The Smarter Way HR Professionals Are Using AI to Communicate with Candidates

Every recruiter has had this happen. A strong candidate goes silent after the second interview. You check your last message. It’s a polite, well-structured note that mentions the role, the timeline, and the next step, written by AI in 12 seconds and sent without a second look. Reads fine. Feels like nobody actually wrote it. The candidate noticed.

Candidate communication is the area of HR where AI delivers obvious gains and obvious risk in the same breath. It can produce 30 personalised, on-tone messages in the time it used to take to write three. It can also produce 30 messages that sound like a single template ran through a thesaurus, which is what most candidates’ inboxes actually look like in 2026. The difference is in the workflow, not the tool.

The reference frame that holds up

LinkedIn’s Future of Recruiting 2025 found that 37% of recruiting professionals are actively using or experimenting with generative AI, with sourcing and outreach as the top use cases. That’s a doubling in roughly two years. The strong recruiters in that 37% aren’t using AI as a message-generation machine. They’re using it as a personalisation lens that takes their own draft and tunes it to a specific candidate.

The pattern is: write the message yourself in three or four sentences, then ask AI to personalise it using the candidate’s actual background. The output reads as if you spent an hour reading their LinkedIn, because effectively you did, the AI just did the reading for you.

The personalisation prompt that doesn’t feel templated

Here’s the prompt that turns a generic outreach into a message a candidate actually replies to. The trick is feeding it the public profile content as raw text, not relying on the model’s general knowledge of the person.

“You are helping me personalise an outreach message for a [JOB TITLE] role at [COMPANY]. Below is my own draft, which captures the role and my tone but is generic. Below that is the candidate’s LinkedIn ‘About’ section and their two most recent role descriptions, pasted verbatim. Rewrite the message to reference one specific, non-obvious detail from their profile that genuinely connects to why this role might interest them. Do NOT compliment them generically (‘your impressive background’, ‘your strong track record’). Do reference something specific (a project they mentioned, a transition they made, a skill the role would let them develop further). Maximum 110 words, conversational, signed off with a clear single next step. My draft: [PASTE]. Their profile: [PASTE].”

What lands in the inbox reads as if it was written by someone who actually looked at the profile. Reply rates from this kind of message are noticeably better than from template outreach, which makes sense, candidates can tell which they’re getting.

The choice of model matters less here than the choice of inputs. ChatGPT, Claude, and Gemini all do this task well. The two things that don’t work: giving the model a candidate’s name and asking it to “look them up” (it can’t, reliably, even with web access enabled), and pasting in a partial profile and hoping the model fills in the rest. Paste the actual text, treat the model as a writing assistant, and the output is good.

The AI-tells that flag unedited output

Candidates are now savvy enough to spot AI-written outreach, and the recruiters who haven’t noticed this are losing real candidates to it. The tells, in order of how badly they read:

The opener “I hope this message finds you well.” The phrase “I came across your profile and was impressed by your background.” Three-sentence paragraphs, all roughly the same length, with one transitional adverb each (“Additionally,” “Furthermore,” “Moreover”). A closing question that feels asked-by-rote: “Would you be open to a brief conversation?” An em dash anywhere. Lists where every bullet starts with the same verb. The phrase “uniquely positioned.” Generic praise of “your experience in [field]” without naming a specific role or output.

A good edit pass is faster than a bad one. Scan the message for the patterns above, replace one or two with phrasing you’d actually use in conversation, and the message reads as human. Skip the edit and you’ll send the same five tells your candidates have learned to recognise.

A failure mode: bulk personalisation that isn’t personalised

Here’s the trap. You build a beautiful personalisation prompt. You then run it across 200 candidates in a batch, with ChatGPT’s API or a no-code automation through Make or Zapier, and you send the outputs without reading them.

About 5% of those messages will contain something hallucinated, slightly off, or misjudged. A reference to a project the candidate didn’t actually work on. A skill they put on their profile sarcastically. A current employer the model swapped for a previous one. The other 95% are great. But the 5% land in inboxes and get screenshot and shared, and the screenshot says more about your employer brand than the 95% combined.

The defensible version of bulk personalisation is human-in-the-loop on every message. AI generates, recruiter scans (10 seconds per message), recruiter sends. That’s still 10x faster than writing each one, and it stops the embarrassing mistakes in their tracks.

The library worth building once

Beyond outreach, candidate communication has roughly eight templates that recur across every recruiter’s week: initial outreach, follow-up to silence, post-screening update, scheduling the next stage, post-interview thank-you, “we’re still deciding” holding message, offer message, and decline message. Build all eight as personalisation prompts now, store them somewhere your team can reuse them (Notion, a ChatGPT Team workspace, or your ATS notes section all work), and the entire team’s outreach quality lifts at once.

The decline message is the one most teams under-invest in. It’s also the highest-impact one for employer brand, because the candidates you decline talk about your process to other candidates you’d want. Here’s the version worth saving:

“Write a decline message to [CANDIDATE NAME] for the [ROLE] at [COMPANY]. They [briefly: how far they got and what was strong about them]. The reason for declining is [PRIMARY REASON, written for the recruiter’s own clarity, will not appear in the message]. Tone: warm, specific, respectful of their time, gives one piece of useful and honest feedback if appropriate. Do NOT use phrases like ‘we wish you the best in your future endeavours’, ‘this was a difficult decision’, or ‘we will keep your resume on file’ unless they’re literally true. Maximum 130 words. Sign off in my voice as [RECRUITER NAME].”

Candidates who get a decline message that reads as if a person wrote it tend to remember the company favourably. That’s the kind of compounding employer brand work AI is genuinely useful for, when the human stays in the loop on every send.

What candidate communication looks like in a year

The recruiters whose pipelines feel different in 2027 won’t be the ones who automated outreach. They’ll be the ones who used AI to personalise more deeply at scale, while keeping the read-and-edit step every single time. The work isn’t gone. It’s shifted from typing to judgment, which is what most recruiters preferred to be doing in the first place.