Writing Job Descriptions for AI Roles That Actually Attract

Senior AI people read job ads the way engineers read code: scanning for signal, discarding boilerplate, and closing the tab at the first red flag. Here is the structure that survives that scan.

Marco Reyes·Head of GEO & Growth, Aiporate··7 min read·Share on XLinkedIn

Key takeaways

  • Senior AI candidates scan for four things in under a minute: the actual problem, the data and stack reality, who they would work with, and the salary range, and vague answers to any of them read as answers.
  • Lead with the problem, not the company boilerplate: the first paragraph should describe the technical problem the hire will own, specifically enough that the right person becomes curious.
  • Honest tech-stack disclosure beats aspirational stack lists: stating what is genuinely modern, what is legacy and what is greenfield builds more trust than a wall of logos, and pre-filters for people who want that reality.
  • A salary range is both a filter and a signal: ads with ranges attract more qualified applicants, and in more and more jurisdictions transparency is becoming law anyway, hiding the range now signals below-market pay.
  • Most repellent red flags are unforced errors: unicorn requirement lists, "rockstar/ninja" language, five-round unspecified processes, and mission statements with no technical substance.

The strongest AI candidates are almost never actively job hunting, and when they do read an ad, they give it the same scan they give a pull request: thirty seconds, looking for concrete signal, primed to close the tab at the first cliché. "Fast-paced environment," "AI rockstar," fifteen required frameworks, no salary, no named problem: tab closed. A job description for an AI role is a technical document with a marketing function, and it wins by being specific where every competing ad is vague. This article covers what senior AI people actually scan for, the problem-first structure that holds their attention, and the red flags that repel them before you ever know they were there.

What senior AI people actually scan for

Eye-tracking an experienced candidate through your ad would show them skipping the intro paragraph entirely and hunting for a handful of concrete facts. If those facts are missing, the gaps get filled with the pessimistic default.

What they look forQuestion in their headIf it's missing, they assume
The problem"What would I actually work on?"There is no defined problem; I'd be the AI figleaf
Data & stack reality"Is there real data and infrastructure, or slides?"Data is a mess and I'd spend a year on plumbing alone
The team"Who would I learn from and report to?"I'd be the only technical person, unsupported
Salary range"Is this even in my bracket?"Below market, and they know it
Deployment evidence"Has anything shipped here?"Eternal proof-of-concept land
The senior candidate's scan, and what missing answers signal

The problem-first structure

This structure front-loads the scan targets and pushes company boilerplate to the end, where it belongs. It fits on one page.

  1. 1The problem (3-4 sentences, first thing on the page): the concrete technical challenge, its scale, and why it is hard. "Our demand forecasts run on a 5-year-old gradient-boosting pipeline that breaks on promotions; you will own its successor end to end."
  2. 2What you will do (5-6 bullets, verbs first): real activities in the first year, including the unglamorous ones, honesty here pre-sells the reality.
  3. 3What you bring (max 6 must-haves, outcome-phrased): taken directly from the requirements profile, plus a short, clearly separated nice-to-have list.
  4. 4Stack and data reality (honest, 3-4 lines): what is modern, what is legacy, what is greenfield, and what the data actually looks like.
  5. 5Team and reporting line: who they work with day to day, who they report to, how many engineers/scientists are around them.
  6. 6Salary range, location policy, process: the range, the remote/on-site reality, and the interview process with the number of rounds and total time commitment.
  7. 7About the company (short, last): two or three sentences, after everything the candidate actually came for.

Honest disclosure: stack, data, salary

Every ad competes with dozens claiming a "modern cloud-native stack" and "data-driven culture." Verifiable honesty is the differentiator, and it filters in the people who want your actual job, not the imaginary one. Salary belongs in this section too: ranges measurably increase qualified application rates, EU pay-transparency rules are making disclosure mandatory across member states, and senior candidates increasingly read a missing range as a lowball in waiting.

Aspirational (reads as spin)Honest (reads as signal)
"Modern, cloud-native ML stack""Models run on a managed cloud platform; feature pipelines are solid, experiment tracking is basic, you would help choose what we adopt next"
"Huge amounts of data""About 40M transactions across three systems; joined and cleaned for two core use cases, raw elsewhere" (illustrative figures)
"Competitive salary""EUR 85,000-105,000 depending on level, plus the criteria that decide where in the range you land"
"Work with cutting-edge AI""Two models in production, one LLM feature in pilot; your first project is taking the pilot to production"
Aspirational phrasing vs. honest phrasing that builds trust

Red flags that repel senior candidates

These are the tab-closers. Each one is common, each one is avoidable, and each one costs you candidates you never see.

