How to write a sourcing prompt that brings back the right profiles

Why the prompt matters so much

An AI sourcing agent understands natural language: you can talk to it as you would to a colleague. That is its strength, and also a trap. A vague request gives vague results, and an overly narrow one makes you miss good profiles.

The good news: there is nothing technical to learn. No Boolean operators, no syntax. You just need to give the agent the right information, in plain words. The prompt works like a brief: the one you would give a researcher who does not know the role yet.

The five elements of a good prompt

1. The role, and its variants

Give the job title, but also the related titles the same job hides behind. An “Account Executive” may be called “Business Developer”, “Key Account Manager” or “Sales Executive” depending on the company. A “React Developer” may describe themselves as “Front-end Developer” or “Software Engineer”.

2. The skills that really matter

Separate must-haves from nice-to-haves. “B2B SaaS sales” is a filter; “HubSpot experience” is a bonus. The agent weighs criteria differently if you tell it which ones are non-negotiable.

3. The level of experience

A range beats a single number: “at least five years” or “between three and eight years”. Say whether the experience must be in a given industry or with a given type of client.

4. Location and way of working

City, region, acceptable radius, remote or not. It is often the criterion that filters out the most profiles: be precise.

5. The context

Company size and industry, type of team, what is at stake in the role. Context helps the agent prefer someone from a start-up for a start-up, or from a large group for a large group.

Three examples you can reuse

SaaS sales

I’m looking for a Senior Account Executive in B2B SaaS, with at least five years selling to mid-sized companies. Based in London or remote. Fintech experience is a plus.

Developer

Find me a senior React developer for a twenty-person start-up in Manchester. Hybrid is fine. TypeScript is a must, Node.js is a plus.

Healthcare

I need a theatre nurse in Bristol or within twenty miles, able to start soon, ideally with private hospital experience.

Each one covers the five elements, without jargon. And each sentence helps the agent justify its choices: “exact title”, “five years of SaaS sales”, “based in London”.

The most common mistakes

• Pasting a whole job description without rereading it. Job descriptions often mix essentials and decoration (“team spirit”, “dynamic”). Let the agent extract the criteria, then check what it kept.
• Stacking criteria. Ten mandatory criteria often means zero profiles. Keep three or four non-negotiables and put the rest as “preferred”.
• Forgetting title variants. It is the number one cause of shortlists that are too short.
• Describing the person rather than the career. The agent works on public professional backgrounds: describe what someone has done, not their personality.
• Not iterating. The first result is a starting point. If the shortlist leans too far one way, change one sentence and run it again.

Reading the results to improve the prompt

The agent’s reasons are your best tuning tool. If several well-scored profiles share the same flaw (an off-topic industry, a location too far away), add it to the prompt as an exclusion. If interesting profiles come back with an average score, a criterion was probably too strict: loosen it.

In two or three rounds, you get a shortlist that looks like the one you would have built by hand, in a fraction of the time.

Type it, or say it

One last tip: you do not have to type. Describing a need out loud, as you would on the phone to a colleague, often gives a more natural and complete prompt. With Profilr, you can type or dictate your search, and even run it from Claude, ChatGPT or Gemini thanks to MCP access, included from the Pro plan.

To see how the agent handles a prompt from start to finish, read AI sourcing agent: how it works, or book a demo with one of your job descriptions.