Personalised Candidate Outreach at Scale Without Becoming Spam

Personalised candidate outreach is the difference between a message someone answers and a message that trains them to ignore your company name. The tension is obvious: personalisation takes time, and recruiters are measured on coverage. AI tooling appears to settle the argument by writing a tailored opener for 400 people at once. It does not, and the reason it does not is measurable in your reply rate and your sender reputation.

This is about outbound messages you send to people who did not apply. Inbound application spam is a different problem with a different fix, and we treat it separately in our guide to stopping application spam.

What a message that gets answered actually contains

Good outreach is short. Three sentences is usually the whole message, and each sentence has one job.

• Sentence one - Job: prove you looked at this specific person; Contains: one concrete detail from their actual work or history
• Sentence two - Job: say why that detail matters for this role; Contains: the link between what they do and what the job needs
• Sentence three - Job: make replying cheap; Contains: one small ask, with salary or range if you have it

What is deliberately absent: company boilerplate, a list of benefits, the phrase "I came across your profile", and any sentence that would survive unchanged if pasted to someone else. That last test is the only personalisation test that matters. Delete the candidate's name from your message. If the rest still reads as written for them, it is personalised. If it reads as a template with a merge field, so will they.

Before and after

Before: "Hi Sarah, I came across your profile and was impressed by your background. We are a market-leading organisation hiring a Senior Management Accountant, and I think you could be a great fit. We offer flexible work, a supportive culture and genuine career progression. Would you be open to a confidential chat this week?"

That message says nothing a recipient could not predict from the first line. It works at a low single-digit reply rate and costs a little reputation each time.

After: "Hi Sarah, you have run month-end for a multi-entity group through two ERP migrations, which is the part most candidates for this role have not done. A manufacturer in Clayton is replacing a retiring financial controller and wants someone who has survived a migration rather than someone who has read about one. It pays 145 to 160 plus super. Worth a fifteen-minute call?"

The second message is barely longer. The difference is that the first sentence could only have been written after reading her history, and the third removes the main reason people do not reply: not knowing whether the money is serious.

The per-candidate detail AI can supply, and the judgement it cannot

Used properly, AI is a research assistant for outreach rather than a writer. It is good at reading a long work history and extracting what is distinctive, summarising a company so you understand the environment someone works in, and grouping a list by the genuine difference between candidates.

What it is poor at is the judgement call underneath: whether the detail it found is the detail worth leading with. A model will confidently open on a certification the candidate finished nine years ago and no longer cares about. It also has no idea whether a candidate is likely to move, which is the actual question.

The workable split is to let the tool do the reading and keep the first sentence yours. A recruiter who reviews a machine-drafted opener and rewrites it in fifteen seconds is still working at scale, and the resulting message passes the delete-the-name test. A recruiter who approves 300 drafts without reading them is running a mail merge with extra steps.

One trap worth naming: do not let a tool infer things it cannot know. A message that claims to know why someone is unhappy in their current role reads as creepy when it is wrong, and worse when it is right. Stick to what is on the record. Related judgement applies to the channel itself, since messaging someone through a professional network carries its own risks, which we set out when looking at the risks of LinkedIn InMail for reference and sourcing work.

How volume burns reply rates and sender reputation

Outreach at scale has two costs that do not appear on the recruiter's dashboard.

The first is audience fatigue. In a market the size of Australia, a specialist talent pool is small. Send four generic messages to the same 600 people over a year and you have taught that pool to ignore your sender name. The damage is cumulative and invisible, because nobody reports a non-reply as a loss.

The second is technical. Email providers score sender reputation on engagement, complaint rate, bounce rate and sending pattern. A sudden jump from 40 messages a day to 400, into a purchased list, with low opens and a handful of spam reports, is exactly the signature filters are built to catch. Once reputation degrades, your legitimate mail, including offer letters and interview confirmations, starts landing in junk, and recovering a domain takes weeks of good behaviour.

Practical guards that cost nothing: keep daily volume per sender steady rather than bursty, verify addresses before sending so bounces stay low, suppress anyone who has not engaged across two prior campaigns, and never send outbound from the same domain that carries your offer and onboarding mail. The outreach you send is also brand contact, which is why it belongs inside the same standard as the rest of your branded candidate experience at scale rather than being treated as a sourcing activity nobody else sees.

Spam Act 2003 duties for Australian recruiters

Recruitment outreach is frequently a commercial electronic message, which puts it inside the Spam Act 2003. The ACMA sets out three obligations for anyone sending marketing emails or messages with an Australian link.

• Consent - Requirement: express or inferred consent before you send; Practical note: the ACMA states you cannot send an electronic message to ask for consent, because that request is itself a marketing message
• Identification - Requirement: the message identifies the sender and includes accurate contact details; Practical note: a personal signature with the agency name and a working phone number satisfies this
• Unsubscribe - Requirement: a functional unsubscribe facility, honoured promptly; Practical note: must keep working after the campaign ends, not just while it runs

Two points catch recruiters out. First, inferred consent is narrower than people assume: the ACMA ties it to a provable, ongoing relationship where the message is directly relevant to that relationship, and having just bought something from you does not create it. Second, if you buy or rent a list, you remain responsible for consent on every address in it, and under the Spam Act the burden of proving consent sits with the sender. Address-harvesting software and harvested-address lists are separately prohibited. Keep a record of when, how and from whom consent was obtained: a candidate who gave you their address at a careers event two years ago is a weaker position than one who ticked a box on your site last month, and the difference only matters once someone complains.

Frequently Asked Questions

Does the Spam Act apply to LinkedIn messages and SMS?

The Act covers commercial electronic messages including email, SMS, MMS and instant messaging. Messages sent inside a platform are also governed by that platform's own terms, which are often stricter than the law. Treat both as binding and assume the stricter one applies.

How many outreach messages per recruiter per day is safe?

There is no legal number, and the technical answer depends on your domain's history. The pattern matters more than the total: steady daily volume with good engagement is safer than the same monthly total sent in two bursts. If you are increasing volume, raise it gradually over several weeks.

Is a follow-up message worth sending?

One is. Two is usually where reply rate per message stops justifying the goodwill spent. Make the follow-up add something, such as the salary range or a detail about the team, rather than just asking again.

Should we tell candidates AI helped write the message?

Not in the message itself, but do not deny it either. The standard to hold is that a human made the judgement about why this person and checked what was sent.

What reply rate should we expect?

It varies too much by role, seniority and market to quote a benchmark honestly. The useful comparison is your own: measure your current template for a month, then run genuinely personalised messages against the same pool and compare. Measure replies, not opens, because opens are unreliable.

Personalised Candidate Outreach Beats Volume

The choice is not between personalisation and scale. It is between spending your finite credibility with a small Australian talent pool carefully or quickly. Pick one role, write twenty messages that pass the delete-the-name test, keep consent and unsubscribe clean, and compare the replies against your last campaign. Speed still matters once they answer, which is where our guide on speed and candidate experience picks up.