Passive Candidates in AI/ML: How to Reach People Who Aren't Looking

The strongest AI/ML specialists are usually not looking for a job. They are building models, improving evaluation pipelines, cutting inference cost, shipping product to production or choosing between several offers that arrived through a trusted network. That is why posting a vacancy and waiting for applications is almost always too passive for this market. At IT recruiting agency EvoTalents we work daily with roles where an open competition produces nothing, and we see one consistent pattern: the conversation with this kind of candidate starts with concrete technical context, not with employer brand.

This article is for founders, CEOs, CTOs and Heads of Talent who are hiring an AI/ML team and have already realised that standard recruiting does not work here. Below we break down why a job post is too passive for this market, how to define the role before sourcing starts, how to separate AI engineering profiles, how to write technical outreach that proves you understand the work, and what the hiring team has to fix on its own side before a candidate will even want to reply.

WHY AI/ML CANDIDATES STAY PASSIVE

Senior AI/ML specialists do not need to browse job boards. Many of them receive several recruiter messages every week, and most of those messages look identical. So by default this candidate is not looking for anything and filters the incoming flow almost automatically.

A passive AI/ML candidate usually agrees to a conversation for one of five reasons:

• a harder technical problem than the one they have now;
• a better environment: stronger data, infrastructure or research conditions;
• clearer impact on the product and on business metrics;
• stronger leadership and faster decision-making;
• a compensation, equity and mission equation that makes sense.

If a company cannot convincingly explain at least two of these reasons, passive candidate sourcing will move slowly no matter how many profiles the recruiter reviews. At EvoTalents we start not with the search for people, but with an honest answer to one question: why should a strong engineer trade their current problem for this one.

WHY A JOB POST IS TOO PASSIVE FOR THE AI/ML MARKET

A vacancy on your website or on a job board works for roles with an active flow of candidates. In AI/ML that flow does not exist. People who can genuinely build production LLM systems, RAG pipelines, agentic workflows or an ML platform barely look at job ads, because they are already being headhunted on a regular basis.

A published vacancy therefore collects mostly those who are already in active search, and that is not always the top segment of the market. The strongest candidates simply never see your ad, which is why they never enter your funnel. Technical sourcing and targeted outreach exist precisely to reach people who do not submit applications.

STOP SEARCHING BY YEARS OF EXPERIENCE

AI hiring breaks when companies apply old experience filters to young technologies. Production LLM, RAG, agentic systems and modern MLOps have existed as mainstream practice for only a few years. A requirement like "5+ years of commercial LLM experience" filters out exactly the people who built these systems from scratch, because that much tenure could not physically have accumulated yet.

Instead of years, look for evidence of the work: systems shipped to production, model evaluation, latency and inference cost optimisation, ownership of data pipelines, fine-tuning, real product decisions built on AI. The hiring signal sits in what the person built, tested and improved, not in the number of years on the CV. A candidate with two intense years of production LLM work is often stronger than someone with six years of general ML who has never shipped an AI feature to a real user.

DEFINE CANDIDATE PERSONAS BEFORE SOURCING STARTS

"AI engineer" is too broad a concept for sourcing. That single title hides people with completely different skills, different target companies and different motivation. Before outreach begins, the market should be split into personas:

• Applied LLM engineer: RAG, agents, evaluation, product integration.
• ML platform engineer: MLOps, deployment, observability, infrastructure, model serving.
• Research-to-product profile: papers, prototypes, experimentation and translating research into production.
• Data / AI product engineer: product sense, analytics, user scenarios and business metrics.
• AI governance or safety profile: evaluation, risk, policy, responsible AI, auditability.

Each persona needs its own list of target companies, its own search strings, its own screening questions and its own angle in outreach. Trying to find everyone with a single query produces a blurred funnel and a weak response rate. At EvoTalents we lock the persona and the ICP profile before the first sourcing wave, because that directly determines who we write to and what we say.

WRITE OUTREACH THAT PROVES YOU UNDERSTAND THE WORK

Passive candidates reply when the message is specific. "We have an interesting AI role" is not specific. A strong first message shows that the author understands what the person actually does, and why this new problem might be interesting to them.

A working structure for first contact looks like this:

• one line on why this particular candidate is relevant;
• one line on the technical challenge that needs solving;
• one line on the impact on product or business;
• one line on the practical setup: location, remote model, seniority and process;
• a low-friction question that is easy to answer.

The job of first contact is more modest than it looks: earn a short conversation. You can sell the company later in dialogue, where context works, rather than in the text of an ad.

USE AI TOOLS, BUT KEEP HUMAN JUDGEMENT

AI helps recruiters find profiles, enrich context, identify adjacent titles, prepare message variants and process large lists faster. The market has already accepted this: 84% of developers use or plan to use AI tools in their work, and 51% of professional developers use them daily.

