An AI agent can automate the repeatable parts of recruiting outreach: checking profiles against the role, drafting messages, sending follow-ups on schedule, stopping when someone replies and sorting the replies. What it should not do alone is make the claim "this role fits you" in the first message, answer sensitive questions or decide who gets rejected. In the EU, three rules set the limits: candidates must be informed within one month (Art. 14 GDPR), unsolicited email contact falls under Section 7 UWG in Germany, and LinkedIn's terms ban third-party automation tools (as of November 2026).
The practical split that works for most teams: you approve the first message, the agent runs the follow-ups. The first message carries a claim a person should stand behind. A follow-up repeats an offer with one new fact, so its timing and stop rules are mechanical. This article shows what to automate, how to set it up in seven steps, how many follow-ups pay off, and what the law and LinkedIn allow.
What does an active sourcing AI agent do, and what stays with you?
An active sourcing AI agent is software that takes a role brief and carries the outreach workflow forward without a click for every step. The test is simple: if every message still has to be written, scheduled and judged by a person, it is an AI feature, not an agent. Agents differ in how much they do alone, so decide per step.
| Step | Agent can do it alone | Keep a human in the loop |
|---|---|---|
| Read brief, find and check profiles | Search, rank, justify each match against the requirements | Approve the shortlist before anyone is contacted |
| Find the contact route | Look up a business email or the LinkedIn route, flag duplicates | Exclude people your company must not contact (competitor staff, people who objected before) |
| First message | Draft from the profile and your pitch | Approve or edit the first line; it carries the claim that the role fits |
| Follow-ups | Send on schedule, one new fact each, switch channel | Approve the templates once |
| Stop rules | Stop on any reply, pause on out-of-office, suppress after a refusal | Check the rule set before launch |
| Replies | Sort into interested, question, not now, refusal; hand over interested people | Answer questions on pay, working conditions and anything sensitive yourself |
| Rejections and decisions | Nothing | Always a person |
How do you automate sourcing and outreach in seven steps?
- Write a decision-ready brief. Must-haves, acceptable trade-offs, the reason someone might move, salary range. An agent that works from a vague brief only sends vague messages faster.
- Let the agent search and check, then review the shortlist. Look at the reasons it gives per profile. Results that appear in an instant are almost always a keyword or semantic filter; a check of every profile against your requirements takes minutes.
- Choose the channel per person. Email and LinkedIn together reach more people than either alone; connection request or InMail compares the LinkedIn options for the first contact.
- Approve the first message. One concrete reason, the range, one easy next step. Patterns are in recruiter outreach templates.
- Set the follow-up cadence. A common pattern is two to three follow-ups over about ten days (see the data below), each with one new fact: team, range, first project.
- Define stop rules before the first send. A personal reply stops every step, an out-of-office pauses, a refusal becomes a permanent suppression across all roles and colleagues.
- Fix who answers. An interested reply needs an owner and a reaction time. An agent that finds people faster than your team can answer only creates waiting candidates.
For the broader question of what else in outreach can be automated, see automating recruiting outreach.
Does AI personalisation work, and how many follow-ups pay off?
Partly yes, and the data points need careful reading. LinkedIn reports for its own AI-assisted messages a 44 % higher acceptance rate and acceptance more than 11 % faster than messages written without AI (LinkedIn Talent Solutions, Hiring Release 2025; company data, not independently verified). It describes acceptance of its own messages, not hiring speed.
On follow-ups, one large outreach platform published figures from more than 4 million recruiting messages (Pin, "Recruiting Outreach Benchmarks 2026", period June 2025 to May 2026, reply rate on delivered messages; vendor data):
| Measure | Value |
|---|---|
| Reply rate, automated email | 4.96 % |
| Reply rate, recruiter-written email | 6.31 % |
| Reply rate, LinkedIn message | 17.08 % |
| Share of all replies that came from follow-ups | 56.2 % |
| Cumulative share of replies after message 3 | 93.2 % |
| Cumulative share of replies after message 4 | 97.7 % |
What to take from it: more than half of the replies come from follow-ups, and almost all replies have arrived after the third or fourth message. A sequence of three to four steps is therefore a reasonable starting point; a fifth message adds little and raises the risk of annoying people. The figures are one vendor's data and show an association, not proof that automation causes the difference. Measure your own: reply rate on delivered messages, positive replies, meetings booked (see the sourcing funnel).
What do GDPR, UWG, LinkedIn and the AI Act allow?
