How to Write a Sourcing Prompt That Finds the Right Candidates in 2026
Most recruiters using AI sourcing tools get mediocre results not because the tool is weak, but because the prompt is vague. "Senior software engineer in Paris" returns 10,000 profiles. Half are irrelevant. You spend an hour filtering. That's not the point of AI sourcing. A well-written prompt does the filtering work upfront. It tells the AI exactly who you're looking for, what context matters, and what you can ignore. The result is a shortlist you can actually use. Here's how to write sourcing prompts that work — with examples you can adapt today. The gap between a recruiter who finds the right candidates in two hours and one who spends two days filtering noise usually comes down to the quality of the initial prompt. Get the four elements right, iterate once or twice, and the AI does the heavy work. Book a demo to see how Kalent turns a plain-text prompt into a shortlist of enriched, contactable candidates.
- Why Prompt Quality Determines Your Results
- The Four Elements of a Strong Sourcing Prompt
- Prompt Templates You Can Adapt Right Now
- Common Mistakes That Kill Prompt Quality
- How to Iterate When Your First Prompt Misses
- Putting It Into Practice on Kalent
- A Note on Prompts for European Searches
- FAQs
Most recruiters using AI sourcing tools get mediocre results not because the tool is weak, but because the prompt is vague. "Senior software engineer in Paris" returns 10,000 profiles. Half are irrelevant. You spend an hour filtering. That's not the point of AI sourcing.
A well-written prompt does the filtering work upfront. It tells the AI exactly who you're looking for, what context matters, and what you can ignore. The result is a shortlist you can actually use.
Here's how to write sourcing prompts that work — with examples you can adapt today.
Why Prompt Quality Determines Your Results
AI sourcing platforms match candidates based on the signal you give them. A thin prompt produces a wide, noisy match. A precise prompt produces a tight, usable shortlist.
Think of it like briefing a researcher. "Find me a marketer" gets you 500 options and no direction. "Find me a B2B SaaS growth marketer with experience scaling paid acquisition in France, who's worked at a Series B or C company" — that researcher knows exactly where to start.
Same logic applies to AI. Better input, better output.
The Four Elements of a Strong Sourcing Prompt
Every effective sourcing prompt covers four things. You don't need to write a paragraph. You need to be precise on each one.
1. Role and Seniority
Start with the job title and level. Be specific — "engineer" and "senior engineer" pull from very different pools.
Weak: Software engineer
Strong: Senior backend engineer, 5+ years of experience, specializing in Python and distributed systems
Skip internal job codes and overly generic titles. Use the language candidates actually put on their profiles.
2. Must-Have Skills or Experience
List two to four non-negotiable criteria — the things that disqualify a candidate if missing. Keep it short. Every item you add narrows the pool, so only include genuine requirements.
Example: Must have Kubernetes experience in a production environment and have led at least one engineering team.
Ten must-haves returns three candidates. Two returns three hundred qualified ones.
3. Location and Language
Specify city, region, or country. If remote is acceptable, say so. If language matters for the role, include it.
Example: Based in London or open to remote within the UK. Must be fluent in English.
For European roles especially, language is a real filter. A French-speaking account manager in Lyon is a completely different search than an English-speaking one.
4. Context and Exclusions
This is the element most recruiters skip — and often the most valuable. Add context about company type, industry background, or what you're specifically not looking for.
Example: Ideally from a fintech or insurtech background. Not looking for candidates from large consulting firms.
Exclusions help the AI avoid obvious mismatches and save you from reviewing profiles that look right on the surface but would never fit the role.
Prompt Templates You Can Adapt Right Now
Three ready-to-use templates across different role types.
Technical Role
Senior data engineer with 6+ years of experience in Python, Spark, and cloud data pipelines (AWS or GCP). Based in Berlin or Hamburg, or open to remote within Germany. Has worked at a tech company or scale-up with 50 to 500 employees. German or English fluent. Not looking for candidates from large enterprise IT departments.
Commercial Role
Mid-market Account Executive with 3 to 7 years of B2B SaaS sales experience. Has consistently hit or exceeded quota. Based in Paris or Lyon. French native speaker, English working level. Experience selling to HR or finance buyers is a strong plus.
Leadership Role
Head of Engineering with experience managing teams of 10 to 30 engineers across multiple squads. Has led a product-focused engineering org at a Series B or C company. Based in London. Experience with platform or infrastructure domains preferred. Not looking for candidates from agency or consulting backgrounds.
