What AI Actually Changed About Outbound Recruiting
Personalisation at scale became free, which means it stopped working. What still gets replies is the part AI cannot generate.
Out of every 100 candidates contacted about a well-paid role with real prospects, roughly 40 respond to a phone call, about 25 respond on WhatsApp and 12 to 15 reply to an email. Those numbers are the foundation of any outbound plan, and they have held up even as generative tools made personalised messaging free to produce. The reason is that what earns a reply was never the personalisation itself.
Last reviewed July 2026. Channel response figures from Sortinghat outreach data, stated as directional.
Key takeaways
- The phone still wins by a wide margin. Around 40 responses per 100 contacted, against 25 on WhatsApp and 12 to 15 by email.
- Personalisation at scale stopped being a differentiator. When a mail-merge can reference someone's last three roles, referencing them proves nothing.
- Relevance beats wording, and always did. A precise segment with average copy outperforms brilliant copy sent broadly.
- These are ceiling numbers. Measured on strong roles with real compensation. A weak role performs worse on every channel.
Why the phone still beats messaging
This surprises teams who assumed messaging had replaced calling, so it is worth being specific.
A call is harder to ignore than a notification and easier to resolve. Thirty seconds settles whether someone is interested, where a message thread takes two days and three exchanges to reach the same point.
A call also carries a signal that a message does not: somebody picked up a phone to speak to you specifically. On senior roles that signal is a meaningful share of the response rate.
The catch is cost. Forty calls is a recruiter's entire day, which makes the highest-yield channel the worst-scaling one. That tension is the whole design problem in outbound.
What generative tools actually changed
They removed the cost of writing something that looks personalised, which removed the meaning of writing something that looks personalised.
A message referencing a candidate's last three roles used to prove somebody read the profile. It now proves somebody connected a data source to a template. Candidates worked this out quickly, and the response to superficially personalised outreach fell accordingly.
What did not change is that a relevant message to a well-chosen person still gets read. The differentiator moved from how the message is written to who receives it.
How to sequence when the best channel scales worst
Given the response numbers, the sequence that works is not best-channel-first. It is cheapest channel that still gets a reply, escalating to the expensive one on a shortlist.
- WhatsApp first. Near-zero marginal cost, around 25 responses per 100, and it reaches people during a working day without interrupting them.
- Email 24 hours later. Carries detail WhatsApp cannot, and reaches people who ignore unknown numbers.
- Call on day four, to non-responders you actually want. This is where the 40 percent rate gets spent deliberately rather than sprayed.
- Final touch on day nine, on whichever channel produced any engagement.
Leading with calls across 300 people wastes the one channel that converts.
Segment before you write anything
This matters more than every other decision in outbound combined.
A usable segment has four constraints: capability, level, location and an availability signal. "Java developers" is not a segment. "Java developers, 4 to 7 years, Pune, whose notice period we confirmed in the last six months" is one.
Narrow segments feel uncomfortable because the list is short. Short lists reply. The 40 percent figure above is measured on people who were a genuine fit, and it collapses on a broad pull regardless of what the message says.
Why response rates decay, and what to do
A sequence that worked last year quietly stops, and teams usually blame the copy. Three more likely causes.
The segment has been contacted before. Response is partly a function of novelty. A pool worked three times in eighteen months responds worse regardless of wording.
The roles got weaker. These figures are conditional on strong compensation. If the desk has drifted downmarket, response falls for reasons no rewrite addresses.
Contact data decayed. A falling response rate and a falling delivery rate look identical in most reporting. Check delivery first.
Frequently asked questions
What is a good candidate response rate?
On well-paid roles with real prospects, roughly 40 out of 100 contacted respond to a phone call, about 25 on WhatsApp and 12 to 15 by email. Weaker roles perform materially worse on every channel, so treat these as a ceiling.
Which outreach channel works best for recruiters?
The phone, by a clear margin, followed by WhatsApp and then email. It is also the worst-scaling channel, since 40 calls is a full recruiter day, which is why it should be spent on a shortlist rather than a list.
Has AI made recruitment outreach less effective?
It removed the cost of producing personalised-looking messages, which removed the signal that personalisation carried. Relevance still works; superficial personalisation no longer does, because candidates recognise a template with a data source attached.
How should you sequence recruitment outreach?
Cheapest effective channel first, escalating to the expensive one on a shortlist. WhatsApp, then email 24 hours later, then a call on day four to non-responders you actually want, then a final touch around day nine.
Why did my outreach response rate drop?
Usually one of three causes: the segment has been contacted repeatedly, the roles have become less attractive, or contact data has decayed. Check delivery rate first, since a data problem and a copy problem look identical in most reporting.
Does personalisation still matter in candidate outreach?
Relevance does. Referencing someone's history no longer proves anything, since it can be generated. What still earns a reply is contacting the right person about a role that genuinely fits their situation.
Establish your own baseline first
Run one campaign to fifty people from your own database, not five hundred from a fresh pull. Fifty is enough to read a signal and small enough that you will actually personalise the segment properly.
Record responses by channel and by day. After three campaigns you have a baseline worth more than any published benchmark, including the numbers on this page.
Run a segmented campaign on your own pool
Bring one role and we will run it across call, WhatsApp and email to show you where your replies actually come from.
Book a demoFounder of Sortinghat, an AI-native ATS and CRM for staffing, search and RPO firms. Writes about recruiter capacity, sourcing economics and what actually changes when AI reaches a delivery desk. More about the author
Related reading