How to write a job description your screening AI can actually score

Most job description advice is written for one reader: the candidate deciding whether to click apply. That advice is mostly good. Put the pay range near the top. Cut the internal jargon. List only the requirements you’d actually reject someone over. A job search report from Resume Genius found that 83% of job seekers say vague responsibilities or unclear requirements in a job description signal a disorganized hiring process, so tightening the posting is worth doing on its own.

None of that advice was written for a job description for AI screening, which is what you’re actually writing the moment you run applications through screening software. Your job description already has two audiences: the candidate deciding whether to apply, and the AI scoring them against whatever you gave it to work with. Optimize only for the human one, and you’ll get a shorter, cleaner posting that scores almost every candidate the same way, because you never told the AI what to look for either.

Your job description has a second reader now

A job description used to have one job: get the right person to apply and the wrong person to skip it. Once you plug that same text into a screening tool, it picks up a second job it was never written for. It becomes the input an AI reads to figure out what “good” looks like for this specific role.

That’s not a minor technical detail. Most screening tools build a screening position the same way: a job description first, then an intake step asking what success looks like in the first six to twelve months and what traits help someone fit your team. Whatever you write in both becomes the criteria every resume, interview answer, and assessment result gets measured against.

Most owners don’t think about the job description this way, because they’ve written dozens of them as pure advertising copy. It’s a habit that made sense for years and quietly breaks the moment an AI starts reading the same document as a rubric instead of a pitch.

Generic descriptions are why every match score looks the same

Here’s what that break actually looks like in practice. Write a job description the way every guide recommends: concise, a handful of non-negotiable requirements, outcomes stated in general terms. Feed it into an AI screener along with fifty or sixty resumes and interview responses. What comes back isn’t wrong exactly. It’s just useless. Candidates cluster in a tight band, mostly in the 60s and 70s, and nothing in the ranking tells you who’s actually different from whom.

The AI isn’t confused. It’s doing exactly what you told it to do, which was not much. A Totaljobs study cited by recruitment researchers found a real gap between how recruiters and candidates experience the same job posting: 74% of recruiters believe a clear job description improves the quality of applications they get, while 69% of job seekers say the expectations in recent job ads still felt unclear to them. Both groups are describing the same underspecified document from two sides. The AI ends up on the recruiter’s side of that gap with even less to go on, because it only has the text, not years of intuition about what “team player” or “fast-paced environment” is supposed to mean.

Where the specifics actually disappear

Three places, consistently:

• The requirements section gets trimmed to the bone. Good advice for candidates, since a shorter list keeps more of the right people applying. Bad for scoring, since “3+ years experience” and “customer service background” don’t give an AI much to differentiate people on.
• Success is described in adjectives, not outcomes. “Fast learner” and “great communicator” show up in almost every job description ever written. They don’t tell a screening tool, or a person, what winning in this role actually looks like six months in.
• The intake step gets skipped or rushed. It’s the extra five minutes between writing the ad and hitting publish, and it’s usually the first thing an owner skips when they’re trying to get a role live before the workday ends.

Write the ad short. Write the rubric specific.

The fix isn’t “write longer job descriptions.” Every piece of advice about keeping postings concise and scannable is still correct, because the candidate reading it hasn’t changed. What changes is realizing there are two different documents doing two different jobs, even when they live in the same form.

What stays on the public posting

The candidate-facing posting stays short: title, pay range, the two or three requirements that would actually disqualify someone, a couple of sentences on what the role does. That’s the ad. It’s for the candidate deciding whether to spend fifteen minutes recording a one-way interview for you.

What actually needs to be specific

The intake is the rubric, and it’s where the specificity has to live. On Truffle, this is two questions most owners blow past: what does success look like in the first six to twelve months, and what traits help someone actually thrive on this particular team versus the traits that tend to sink people. Filling those in with real specifics, not adjectives, is what improves match accuracy by up to 25% on Truffle. That’s not a small tweak. It’s the difference between an AI that’s guessing at your standard and one that’s actually measuring against it.

Match scores themselves are built from a handful of concrete criteria: how well someone’s answers align with the role, their apparent motivation, communication style, intent, collaboration, and creativity. Every one of those gets pulled from what you put into the job description and the intake, not invented by the AI on its own. Vague inputs produce vague weighting across all six. Specific inputs let the scoring actually separate a 92% from a 68%, instead of leaving everyone bunched in the middle.

