How to Automate Candidate Outreach That Gets Replies
How-to — — by Mahmoud Zalt
Automate the research, the drafting and the follow-ups without the rule breaking. What to personalise, what to template, and what the law expects.
Outreach is the part of hiring that quietly consumes an afternoon and produces nothing you can point at. Twenty profiles read, twelve messages written, three replies, two of them saying no. Then a week goes by and nobody follows up with the nine who never answered.
The follow-ups are where the automation pays off, not the first send. This guide walks through what to automate, what to keep human, and the rules that apply once you start messaging people who never asked to hear from you.
What can you actually automate in candidate outreach?
Four things, cleanly. Building the list, researching each person, drafting a first version of the message, and scheduling the follow-up sequence. The judgement about whether this person is worth contacting, and the opening lines that prove you read something, stay with you.
- List building: assembling candidates from sources you are permitted to use, with the criteria written down rather than improvised per search.
- Research: a short brief per person covering what they work on, what they have shipped, and anything recent worth referencing.
- Drafting: a first version of the message with the role details filled in, leaving the opening for you.
- Sequencing: reminders and drafted follow-ups on a schedule, plus automatic suppression the moment someone replies or opts out.
Why do most automated sequences get ignored?
Because they personalise the wrong field. Inserting a first name, a job title, and a company into a template produces a message that is technically about the recipient and obviously not written for them. Everyone receiving these has received a hundred of them, and the shape is recognisable in about two seconds.
What earns a reply is a sentence that could only have been written to that person. Something they built, wrote, spoke about, or maintain. That sentence is short, it goes first, and it is the only part of the message that has to be original.
| Part of the message | Automate it? | Why |
|---|---|---|
| The opening line | No | This is the whole reason someone keeps reading. It has to reference something specific and true. |
| Why you are contacting them specifically | Half | Research can be automated. The link between their work and your role is your call to make. |
| The role summary | Yes | It is the same for everyone. Write it once, well, and reuse it. |
| Compensation and location | Yes | Fixed facts. Include them early, because leaving them out costs replies. |
| The ask | Yes | One clear low-effort question works across every message. Do not invent a new one each time. |
| The opt-out | Yes | Should be in every message automatically, and honoured automatically. |
| Follow-up timing | Yes | This is the highest return automation in outreach and the one people skip. |
Vendors in this space publish reply rate figures that are usually flattering, and they are marketing rather than independent research, so treat any specific percentage you see with suspicion including the ones in tool comparisons. What is not in dispute is the direction: messages that show evidence of reading get answered more often than messages that do not.
Build a sequence that respects the person on the other end
- Write the role summary once and get it right — Two or three sentences covering what the person would own, the team size, the compensation range, and where the work happens. This block never changes, so it deserves an hour.
- Generate a research brief per candidate — Public work only. What they build, where they have written or spoken, what they seem to care about professionally. Half a page is plenty and it is what makes the opening line possible.
- Write the opening line yourself — One or two sentences referencing something specific from the brief. If nothing in the brief gives you something honest to say, that is a signal to drop them from the list rather than to write filler.
- Cap the sequence at three touches — The first message, one nudge after four or five days, and a short close after another week. Beyond that you are not persistent, you are noise, and the reply rate will tell you so.
- Suppress on reply and on opt-out, automatically — A follow-up that lands after someone already said no is the fastest way to burn a name. Wire the suppression before you send the first message, not after the first complaint.
- Send from a real person's address — The name on the message should be the person who will actually answer the reply. Nobody wants to respond to an alias, and a reply landing nowhere is a wasted send.
- Track replies by opening line, not by template — You already know the template works or it does not. What you want to learn is which kind of specific detail earns responses in your market, and that only shows up if you log it.
A useful test before any batch goes out: pick three messages at random and read them as though you received them cold on a Tuesday morning. If the first line could be swapped between all three without anyone noticing, the sequence is not ready.
What rules apply to automated candidate outreach?
Three things bind most small teams: platform terms, data protection law, and the rules around automated hiring decisions. They apply independently, so satisfying one says nothing about the others.
