Prompt Engineering in Recruitment: What Most Sourcers Get Completely Wrong

The Recruiter's Guide to Prompt Engineering for Talent Search

I started my PR career when CVs still landed on my desk with a thud rather than a digital ping. Back then, 'sourcing' meant rifling through a Rolodex of journalists and praying someone picked up the landline. Now? We're feeding text prompts to machines that can scan thousands of profiles in seconds.

Speaking with recruiters these past few months, it's staggering how many are still treating AI sourcing tools like basic keyword search. They plug in "PR Account Director London" and wonder why they're drowning in an ocean of irrelevant results.

If you're in recruitment and haven't mastered prompt engineering, you're functionally illiterate in 2026.

What Nobody Tells You About AI Sourcing

I met with a former colleague last week who's now heading talent for a mid-sized PR agency. She confessed they'd been using the same generic prompts since implementing their fancy AI sourcing tool six months ago. The results? Mediocre candidates and frustrated hiring managers.

What had she missed? The hidden architecture of effective prompts that actually produce results worth paying for.

But first, let's demolish a persistent myth: prompt engineering isn't about learning some magical incantation or secret code. It's about thinking carefully about what you actually want - and then communicating that with precision.

The problem is most recruiters' prompts read like they're ordering coffee. "I want a candidate with strategic comms experience and crisis management skills. Must have worked with tech clients."

That approach worked fine with human researchers. It fails miserably with AI.

The Structure of High-Performance Prompts

Any decent prompt for talent search has four components:

1. Role context - What the candidate will actually do day-to-day
2. Performance indicators - What success looks like in this role
3. Exclusion criteria - What you explicitly don't want
4. Format instructions - How you want results delivered

Most recruiters get the first one right. They utterly botch the rest.

Let me share a prompt template that's worked consistently for specialist PR roles:

"Find PR professionals with expertise in [specific sector] who have led campaigns resulting in [concrete outcomes]. They should have experience managing [specific stakeholder types] and navigating [industry-specific challenges]. Exclude candidates who have only worked in-house OR who have jumped between agencies more than twice in the past three years. Present results with each candidate's most notable campaign achievement first, followed by tenure history."

See the difference? This isn't just asking for skills - it's directing the AI to evaluate candidates based on outcomes and stability patterns while providing clear exclusion criteria.

Specificity Trumps Breadth

Here's where most sourcers self-sabotage. They think broader prompts cast wider nets. In reality, the opposite is true.

Look at these two prompts:

"Find PR professionals with healthcare experience"

vs.

"Find PR professionals who have managed communications for pharmaceutical product launches, specifically those involving regulatory approval processes in post-Brexit Britain"

The first will drown you in hundreds of candidates who once wrote a press release about hand sanitiser. The second might deliver seven candidates - but they'll be the right seven.

Context Is Everything

Most AI sourcing tools now accept what I call "contextual padding" - background information that frames the search without being part of the search itself.

This makes a massive difference.

Compare:

"Find PR directors with experience in sustainability communications"

vs.

"My client is a renewable energy startup preparing for Series B funding. They've faced criticism for greenwashing from environmental groups. They need a PR director to rebuild credibility with investors and environmental stakeholders. Find candidates with relevant experience."

The second prompt gives the AI crucial context about the actual business problem that needs solving, not just a job title and keyword.

Negative Prompting: The Underused Technique

One technique that's transformative but underused is negative prompting - explicitly telling the AI what you don't want.

I was helping a boutique agency find a consumer PR specialist recently. After getting flooded with corporate communications candidates, I rewrote the prompt to include: "Exclude candidates whose experience is primarily in corporate communications, public affairs, or internal communications. The ideal candidate should not have worked exclusively with luxury brands."

The quality of matches immediately improved.

So who's doing this well? From conversations with industry colleagues, specialist agencies focused on hard-to-fill roles seem to be developing the most sophisticated prompting strategies. They're investing time in prompt libraries and refining based on feedback loops.

The Technical Mechanics

Prompt engineering isn't just about what you ask but how you structure the ask. Syntactical precision matters enormously.

For instance:

"Find PR candidates with international experience who've worked on award-winning campaigns"

This could mean:

• Candidates with international experience who ALSO worked on award-winning campaigns
• OR candidates with either international experience OR award-winning campaign work

Better version:

"Find PR candidates who meet ALL of the following criteria: 1) Have worked on accounts across at least two different countries 2) Have contributed to campaigns that won industry recognition through awards 3) Can demonstrate measurable results from their international work"

See how that clarifies the Boolean logic? It's unambiguous to both human and machine readers.

The Format Directive

A criminally overlooked aspect is telling the AI how to present results. Without this guidance, you'll get whatever default format the tool uses - usually not optimised for quick assessment.

Add this to your prompts:

"Format results as a table with these columns: Name | Standout Achievement | Years in PR | Current/Last Employer | Specialisation | Red Flags (if any). Sort by relevance to the specific sector experience mentioned above."

This simple addition can save you hours of manual filtering.

Practical Examples That Work

Here's a prompt I used recently for a senior consumer healthcare role that produced exceptional results:

"I need PR directors or associate directors who have led consumer health campaigns that achieved measurable behaviour change outcomes. The ideal candidate has navigated UK healthcare regulatory frameworks when promoting prescription or OTC products. They should have experience briefing and managing medical key opinion leaders. Exclude candidates who focus primarily on pharma investor relations or internal comms. Present results with each candidate's most successful campaign ROI metrics first, followed by regulatory experience summary."

This prompt worked because it:

• Focused on outcomes, not just experience
• Specified the regulatory context
• Included clear exclusion criteria
• Directed the formatting toward what mattered most

Of course, your mileage will vary depending on which sourcing platform you're using. Some of the major recruitment platforms have built impressive custom sourcing tools with their own query language quirks.

The OHub's talent search platform has developed one of the more intuitive interfaces for PR recruitment specifically, with contextual prompt builders that guide you through this process.

The Limitations You Need to Know

Prompt engineering isn't magic. Even the best prompts will occasionally deliver duds, especially for roles requiring nuanced cultural fit assessment.

And let's be brutally honest - AI sourcing can replicate and amplify existing biases if your prompts aren't carefully constructed. The ICO published its "Recruitment Rewired" report on 31 March 2026, following engagement with over 30 employers between March 2025 and January 2026. The report found widespread gaps in bias monitoring across automated hiring tools and sets out clear regulatory expectations for organisations using AI in recruitment. According to Bird & Bird's analysis of the ICO's findings, over 70% of organisations anticipate increasing their use of AI in recruitment over the next five years, making the ICO's guidance on bias, transparency and candidate rights essential reading now, not later.

The machines aren't replacing good recruiters. They're just changing what "good" means in 2026. The recruiters who thrive won't be the ones who fight the technology, but those who become fluent in directing it.

If I could leave you with one actionable tip: start keeping a prompt journal. Document what works, what bombs, and how you've refined your approach. Six months from now, you'll have a personalised playbook that no competitor can copy.

Your prompts aren't just search queries. Six months of refinement, feedback loops and documented failures add up to a sourcing playbook no competitor can replicate. That's intellectual property worth protecting.

Sophie Chen has led consumer PR campaigns across London agencies for 12 years and now advises recruitment firms on specialist PR talent acquisition strategies.