You’re not making the posting longer, you’re filling in a different field

The objection here is obvious: doesn’t this contradict the concise, scannable advice everyone else gives, including the 83% of candidates who say vagueness signals disorganization? It doesn’t, once you separate where each kind of detail lives.

The posting candidates see never gets longer. The intake fields they never see as prose do the calibrating instead. You’re not choosing between a good candidate experience and a usable match score. You’re recognizing that the ad and the rubric were always two different things, even back when you wrote job descriptions on a single page and both audiences were people.

This is also where an inclusive, well-written posting and a well-calibrated intake work together rather than against each other. A posting that lists only genuine requirements naturally produces a wider, more varied pool. A specific intake is what lets you actually make sense of that wider pool once it arrives, instead of drowning in it the way most owners do when they handle too many applicants with nothing but a spreadsheet and a gut feeling.

What this looks like when you actually do it

The generic version

Say you’re hiring a front-desk coordinator. The generic version of the intake says: friendly, organized, good with people. Fed into any AI screener, that’s three adjectives every candidate’s resume and interview will superficially claim to match, and the scores prove it: a wall of candidates sitting between 65% and 78%, no real signal about who to call first.

The specific version

Now write the same intake the way you’d brief a recruiter you were paying to fill the role. Success in six months looks like: the front desk answers the phone within two rings during the lunch rush, and member complaints about scheduling mix-ups drop to near zero. The trait that’s mattered most on this specific team: someone who double-checks their own work without being asked, because the last two hires who didn’t have caused double-bookings that cost you client trust. The trait that tends to sink people here: anyone who needs constant direction, because there’s rarely a manager on the floor to give it.

That’s not a longer job posting. It’s the same three or four sentences, aimed at a different question. Feed that into the same screening flow and the AI has something to actually check answers against, not just agree with. The Question-Level Evaluation on each interview response starts citing specifics (“mentions catching a scheduling error before it became a problem” versus “says they’re detail oriented”) instead of restating the same three adjectives back at you. Candidates spread out. A few clear 90%. Most sit in the 70s. A handful genuinely don’t fit, and now you can see why instead of guessing.

Truffle is a candidate screening platform that combines resume screening, one-way video interviews, and talent assessments, and all three read from the same job description and intake. Get specific once at the front of that flow, and the resume review, the interview scoring, and any assessment you layer in are all measuring candidates against the same real standard instead of three different guesses at one.

The job description was never just an ad

Once an AI is scoring against it, your job description stops being an ad and starts being an answer key. That reframe changes more than one posting. It changes how you write qualification questions, how you brief anyone helping you screen, even how you talk to a hiring manager before a role goes live, because all of it is now instruction to a reader that isn’t a person.

Writing well for that reader is a learnable skill, and it looks less like copywriting and more like briefing a recruiter you don’t have. You’re not deciding for anyone. You’re giving the tool doing your first read the same standard you’d give a person, so the shortlist it hands back actually means something when you sit down to read it.

Frequently asked questions about writing job descriptions for AI screening

Does a more detailed job description slow down candidates or hurt completion rates?

Not if you keep the split. The public posting stays short and scannable, since that’s still what gets candidates to apply. The added specificity goes into intake fields the AI uses for scoring, not into a longer wall of text candidates have to read first.

What should I actually put in the intake beyond the basic requirements?

Two things matter most: a concrete picture of what success looks like in the first six to twelve months, stated as an outcome rather than an adjective, and the specific traits that have helped or hurt people on this exact team before. Both give a screening tool something real to check candidates against.

Will this work for a role I’ve never hired for before?

It’s harder without a pattern to draw on, but you can still be specific about the job itself: what the person will actually do in a typical week, what a good first ninety days produces, and what would make you regret the hire. Specific and hypothetical still beats vague and generic.

Can I update the job description and intake after candidates have already started applying?

On Truffle, most of a live position stays editable, including intake answers and qualification questions, so you can sharpen your criteria mid-stream if the first batch of match scores isn’t telling you much. The position title, core job description, and response format lock once the position goes live, so get those roughly right before you publish and refine the rest as you learn.