AI Prompts for Recruiting and Staffing Agencies

TL;DR: Recruiting agency AI works best on writing tasks: job ads, search strings, screening questions, outreach and client updates. Below are 14 copy-paste recruiter prompts, including bias-check prompts and what they can and cannot catch. Keep a recruiter reviewing every output, never let a model reject candidates on its own, and check hiring-AI rules with counsel.

A staffing desk runs on words. Intake notes become job ads, job ads become search strings, search strings become outreach, and calls become shortlists a client can read in two minutes. That is why recruiting agency AI adoption has been fast: most of the day is drafting, and drafting is what language models are good at.

The same models can also carry bias from their training into the decisions that matter most, and in several places the law has started to treat hiring tools as a special case. So this guide splits the work into two groups. The first is the writing where AI saves time with little risk. The second is anything that touches who gets considered, where the prompts are built to support a recruiter's judgment rather than replace it.

A note on scope: this page explains prompts and points to public sources. It is not legal advice. Employment law differs by country, state and city, and your agency's counsel is the right person to say what applies to you.

Where does AI help a recruiting agency most?

In the tasks where a recruiter can check the output quickly and the cost of a mistake is small.

The last row is the one to be careful with, and the section on bias explains why.

Which AI prompts help with sourcing?

Sourcing prompts turn a messy intake into something you can search with and post. They need no candidate data at all, which also makes them the safest place to start.

Prompt 1: Intake notes to a structured brief

You are a recruiter at a staffing agency. Turn these client intake notes
into a structured role brief.

NOTES:
[paste intake call notes]

Output:
- Job title (and 2-3 alternative titles candidates actually use)
- Must-have requirements (only those the client stated as essential)
- Nice-to-have requirements
- Pay range, location, work pattern (write "NOT STATED" if missing)
- Questions to send back to the client

Do not add requirements the client did not mention. Flag any requirement
that is not clearly job-related, such as age, years since graduation,
or "cultural fit" without a definition.

Prompt 1 earns its place before anything else is written. Most search problems start as intake problems: a client says "senior" and means five years, or lists a nice-to-have as essential. Getting the brief right, and sending the gaps back as questions, saves a week of sourcing the wrong people.

Prompt 2: Search strings

Using this role brief:
[paste brief]

Write:
1. A Boolean search string for LinkedIn Recruiter-style search
2. A simpler version for a job board with basic keyword search
3. A list of synonym groups I can add or remove

Use only skills, titles and qualifications. Do not include terms that
target or exclude people by age, gender, ethnicity, nationality,
religion, disability or any other protected characteristic, directly or
by proxy (for example, graduation years or specific clubs).

Test every string before you trust it. Models produce Boolean syntax that looks right and sometimes uses operators a given search tool does not support, or nests brackets in a way that silently changes the logic. Run it, look at the first page of results, and adjust. The proxy rule in the prompt matters too: excluding people by graduation year is excluding people by age, whether or not the word appears.

Prompt 3: Job ad rewrite

Rewrite this job ad so it is clear, specific and inclusive.
[paste draft ad]

Rules:
- Lead with what the person will do, then what they need.
- Keep must-haves to the essentials from the brief.
- Include pay, location and work pattern if provided.
- Replace jargon and vague phrases ("rockstar", "work hard play hard")
  with concrete descriptions.
- Under 350 words.
After the ad, list every change that affects requirements, so the client
can approve them.

The final instruction in Prompt 3 matters in agency work. The ad represents your client, and any change to requirements should go back to them rather than slip through in a rewrite.

Which AI prompts help with screening questions?

Structured questions, asked the same way of every candidate and tied to the job, are fairer and easier to compare than improvised ones. AI is good at drafting them from a brief.

Prompt 4: Screening questions from the brief

Write 8 screening questions for this role.
[paste brief]

For each question give:
- The question, worded identically for every candidate
- Which requirement from the brief it tests
- What a strong, acceptable and weak answer looks like

Rules: every question must relate to a requirement in the brief. No
questions about age, family, health, religion, nationality, or anything
else not needed to judge ability to do the job.

The strong, acceptable and weak examples are what make these questions usable across a desk. Two recruiters asking the same question will judge the answer the same way only if they share a picture of what good looks like.

Prompt 5: Knockout question check

These are the knockout questions our application form uses:
[paste questions]

For each, tell me:
1. Which job requirement it enforces
2. Whether it could screen out capable people for reasons unrelated to
   the job (for example, requiring a degree where the brief does not)
3. A more precise alternative if needed

Prompt 6: Call notes to a scorecard

Below are my notes from a screening call. Map them to the scorecard.

