The reply rate on generic recruiter outreach has been in freefall for two years. Not because candidates got ruder. Because inboxes got smarter, and readers can now smell a template within the first six words.
The fix is not another sequence. The fix is writing like a specialist who has been paying attention.
Why templates stopped working
A well-tuned template used to convert at 8 to 12 percent. In 2024 the same template converted at 3 to 5 percent. In 2026 it converts at under 2 percent, and that is before you account for the fact that anything mass-sent to a senior engineer at a Series B is now silently binned by a Gmail rule they wrote themselves.
Three things changed at once:
- Everyone got the same tool. The moment "Hi {first_name}, I saw you were at {company}" became infinite and free, it became worthless.
- Readers learned the shape. A three-sentence opener, a role bullet, a Calendly link. Readers now recognise the silhouette before they read the words.
- Filters got clever. Outlook and Gmail both quietly deprioritise messages that pattern-match to bulk outreach, even when sent one-to-one.
What actually gets read
Reply-worthy outreach in a specialist niche has four properties. Not three. Not five.
- A subject line that reads like a human wrote it to one person. Lowercase is fine. No emoji. No "quick question". Something like: saw your talk at RSNA, one question about the workflow bit.
- A first sentence that proves you read something. Not the company page. Something specific: a talk, a paper, a repo, a hire, a product change. If you cannot find one, you are not ready to send.
- One paragraph on why you are writing. The role, the reason it might be interesting for this person specifically, and the trade-off you already know they will care about.
- A close that asks for a small thing. Not a call. A reply. "Worth me sending the spec?" gets replied to. "Have you got 20 minutes on Thursday?" gets ignored.
The tone problem
Half the AI outreach you see fails not because it is generic but because it is tonally wrong for the reader. A senior clinician does not want the same voice as a founding engineer. A design director does not want the same voice as a partner at a top-tier firm.
Holi rule of thumb, and the one we bake into every draft: match the register of the last thing this person published. If they write short, you write short. If they write academic, drop the exclamation marks. If they write with dry humour, you get one shot at dry humour and you had better land it.
A note on personalisation-at-scale
There is a version of this you can automate badly. Feed a resume and a company page to a model, ask for "personalised outreach", and you get a note that mentions their tenure and their last role, and reads exactly like the notes the other 41 recruiters sent them this month.
Personalisation is not "mentioning a fact about them". It is writing something that could only have been written this week, by someone who understands the niche they operate in. Those are different things. The first is worthless. The second is why the reply arrives.
What good looks like in practice
Here is the structure Holi drafts against, and what specialist recruiters we work with typically send:
subject: quick note on the OR-scheduling piece
Saw your write-up on why the OR-scheduling handoff keeps breaking, the bit about anesthesia being the missing signal was the sharpest framing I have read on it this year.
The reason I am writing: a health system I am working with is rebuilding that exact handoff, and they are looking for a clinical lead who has actually run the workflow rather than just modelled it. Comp is honest, the CMO is technical, and they are not pretending it is a six-month problem.
Not a pitch, just wondering if it is worth me sending the one-pager. Reply "yes" and I will.
Four short paragraphs. One specific signal. One trade-off already surfaced ("not pretending it is a six-month problem"). One small ask.
That gets replied to. The template does not.