending, not the part that gets read, which is the specific line only a real look at the team could produce. Scale is also the thing that broke this format: the more identical the paragraph, the faster it reads as one. Check the terms of any platform before automating messages on it, since the rules on bulk sending are set by each service and change.
References
- 1. How HR Leaders Can Thrive in a Complicated Job Market workday.com Sizes the gap between how fast applications grew and how fast openings grew, the math outreach was meant to escape.
- 2. An AI Trust Crisis: 70% of Hiring Managers Trust AI to Make Faster and Better Hiring Decisions, Only 8% of Job Seekers Call it Fair greenhouse.com Supports that job seekers are sending more applications than a year ago, the same volume pressure behind outreach messages.
- 3. Greenhouse 2025 workforce and hiring report cdn.prod.website-files.com Shows the most common AI uses in a job search are preparation and analysis, not automated outreach, supporting the recommended use of a tool here.
3 sources, numbered by first appearance. How Olive sources claims
General guidance for hiring teams. What works at one company and one volume may not transfer to yours.
Olive assesses how a person works with AI. It does not detect AI-written documents, and it never produces a score, a ranking, or a match percentage for a person. Candidates read the same report the employer reads.