  1. 1The unicorn list: ten-plus required skills spanning research, engineering and DevOps signals the company does not know what it needs, seniors read it as guaranteed role confusion.
  2. 2"Rockstar," "ninja," "guru," "AI wizard": personality-cult vocabulary reads as a culture that rewards heroics over engineering.
  3. 3No salary range in 2026: increasingly read as "below market and hoping you won't ask," and soon simply non-compliant in much of the EU.
  4. 4Vague process ("several interviews"): seniors budget their time; an unspecified process signals a disorganized one.
  5. 5Mission-only ads: three paragraphs of vision with no named problem, stack or team tells a technical reader there is nothing underneath.
  6. 6Buzzword stacking ("GenAI, AGI-adjacent, blockchain-ready"): technology name-dropping without coherence signals leadership chasing trends, not solving problems.

A miniature example: the opening that survives the scan

An illustrative opening block for a fictional logistics company, the first thing a candidate reads. Note that it answers three scan targets in the first five lines.

ElementText (condensed)
Problem lead"Our route-planning runs on hand-tuned heuristics that leave an estimated 8-12% efficiency on the table (internal estimate). You will own the ML system that replaces them, from first model to production rollout across 40 depots."
Stack honesty"Python services on AWS; clean telemetry data for 18 months, older data patchy. No ML in production yet, you would be first, with a platform engineer dedicated to the rollout."
Team"You join a data team of four (two engineers, one analyst, one PM), reporting to the Head of Data, who ships code herself."
Range & process"EUR 90,000-110,000 by level. Process: intro call, technical deep-dive on a past project, half-day on-site with the team, offer, three weeks end to end."
Illustrative example: opening block of an ML engineer ad (fictional company)

Common mistakes when writing the ad

  1. 1Writing the ad before the requirements profile exists, so the ad becomes the spec, backwards and unprioritized.
  2. 2Letting HR templates dictate structure: company history first, problem last, exactly inverted from how candidates read.
  3. 3Listing every tool anyone on the team uses as a requirement instead of describing the stack as context.
  4. 4Hiding the seniority level: "senior" in the title with junior scope in the bullets wastes everyone's process time at offer stage.
  5. 5Never updating the ad: a description that does not change after learnings from the first ten conversations is a template, not a document.

Frequently asked questions

Should we really publish the salary range?

Yes. Ranges measurably increase qualified applications, senior candidates increasingly skip ads without them, and EU pay-transparency rules are making disclosure a legal requirement across member states. The only strategic question left is how wide the range is, not whether to show it.

How long should an AI job description be?

One page of substance. The scan targets, problem, responsibilities, requirements, stack reality, team, range, process, fit comfortably in 400-600 words. Longer ads usually mean the requirements profile behind them was never prioritized.

How honest should we be about legacy systems and messy data?

More honest than feels comfortable. Candidates discover the reality in the first week anyway; discovering it after signing is how early attrition happens. Framing matters: legacy plus a mandate to modernize is genuinely attractive to many senior engineers, hidden legacy is attractive to no one.

What if we cannot name the exact problem yet?

Then the honest ad says so: "first mandate: assess these three candidate use cases and build the first one." That is a real, attractive senior brief. What repels is pretending certainty with vague mission language, seniors can tell the difference instantly.

Head of GEO & Growth, Aiporate

Marco leads generative engine optimization and organic growth at Aiporate. He has run search and content strategy through the shift from ten blue links to AI answers, and helps SaaS brands stay visible where buyers now decide, inside the models.

Need the team to make this real?

Describe your need in plain English, get the exact hire, forward-deployed talent or a fractional leader, vetted and matched in 72 hours.

Scope your need →

Keep reading

The Weekly Brief

Intelligence for building AI-native organizations.

One email a week: the sharpest thinking on AI hiring, infrastructure, teams and strategy, for the people building the future of work.

Join operators, founders and CTOs. No spam, unsubscribe anytime.