- LinkedIn prohibits automated messaging. Section 8.2.13 of the User Agreement forbids bots or other unauthorised automated methods used to access the service, add or download contacts, or send or redirect messages. Their help pages state that accounts using third-party automation extensions face restriction or suspension.
- GDPR applies to sourced data. If you build a list of people in the EU from public profiles, Article 14 requires you to tell them what you hold and why, and where the data will be used to contact them, that information must be provided at the latest with the first communication.
- Targeted job advertising is named as high risk in the EU AI Act. Annex III point 4(a) covers AI systems used for recruitment or selection, expressly including placing targeted job advertisements, so an audience-picking model is inside the same regime as a CV filter.
- Opting out has to actually work. Every message needs a simple way to stop hearing from you, and the suppression must apply across every channel you use, not only the one they replied on.
The practical read is that email you send from your own system is far easier to keep clean than automation pointed at someone else's platform. If a tool promises unattended messaging inside a network that forbids it, the account it puts at risk belongs to your company, not to the vendor.
This is the shape of work we built Sistava around. You hire an AI Employee, describe the role and the kind of person you want to reach, and it does the research, drafts each message with the brief attached, and keeps the follow-up schedule. Nothing goes to a candidate until you have read it and said yes.
What automation will not do for your outreach
It will not make a weak role attractive. If the compensation is below market, the title is unclear, or the description reads like it was written by committee, a better sequence just delivers that message to more people. Outreach amplifies whatever you are actually offering.
- It cannot replace a referral. A message forwarded by someone the candidate already trusts outperforms anything you can write cold.
- It cannot judge fit. A profile that matches on keywords tells you nothing about whether this person wants the job you have.
- It cannot repair a burned list. Once a community decides your company spams, that reputation outlives the campaign by years.
- It cannot carry the conversation. The reply is where hiring actually starts, and that has to be a person who can answer real questions.
The honest version of this whole topic is that automation buys back the hours around the message, and the message itself still has to be worth reading. Teams that get results from automated outreach are usually sending fewer messages than before, not more, because the research step disqualifies people earlier.
Start with one role and twenty candidates. Research briefs generated, opening lines written by you, three touches, suppression wired up, replies logged against what you said. If that works, widen it. If it does not, the fix is almost always in the first two lines rather than in the tooling.
Most small teams do not need a sourcing platform for this, they need the research and the schedule handled by something that does not forget. That is the job an AI Employee takes on at Sistava, running the same routine every week, keeping the record straight, and stopping for your approval on anything that leaves the building.
FAQ
Can I automate LinkedIn messages to candidates?
Not within LinkedIn's terms. Section 8.2.13 of their User Agreement prohibits bots and other unauthorised automated methods used to access the service or to send or redirect messages, and their help pages state that accounts using third-party automation extensions face restriction or suspension until the software is disabled. Automate the research and the drafting, and send by hand, or run your sequence over email you control.
How many follow-ups should a candidate sequence have?
Three touches total is a sensible ceiling: the first message, a short nudge after four or five days, and a brief close a week later. Past that, response rates do not improve and the cost lands on your employer reputation. Suppress everything the moment someone replies or asks you to stop.
Do I need consent to email a candidate I found online?
It depends on where they are. Under the GDPR, if you collect someone's personal data from a source other than the person, Article 14 requires you to tell them what you hold, why, and on what legal basis, and that information must reach them at the latest with your first communication. Rules differ by country and this is not legal advice, so check your own position before you build a list.
What actually makes a cold candidate message get a reply?
A specific opening that proves you read something they made, followed by compensation and location stated plainly, and one low-effort question at the end. The parts you can template are the role summary and the ask. The opening line has to be original, and if you cannot write an honest one, that person probably does not belong on the list.
Is using AI to target who sees a job advert regulated?
In the EU, yes. Annex III point 4(a) of the AI Act lists AI systems intended for recruitment or selection, expressly including placing targeted job advertisements, among the high-risk uses. That puts audience targeting in the same regulatory category as CV filtering, with duties around risk management, logging and human oversight.
Should the messages come from a person or from the company?
From the person who will read the reply. A named sender with a real inbox produces better conversations than a careers alias, because the candidate can tell their question will reach someone who can answer it. It also means the follow-up feels like a person remembering rather than a system firing.