GDPR: lawful basis and the one-month rule
If you collect personal data about a person from public sources, you must inform them within a reasonable period and at the latest within one month, and at the latest at first contact if you use the data to contact them (Art. 14 (3) GDPR). The usual lawful basis is legitimate interest under Art. 6 (1) (f), which needs a documented balancing of interests. The first message should therefore carry the information notice or a link to it, and the person must be able to object easily. A fully automated decision with legal effect on a person, such as an automatic rejection, falls under Art. 22; keep a human in that step.
Germany: email falls under the UWG
Advertising by email to a person who did not agree generally needs prior consent under Section 7 UWG. Whether a recruiting approach counts as advertising is judged by the context, so treat automated cold email in Germany as a question to settle with your data protection officer before launch. The details, with the information notice for a first email, are in contacting candidates by email. This article is not legal advice.
LinkedIn: automation tools can cost you the account
LinkedIn states that it does not permit third-party software that automates activity on its site, and that members using such tools risk having their accounts restricted (LinkedIn Help: prohibited software and extensions, checked November 2026). Before you connect an account to any agent, ask the vendor how it acts on LinkedIn, and read LinkedIn automation: safe or banned?
EU AI Act: what applies when
Transparency duties for AI systems that interact with people apply from 2 August 2026 (Art. 50). Obligations for high-risk AI in recruitment (Annex III) apply from 2 December 2027 under the Digital Omnibus (Regulation (EU) 2026/1744). Drafting outreach is usually not the high-risk part; ranking or filtering candidates can be. If candidates can chat with an AI agent, tell them they are talking to one.
What does Sprad do here, and where are its limits?
In Sprad Active Sourcing, one search runs across LinkedIn, XING, GitHub and your own talent pool, and every profile is checked against your requirements. Outreach runs on LinkedIn (connection request, message, InMail) and email, with AI templates and follow-ups, and all messages per person sit in one inbox. The system warns if someone was already contacted, also for another role or by a colleague. Agents can run fully automatically or with your approval, so the first message can wait for your release while the follow-ups run on schedule. A reply stops the sequence and hands over to you.
- Price (as of November 2026): the first LinkedIn and mailbox account is included in every Atlas Recruiter seat (299 € per seat and month); each further account costs 2,000 credits (about 180 € per month). With Recruiter Lite (79 € per seat and month) the outreach agent is an add-on at the same 2,000 credits. Starting is free, with 100 credits a month.
- Limits: the deep search takes a few minutes because every profile is checked; there is no SMS channel; the LinkedIn steps run through your connected account, so the LinkedIn terms above apply to you.
What should you check before signing with any vendor?
- Ask to see the second and third message of a sequence, not only the first.
- Does a reply stop all steps, also on the other channel?
- Is a refusal a suppression across roles and colleagues?
- Does the first message carry or link the privacy notice, and can you configure it?
- Is there an approval mode and a log of what was sent, to whom and why?
- How does the tool act on LinkedIn, and who carries the account risk?
- What happens to your data and talent pool if you cancel?
FAQ: AI agents for recruiting outreach
Can an AI agent personalise outreach?
Yes, if it has real material: the profile, the brief and your pitch. Personalisation means a checkable reason why this role fits this person. A template with the first name swapped in is not personalisation, and an agent personalises only as well as the brief allows.
Do we need consent before an AI agent contacts a candidate?
Under the GDPR, legitimate interest can be the basis, with a documented balancing and the information notice. For email in Germany, Section 7 UWG makes this stricter. Settle the channel with your data protection officer first.
How many follow-ups should an agent send?
Two to three follow-ups over about ten days is a sound start. In the vendor data above, 93 % of replies had arrived after the third message. Each follow-up should add one new fact and be easy to stop.
Which parts should never be automated?
Rejections and hiring decisions, answers on pay and sensitive topics, and the claim in the first message that the role fits the person. Use approval mode for those.
Is automated LinkedIn outreach allowed?
LinkedIn's terms do not permit third-party tools that automate activity on the site, and accounts can be restricted. Ask any vendor exactly how it acts on LinkedIn, and weigh the account risk.
When does the EU AI Act apply to outreach agents?
Transparency duties from 2 August 2026; high-risk duties for recruiting AI from 2 December 2027. Drafting messages is rarely the high-risk part; ranking and filtering candidates can be.
Jürgen Ulbrich
Jürgen Ulbrich has more than a decade of experience in developing and leading high-performing teams and companies. As an expert in employee referral programs as well as feedback and performance processes, Jürgen has helped over 100 organizations optimize their talent acquisition and development strategies.