SCORECARD CRITERIA:
[paste criteria from Prompt 4]

NOTES:
[paste notes, contact details removed]

For each criterion, quote the part of my notes that supports it, or
write "NO EVIDENCE IN NOTES". Do not score the candidate and do not
summarise anything not related to the criteria, including personal
details mentioned in passing.

Prompt 6 is built to stay out of the decision. It organises your notes against your criteria and shows the evidence. The recruiter still scores.

Which AI prompts help with candidate outreach?

Outreach is where AI saves the most typing, and where carelessness is most visible to candidates. The rules in these prompts exist to stop invented flattery.

Prompt 7: First outreach message

Write a short outreach message to a potential candidate.

ROLE: [title, company type, location, pay range if shareable]
WHY IT IS INTERESTING: [2-3 real selling points]
WHY THIS PERSON: [1-2 facts from their public profile that match the brief]
CHANNEL: [LinkedIn message / email]

Rules:
- Under 90 words.
- Use only the facts above. Do not invent details about the person
  or the company.
- One clear question at the end.
- No "I came across your impressive profile" style openers.

Prompt 8: Follow-up after no reply

Write one polite follow-up to the message below, sent 5 days later.
[paste original message]
Under 50 words. Add one new, true piece of information about the role
from this list: [facts]. Make it easy to say no.

Prompt 9: Rejection message

Write a respectful message telling a candidate they will not be put
forward for [role] at this stage.
Context: [interview stage reached, anything they asked for]
Rules: under 100 words, no reasons that relate to anything other than
the role's requirements, no promises about future roles unless I say so:
[yes/no].

For more outbound patterns, the cold outreach prompts collection covers sequences and subject lines in more depth.

How should recruiters review AI outreach before it is sent?

Read every message as the candidate will. A quick checklist catches almost every problem:

• Is every fact true? Models fill gaps with plausible details: a project the candidate never worked on, a benefit the client never offered. Delete anything you cannot point to.
• Is it the right person and role? Batch drafting makes it easy to send the right message to the wrong name.
• Does it sound like your agency? A shared context describing your tone keeps messages consistent across recruiters.
• Is the pay and location honest? If it is shareable, say it. Vague outreach wastes both sides' time.
• Would you be comfortable if the candidate saw the prompt? If the answer is no, rewrite the prompt.

Volume is the other trap. AI makes it cheap to send hundreds of messages, and candidates can tell. Fewer, specific messages to well-matched people usually get more replies than a large generic send, and they protect the agency's reputation in a small market.

Which AI prompts help with client communication?

Clients want short, comparable summaries. AI is useful here, as long as the summary contains only what is relevant to the job.

Prompt 10: Shortlist summary for a client

Summarise these candidates for the client, one block each.
[paste recruiter scorecards, not raw CVs]

For each: current role, relevant experience against each must-have,
availability, salary expectation if shared, and one risk to discuss.
Only include job-relevant information. Do not mention or infer age,
gender, ethnicity, family situation, health or any other protected
characteristic, even if it appears in the input.

Prompt 11: Weekly client update

Write a weekly update for [client] on the [role] search.
Pipeline numbers: [sourced / contacted / screened / submitted / interviewing]
What changed this week: [notes]
Blockers: [notes]
Under 150 words. Plain language. End with one decision or input we need
from the client, if any.

How do bias-check prompts work, and what can they miss?

A bias-check prompt asks a model to review your own wording for language that is not job-related. It is useful as a second reader. It is not an audit, and it is not a guarantee.

Prompt 12: Job ad bias check

Review this job ad for wording that could discourage or exclude
qualified applicants for reasons unrelated to the job.
[paste ad]

Check for: gender-coded words, age-coded phrases ("digital native",
"young team"), requirements that act as proxies for protected
characteristics, unnecessary physical or language requirements, and
vague "fit" language. For each finding, quote the phrase, explain the
concern in one sentence, and suggest neutral wording. If you find
nothing, say so; do not invent findings.

Prompt 13: Consistency check across candidates

Below are my written summaries of several candidates for the same role.
Candidate names are replaced with letters.
[paste summaries]

Compare how I described them. Flag places where similar evidence is
described with different strength of language, where one summary
mentions personal details the others do not, or where a criterion is
applied to some candidates and not others. Do not rank the candidates.

Prompt 14: Name-swap test

I will give you the same candidate summary twice with different names.
Rate each version against the criteria below independently, then
compare your two ratings and explain any difference.
CRITERIA: [paste]
VERSION A (name: [name 1]): [summary]
VERSION B (name: [name 2]): [identical summary]

These prompts help in three ways. They catch coded phrases people stop noticing. They make inconsistencies across candidates visible, which is hard to spot when you wrote the summaries over a week. And Prompt 14 turns a known weakness of language models into a test you can run on your own tool.

