How to Search Your Database Like You'd Ask a Colleague

A while ago we compared boolean search with AI search. Short version: for everyday recruiting, typing a sentence beats building a query string.

But "type a sentence" raises an obvious question. Which sentence?

We see the full range in Adeptiq. Some recruiters get great results from day one. Others type "java developer", get a wall of profiles, and conclude AI search is overrated. The difference is rarely the database. It's the question.

Here's the mental model that fixes it: imagine walking over to a colleague who knows every candidate you've ever met, and briefing them. That's the query.

Describe a person, not a keyword

A colleague wouldn't know what to do with "java developer" either. Which one? For what?

Compare:

java developer

with:

Java developer with at least 5 years of experience, has worked in banking or insurance, based within commuting distance of Antwerp

The first is a keyword. The second is a brief. It tells the search who you're actually looking for: the seniority, the industry background, the location. AI search uses all of it, and it understands that someone whose CV says "backend engineer at KBC" is a strong match even though the words "java developer in banking" appear nowhere.

One or two sentences is usually enough. Role, seniority, one or two hard requirements, location if it matters. Done.

What works

Plain requirements. Write them the way you'd say them. "Fluent in French and Dutch." "Available for freelance." "Has managed a team." No special syntax, no field names.

Industry and context. "Sold SaaS to SMEs" finds different people than "sales". Context is where AI search beats keyword matching by the widest margin, because relevant candidates rarely use your exact words in their CV.

Iterating. Your first search is a draft. Look at the results, notice what's off, and adjust: "same, but only profiles with agency experience" or tighten the location. Treat it like a conversation, because that's what it is.

The read-aloud test. Say your query out loud. If a colleague would understand it, the search will too. If a colleague would ask "what do you actually mean?", refine it first.

What doesn't work

The kitchen sink. Twelve requirements in one query dilutes everything, because almost nobody matches all twelve and the search has to guess which ones you care about. Pick the three or four dealbreakers, search, then narrow down. Nice-to-haves come later.

Vibes. "A real hunter with great energy" finds nothing useful, because CVs don't record energy. Translate the vibe into evidence: "outbound sales experience, comfortable with cold calling, short sales cycles". If it leaves a trace on a CV, you can search for it.

Mind reading. Salary expectations, willingness to relocate, reasons for leaving: none of that is in the CV, so no search will find it. Use search to build a shortlist of plausible people. Use a phone call for the rest.

Old habits. We still see queries full of AND, OR, quotes and brackets. It won't break anything, but you're doing work the AI already does for you. Synonyms, related job titles, spelling variants: handled. Just write the sentence.

Prompts to steal

Adapt these to your own roles:

• "Senior accountant with Big Four experience, fluent in Dutch and English, in or around Ghent"
• "Sales profile who has sold software to SMEs, at least 3 years of experience, comfortable with cold outreach"
• "Project manager from the construction sector who has run projects above 5 million euro"
• "German-speaking customer support profile with e-commerce experience, open to remote work"
• "Freelance React developer with agency experience, available in Belgium"
• "Warehouse supervisor who has managed shifts of 20 or more people, near Liège"

Notice what they have in common: each one reads like the first sentence of a real conversation about a real vacancy.

The point

Boolean search made you learn a syntax. AI search asks you to drop one. If you can brief a colleague, you can search your database, and the candidates you'd forgotten about are usually already in there.

Try it on your own data. Start for free, import a stack of CVs, and ask for the person you need.