Write Job Descriptions That Attract Better Candidates With Claude

Claude can produce a first-draft job description, but only if you give it the facts and a clear brief. Anthropic's own prompting guidance recommends 3-5 examples, explicit context and structured tags, and each of those maps onto a job description task. The draft is the easy part. The pay range, the real requirements and the legal check still have to come from you.

This post gives a prompt template built from Anthropic's published advice, a bias review step, and the US rules on pay and advertising language as of 16 October 2026. Treat it as a structured method, not a proven result.

How do you get a good job description out of Claude?

Give Claude a role, the facts of the job, one or two good examples and a precise output format. Anthropic's documentation says Claude responds well to clear, explicit instructions, and compares it to a new employee who lacks context on your norms and workflows.

The docs also offer a "golden rule": show your prompt to a colleague with minimal context and ask them to follow it. If that person would be confused, Claude will be too. For HR, the test is simple. Could a new recruiter write a decent job description from your prompt alone?

Anthropic's other points that apply here:

Source for all rows: Anthropic, Claude Platform Docs, accessed 16 October 2026.

What facts should you collect before you prompt?

Collect the inputs Claude cannot guess: the pay range, the real must-haves, the duties of the first 90 days, the location rules and the application deadline. Claude has no access to your compensation bands or your hiring manager's real expectations, so anything you leave out it will fill with generic language.

A short intake list that works for most roles:

1. Job title, level, team, reporting line and work location (including remote rules).
2. The pay range and what is included (base, bonus, benefits).
3. Five to seven real duties, in order of time spent.
4. Three to five must-have skills, and a separate list of nice-to-haves.
5. What success looks like at 90 days, in the manager's words.
6. Application steps and, where required, the closing date.
7. One or two past ads from your company that performed well, as examples.

Keep personal data out. A job description needs role information, not names, reviews or details of the person who held the job before. Check your company's AI policy before pasting internal documents into any tool.

What prompt should you paste into Claude?

Use the template below. It follows the structure Anthropic describes: a role, tagged inputs, examples and a clear output format. Put long documents such as past ads or a manager's notes near the top, and the instructions at the end. Anthropic's docs say placing the query at the end can improve quality by up to 30 percent in its tests, and the figure is a company-reported result on complex, multi-document inputs.

You are a recruiter who writes plain, specific job descriptions
that a candidate can read in two minutes on a phone.

<examples>
<example>[Paste a past job ad you consider good]</example>
<example>[Paste a second one for a different role]</example>
</examples>

<role_facts>
Title and level: [ ]
Team and reporting line: [ ]
Location and remote rules: [ ]
Pay range and what it includes: [ ]
Main duties, in order of time spent: [ ]
Must-have skills (max 5): [ ]
Nice-to-have skills: [ ]
What success looks like at 90 days: [ ]
How to apply and closing date: [ ]
</role_facts>

<instructions>
Write a job description with these sections: Summary (3 sentences),
What you will do, What you need, Nice to have, Pay and benefits,
How to apply.
Use only the facts in <role_facts>. If a fact is missing, write
"[MISSING: item]" instead of inventing it.
Describe duties as actions with outcomes. Keep requirements to the
list I gave. Write in everyday words at a general-reader level.
Use the same tone as the examples, but do not copy their sentences.
Under 450 words.
</instructions>

The "[MISSING: item]" line matters. It gives Claude a positive instruction for gaps, which matches Anthropic's advice to say what to do instead of what not to do. It also makes invented requirements easy to spot.

How do you check the draft for biased or exclusionary wording?

Run a second prompt that reviews the draft, then read the flagged lines yourself. The legal standard is set by the EEOC: it is illegal to publish a job advertisement that shows a preference for, or discourages, someone from applying because of race, color, religion, sex (including transgender status, sexual orientation and pregnancy), national origin, age (40 or older), disability or genetic information.

The EEOC gives two examples: an ad that seeks "females" or "recent college graduates" may discourage men and people over 40 from applying. A review prompt could read: "List any phrase in this job description that favours or discourages an age group, gender, disability status or national origin. Quote each phrase and suggest a neutral rewrite."

Two published sources show what to look for:

• Gendered wording. A 2011 study in the Journal of Personality and Social Psychology found that ads for male-dominated areas used more masculine wording such as "leader, competitive, dominant". In experiments, women found jobs less appealing when ads leaned masculine, and a sense of belonging, not perceived skills, explained the effect.
• Age code words. AARP has pointed to "digital native" as a possible code word for age bias in postings.

A model's review is a screen, not a legal clearance. A person with HR or legal responsibility should approve the final text.

What pay and legal details must the job description include?

It depends on location. In New York City, employers with four or more employees must include a pay range in job ads, a rule in place as of 1 November 2022. It covers jobs, promotions and transfers in the city. Temporary help firms are not required to comply, but employers working with them must.

In Colorado, the state labor department says Part 2 of the Equal Pay for Equal Work Act sets requirements to include compensation, benefits and application deadlines in job postings, and to keep job description and wage rate records.

Other states are moving the same way. According to recruitment advertising vendor Joveo (published 29 September 2026), Illinois, Minnesota, New Jersey, Vermont and Massachusetts require ranges alongside earlier adopters California, Colorado, New York and Washington, with employer-size thresholds that differ by state. Joveo itself says to confirm the current statute for each state where you hire.

Put the pay range in the <role_facts> block. Do not ask Claude to propose one. A model has no access to your pay bands, and a range that is wrong creates a compliance problem. Do not use this post as legal advice; confirm with counsel or the relevant agency.

How do you edit and test the final draft?

Treat the output as a first draft and review it in three passes: facts, fairness and readability. Every requirement should match something the hiring manager confirmed, and every "[MISSING]" tag should be resolved or removed.

A short review routine:

1. Facts pass. The hiring manager confirms duties, must-haves and reporting line.
2. Pay pass. HR confirms the range and benefits lines match the pay band and the local posting rules.
3. Fairness pass. Run the review prompt above, then check each flagged phrase by hand.
4. Cold read. Ask someone outside the team what the person will do in week one. If they cannot say, rewrite the duties.
5. Save the winner. When a posted ad performs well, add it to your <examples> block for next time.

Once the text is final, it needs to reach candidates through your applicant tracking system and job boards. For that side of the process, see our comparison of Greenhouse, Lever and Ashby, the guide to free ATS options for first hires, the data on which job boards bring real hires and the piece on what skipping compliance tools can cost.

What's still unclear?

Several things are not established by public sources.

• Effect on applicant quality. This post cites no evidence that AI-drafted job descriptions raise applicant quality or volume. The method above is a way to write clearer ads faster, not a proven way to get better hires.
• Vendor claims. Anthropic's 30 percent figure is a company-reported test result on a specific type of input. It may not match a short job description task.
• State coverage. The multi-state list comes from one vendor blog. Thresholds, effective dates and penalties differ, so check each statute or agency page.
• Bias checks. A model's flag list is not a validated word list, and word lists alone do not guarantee a fair ad.
• Scope. This post covers drafting only. Using AI to screen or rank candidates is a different activity with its own rules; see our comparison of Paradox and HireVue for that side.

The weakest point of the argument is simple: a clear, well-written ad does not fix an unclear role. If the hiring manager cannot say what the person will do and what the job pays, Claude will produce polished text around the gap.