Each of these gives the AI clear signal on role, level, skills, location, and context — specific enough to be useful, not so narrow the pool disappears.
Common Mistakes That Kill Prompt Quality
Copying from the job description
Job descriptions are written to attract candidates, not to power AI search. They're full of phrases like "passionate team player" and "fast-paced environment." Strip those out. Use only the factual, searchable criteria.
Over-specifying
Eight must-have skills, a specific company size, a specific funding stage, and a specific tool stack will return zero results. Ask yourself: what are the two things this person absolutely must have? Start there.
Vague location
"Europe" is not a location. "France" is better. "Paris or Lyon" is best. Precision here directly affects relevance.
Skipping exclusions
If you've run a search before and kept seeing the same irrelevant profile types, add an exclusion. It takes five seconds and saves significant filtering time.
How to Iterate When Your First Prompt Misses
Even experienced recruiters need two or three passes to dial in a search. Here's a simple approach.
Run your first prompt and review the top 20 profiles. Ask:
- Are these too senior or too junior?
- Are they from the right type of company?
- Is the location right?
- Are there obvious skill gaps?
Adjust one variable at a time. If profiles are too senior, soften the seniority requirement. If you're getting agency consultants when you want in-house candidates, add that exclusion. Small changes produce meaningfully different results.
Putting It Into Practice on Kalent
On Kalent, you type your sourcing prompt in plain language and the platform matches it against 200M+ profiles across Europe and the US. No Boolean syntax. No filters to configure. You describe the candidate you want, and Kalent returns enriched profiles with AI-generated summaries and verified contact details — around 80% mobile phone and email coverage per profile.
Once you have your shortlist, the Conversational Outreach Agent handles engagement across LinkedIn, email, SMS, and WhatsApp from the same interface. Prompt to outreach, no tool-switching required.
The prompt quality still matters. Better input means a better shortlist, and a faster path to interviews.
A Note on Prompts for European Searches
European sourcing has specific nuances that US-centric tools often miss. Language requirements, regional labor markets, and the real differences between candidate pools in Paris, Amsterdam, and Munich all affect your results. Your prompt needs to reflect that.
When sourcing for European roles, always specify:
- Country or city — not just "Europe"
- Language requirements, especially where a local language is essential
- Whether cross-border candidates are acceptable
A prompt that works well for a US search won't automatically translate to a European one. Adjust for geography every time.
FAQs
What is an AI sourcing prompt?
A plain-language description of the candidate you're looking for. You type it into an AI sourcing platform, and the platform matches it against a database of profiles to return the most relevant results. No Boolean syntax required.
How long should a sourcing prompt be?
Two to five sentences is usually enough. Cover role and seniority, two to four must-have criteria, location, and any relevant context or exclusions. Longer prompts don't automatically produce better results — only add detail that genuinely filters.
What makes a sourcing prompt too vague?
If it only includes a job title and a city, it's too vague. Without skill requirements, seniority, and context, the AI has no meaningful signal and returns a wide, noisy pool.
Can I use the same prompt for different markets?
Not without adjusting for location and language. Candidate pools, language requirements, and company types differ significantly across regions. A prompt written for a US search needs those fields updated before it works in Europe.
How do I know if my prompt is working?
Review the top 20 profiles from your first search. If most match your picture of the ideal candidate, the prompt is working. If you're seeing obvious mismatches, identify which variable is off — seniority, location, company type — and adjust that one element.
Do AI sourcing platforms require Boolean search skills?
Kalent is built for natural-language input, so Boolean isn't required. You describe the candidate in plain text and the AI handles the matching logic. That said, the quality of your description still determines the quality of your results.
What's the difference between a sourcing prompt and a job description?
A job description is written to attract candidates — it includes company context, benefits, and aspirational language. A sourcing prompt is written to find candidates — it includes only the factual, searchable criteria the AI needs to match profiles. Strip out the marketing language before using a job description as a prompt.
The gap between a recruiter who finds the right candidates in two hours and one who spends two days filtering noise usually comes down to the quality of the initial prompt. Get the four elements right, iterate once or twice, and the AI does the heavy work.
Book a demo to see how Kalent turns a plain-text prompt into a shortlist of enriched, contactable candidates.