At the same time, using AI tools and hiring strong AI/ML specialists are two different things. In AI/ML hiring human review is mandatory, because strong candidates often have non-linear profiles. A backend engineer who shipped two LLM features to production in the past year may be more relevant than someone with a formally "correct" title. A data scientist who built a model evaluation system is sometimes more valuable than a formal ML Engineer. An automated filter will miss these candidates, while an experienced recruiter will see the signal behind the wording.

WHAT THE HIRING TEAM HAS TO FIX BEFORE CANDIDATES WILL WANT TO REPLY

Often the problem is not sourcing, it is that the role is not ready for the market. Before launching outreach, at EvoTalents we check several things on the client side, because these are what kill the response rate most often:

• A poorly designed AI role, where a single vacancy combines ML, backend, data and product research with no priorities.
• A salary calculated from the budget rather than from the market.
• A weak AI and product story: "AI-first" with no depth at the level of models, data and infrastructure.
• Slow decision-making and no clarity on who actually makes the final call.
• Too many people influencing the decision with no single owner of the role.

If these things are not fixed, even perfect outreach will produce a weak result. The first task is therefore to make the role attractive and clear, and only then to scale sourcing.

FIRST 90 DAYS: A PASSIVE AI/ML SOURCING PLAN

The first month has to answer the main question: does the market believe this role is attractive. If the response is weak, what needs to change is the message, the role scope or the compensation, not the volume of sourcing.

• Days 1-7: define the role archetype, must-have competencies, target companies and the outreach narrative.
• Days 8-21: run the first sourcing waves across exact and adjacent titles, tracking response by persona.
• Days 22-35: recalibrate the brief with the hiring manager based on real candidate feedback rather than assumptions.
• Days 36-60: go deeper into referrals, communities, open-source signals and competitor mapping.
• Days 61-90: build a warm pipeline that can be reused for the next AI/ML roles.

THE INTERVIEW PROCESS: HOW TO KEEP A PASSIVE CANDIDATE ENGAGED

Passive candidates drop out of slow processes quickly. They did not apply, so the company has to keep earning their attention. A long, vague or inconsistent process reads as a signal of how the company works internally.

• Limit the process to the stages you genuinely need.
• Use structured technical interviews and realistic work samples.
• Do not hand out unpaid projects that are effectively production work.
• Give feedback fast.
• Keep the hiring manager visible and prepared.
• Explain the mission, the roadmap and how decisions get made honestly.

A good passive candidate evaluates the company as carefully as the company evaluates them. Speed and clarity of process are part of the offer, not just good manners.

EVOTALENTS CASE: TURNING ZERO APPLICATIONS INTO MARKET SIGNAL

Client: anonymised technology company. Level: senior AI/ML or specialist engineering role. Format: specialist sourcing and candidate engagement.

Situation

The company needed AI/ML talent, but direct applications did not match the required level of expertise. The strongest candidates were already employed, selective and unresponsive to typical recruiter messages. A standard job post was not attracting the people who had actually built systems like this.

The EvoTalents approach

• Built a map of target companies and adjacent candidate pools first, and only then started outreach.
• Built the messaging around the technical problem, product context and decision scope rather than general AI hype.
• Used candidate objections to sharpen the brief and adjust the positioning of the role.
• Helped the hiring team see exactly where the offer, role scope or geography needed adjusting.

Result

• Stronger target market identified for the role: Serbia
• Market-based compensation level: $5,000-7,500
• Strong profiles in the mapping: around 15-20 matching Python, LLM APIs and RAG
• Client received a market signal before committing to a full search

The wider lesson applies to any passive AI/ML sourcing: candidates respond when outreach is grounded in real technical relevance, realistic compensation and a process that moves fast enough to keep momentum.

MOST COMMON MISTAKES

• Using "5+ years of LLM experience" as a hard filter.
• Sending generic messages to candidates who already receive dozens of them.
• Failing to separate applied AI, ML platform, research and governance personas.
• Running a slow process for people who were not thinking about changing jobs at all.
• Ignoring candidate motivation and focusing only on technical keywords.

FAQ

Where do you find passive AI/ML candidates?

Start with target companies, open-source ecosystems, research communities, AI product teams and groups working on ML platforms. Then add referrals and adjacent titles. The key is not to rely on job boards alone, because that is exactly where strong passive candidates are least likely to be.

What makes AI/ML outreach effective?

Specificity. Mention the technical problem, explain why this particular candidate looks relevant, describe what the company is building and what the process looks like. General excitement about AI is not enough for senior candidates who receive many similar messages.

How do you assess a passive AI candidate quickly?

Use structured interviews, evidence of systems shipped to production and realistic work samples. Focus on production impact, model evaluation and real trade-offs rather than the number of years on a CV.

Why do passive candidates drop out?

They leave the process when it is slow, when the role is vague, when compensation is unclear and when technical interviews feel like unpaid work. Since this candidate never applied, any friction in the process costs far more than it would in ordinary hiring.

When is it worth engaging a recruiting agency for AI/ML hiring?

When the role is strategic, niche or confidential and direct applications do not deliver the required level. In those cases it is worth starting with market signal and market mapping rather than mass CV submission. It saves the team's time and shows the real picture of available talent before a full search begins.