That weakness is real. A University of Washington team, reported by UW News in October 2024, varied names associated with white and Black men and women across more than 550 real resumes and had models from Mistral AI, Salesforce and Contextual AI rank them against job listings. The study found the models "favored white-associated names 85% of the time, female-associated names only 11% of the time, and never favored Black male-associated names over white male-associated names." Those were specific models in a specific setup, and your tool may behave differently. The point is that you should test rather than assume.

The limits follow directly. A model asked to check for bias is the same kind of system that can carry bias, so it can miss problems or flag harmless ones. A clean result from Prompt 12 means the model found nothing, not that nothing is there. Treat the output as a list of things to look at, and keep the decision with a person.

What do recruiting agencies need to know about AI and employment law?

That existing discrimination law already applies, and that some places now regulate the tools themselves. Here are three examples, quoted from the primary sources and checked on 17 September 2026. They are examples, not a complete list, and they are not legal advice.

United States, Title VII. The Civil Rights Act names employment agencies directly: "It shall be an unlawful employment practice for an employment agency to fail or refuse to refer for employment, or otherwise to discriminate against, any individual because of his race, color, religion, sex, or national origin". Nothing in that text exempts a decision because software helped make it. State and local laws can add further protected characteristics.

New York City, Local Law 144. The city's Department of Consumer and Worker Protection says the law "prohibits employers and employment agencies from using an automated employment decision tool unless the tool has been subject to a bias audit within one year of the use of the tool, information about the bias audit is publicly available, and certain notices have been provided to employees or job candidates." Whether a particular use of AI counts as such a tool depends on the law's definitions, which is a question for counsel.

European Union, AI Act. Annex III of Regulation (EU) 2024/1689 lists, as high-risk, "AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates". Article 6(3) sets out conditions under which a listed system is not treated as high-risk, including where it is intended "to perform a narrow procedural task". Which side of that line a tool falls on, and when the obligations apply to you, are questions to settle with advice.

The practical takeaway does not need a lawyer: the closer an AI tool gets to deciding who is considered, the more review, documentation and legal input it needs. The writing prompts above sit far from that line. Ranking and filtering sit on it.

How should an agency handle candidate data in AI tools?

Carefully, and with as little of it as possible. A few habits keep most agencies out of trouble:

• Check the tool first. Use only AI tools your agency has approved for candidate data, under terms your clients' contracts allow.
• Use less data. Many prompts above need no personal data at all. For the ones that do, remove contact details and anything not needed for the task.
• Replace names with letters when comparing candidates, as Prompt 13 does. It reduces both privacy exposure and the chance of name-based bias.
• Keep the record. Save the criteria, the evidence and the human decision, not just the AI output.

For personal details you do need to reuse, such as your own name, agency signature or a client's company description, reusable variables keep them consistent without pasting them each time. The guide to variables for personal details covers how to do that without oversharing. Before you trust any new prompt, it is also worth trying to break it, as described in red-teaming your own prompt.

Which prompts should a new recruiter learn first?

Start with the ones that need no candidate data and produce output that is easy to check. A sensible order for a new starter:

1. Prompt 1 and Prompt 3, because a clear brief and a clear ad make every later step easier.
2. Prompt 2, tested against real searches with a senior recruiter looking over the results.
3. Prompt 4, so screening calls are structured from day one.
4. Prompt 7 and Prompt 8, reviewed by a senior recruiter for the first few weeks.
5. Prompts 12 to 14, once they have written enough of their own material to review.

Prompts 6, 10 and 13 handle candidate information, so they come after the agency's data rules have been explained, not before.

How can a staffing agency keep its prompts consistent?

Agencies lose good prompts the same way they lose good client notes: they live in one recruiter's head or chat history. When that recruiter leaves, the wording the client approved leaves too. The guide on how agencies stop losing AI work when staff leave covers the process side.

On the tooling side, Prompt Architects lets an agency keep reviewed prompts, client contexts and variables in one library that works inside ChatGPT, Claude and Gemini. With Team Sharing, a senior recruiter can share an approved outreach or bias-check prompt with the desk. Teammates can use it but not edit it; only the author can change a shared prompt, which keeps approved wording stable. One caution from how the feature works: a shared item follows its creator, so keep core prompts under an account that will stay with the agency.

The Free plan includes 5 prompt enhancements per day, according to our FAQ page. Saving prompts and sharing them with a team need a paid plan. Prompt Architects writes and organises prompts. It is not an applicant tracking system, it does not screen candidates, and it does not make hiring decisions.