AI for recruiters: 10 prompts, and what New York City's bias audit rule says about ranking

What is the best AI for recruiters: a $20 chat app or a recruiting suite?

LinkedIn Recruiter Lite and Recruit CRM print no price on their pricing pages. So of the six recruiting suites we checked on 3 October 2026, four can be priced, and only Manatal has a paid plan that costs less than ChatGPT Plus or Claude Pro on monthly billing.

Loxo Core ($149 a month, quoted billed yearly) and Workable Standard ($299 a month for a whole company of 1 to 20) run off this scale, so the table carries them with a billing column that keeps each price on its own terms: US dollars, each vendor's own page on 3 October 2026. The fourth column is the one the legal section returns to, because a score or a ranking is what New York City's rule names.

The suites and the chat apps are different purchases. The suites sell data and plumbing: contact credits, a candidate database, a pipeline, sequences that actually send. A chat app sells the words, and it ranks nobody unless you type the request, which keeps the screening decision with you. Whizi loses to every suite above on the data half: it has no candidate database, no contact credits, no ATS and it sends no InMail. What it adds is GPT, Claude and (on Pro) Gemini behind one login, and projects that hold up to 10 pinned files and 32,000 characters of instructions, so the intake notes for a requisition ride along on every chat about that role.

What are the best AI prompts for recruiters, from intake call to first message?

The best recruiter prompts hand the model your notes and ban it from adding requirements, flattery or guesses. These four cover the start of a search. Paste them as written and fill the brackets. Each one works the same in ChatGPT, Claude or Gemini. The HR prompts pack carries the rest of the hire, from screening a pile of applications to drafting the offer letter.

Prompt 1, intake notes to a scorecard:

Here are my notes from the intake call with the hiring manager for [role] at [company], [city or remote]. Turn them into: five must-haves, each written as something a resume or an interview answer can show; up to three nice-to-haves; the pay range exactly as stated; and three questions I still need to ask the hiring manager. Do not add any requirement, degree or years of experience that is not in my notes. Notes: [paste].

Prompt 2, the search strings:

Using these five must-haves [paste], write one Boolean search string for LinkedIn and one Google X-ray string starting with site:linkedin.com/in for [role] in [location]. Include common title synonyms in an OR group, and exclude recruiter, intern and student. Under each string, explain every OR group in one line so I can prune it.

Prompt 3, the first message:

Write a LinkedIn message under 90 words to a [title] at [current company]. Open with this specific detail from their profile: [paste one line]. Say what the role is, the pay range [low to high], whether it is remote, and end with one yes or no question. No flattery, no exclamation marks, no "I came across your profile".

Prompt 4, the second message, five days later:

Write a follow up under 40 words to the message below. Do not repeat the pitch. Add one new fact about the role: [paste]. Make it easy to say no. Message: [paste].

Prompt 3 carries the pay range for a reason. The job description prompt lists the 13 states and DC that require a range in the posting itself, and a candidate who sees the number in the first message replies on the number. If you also email hiring managers to win new clients, AI for sales reps quotes the FTC's CAN-SPAM guide, which makes no exception for business-to-business email.

Which prompts work for screening, scheduling, interviews and the client submittal?

The screening prompt below asks for evidence per must-have and forbids a score, a rank or a recommendation. Those are three of the four output forms New York City's rule lists (the fourth is a tag), and the next section quotes the rule. Replace the candidate's name with Candidate A before you paste: names are what the bias studies below swap to change the result.

Prompt 5, resume against the scorecard: For each must-have below, quote the line from this resume that shows it, or write not shown. Then list anything on the resume that I should ask about in a phone screen. Do not score, rank, rate, tag or recommend the candidate. Do not comment on name, age, graduation year, gaps, photos, school prestige or anything else not tied to a must-have. Must-haves: [paste]. Resume: [paste, name removed].

Prompt 6, the phone screen guide and the invitation: Write six questions for a 20 minute phone screen for [role], one per must-have plus one on pay and notice period. Ask every candidate the same questions. For each, write two lines on what a strong answer contains. No questions about age, family, health, religion, national origin or anything a resume cannot change. Then write a three line invitation to the screen offering these two times with the time zone: [slots].

Prompt 7, my notes to a summary: These are my own typed notes from a screen with Candidate A. Sort them under each must-have, quote the candidate's words where I wrote them down, and list open questions. Do not add anything I did not write, and do not give a hire or no hire view. Notes: [paste].

Prompt 8, the client submittal: From the summary below, write a 150 word submittal to the hiring manager: two lines on fit against the must-haves, current and expected pay, notice period, and the one concern I flagged. Use only facts in the summary. Summary: [paste].

Prompt 9, the decline: Write a decline email under 80 words to a candidate who interviewed for [role]. Thank them by first name, say we chose someone whose experience matched [one must-have] more closely, offer to keep them in mind for [type of role]. No reason that touches anything personal, no "unfortunately" twice.

Prompt 10, the close: From my notes, list what this candidate said matters to them in the next job, then draft three talking points for the offer call that answer those points with facts from the offer: [paste the offer terms]. No pressure lines, no invented deadlines.

Scheduling is the one step here a chat app can't finish: Prompt 6 writes the invitation, but the model can't see a calendar or book a slot. That's a suite feature. Workable Standard lists "Self-scheduled interviews" with two way calendar sync for Gmail and Microsoft 365, and Manatal Professional lists calendar integration, both as of 3 October 2026.

Prompts 7 and 8 start from your typed notes, not from a recording, which keeps the summary to what you heard. Recording a call brings its own consent rules (see the FAQ). If the same candidate is preparing with a model on the other side, their prompts are the mirror image of Prompt 6.

Is it legal to use AI to screen resumes?

None of the five US rules in the table below bans using AI to screen resumes; each regulates the output, the notice or the outcome. This section quotes what the rule texts and regulators say, dated, and goes no further: it is not legal advice, and whether a particular workflow is covered is a question for employment counsel. New York City's rule, in force since 5 July 2023, is the most specific. Its definition of simplified output lists a score, a tag or categorization, a recommendation, and a "ranking (e.g., arranging a list of candidates based on how well their cover letters match the job description)". The same definition excludes tools that "translate or transcribe existing text".

The adopted rule text applies to employers and employment agencies. It counts a tool as substantially assisting a decision when the employer relies "solely on a simplified output", weights it "more than any other criterion in the set", or uses it "to overrule conclusions derived from other factors including human decision-making". It also limits the definition to techniques for which "a computer at least in part identifies the inputs" and their weights. The rule gives no example involving a general chat app.

Four more rules apply to the same decision, and the federal one is the oldest. The table lists each as its own text states it, United States only, as of October 2026.

The reason the rules focus on ranking is measured, not theoretical. A University of Washington study published 31 October 2024 ran more than 3 million comparisons of over 550 real resumes against more than 500 job listings through three open models from Mistral AI, Salesforce and Contextual AI. Resumes with white-associated names were preferred 85% of the time and female-associated names 11% of the time, and the systems never favored Black male-associated names over white male-associated names. Those were open models, not ChatGPT or Claude, so read the numbers as the reason Prompt 5 removes the name, not as a measurement of your chat app.

Enforcement so far, in the regulators' own numbers: the New York State Comptroller's audit of 2 December 2025 found the city received 2 AEDT complaints from July 2023 to June 2025, and that it surveyed 32 companies and "identified just a single issue of non-compliance". Reviewing the same companies, the state auditors "identified at least 17 instances of potential non-compliance". A 2024 FAccT study of 391 New York employers found 18 had posted a bias audit and 13 a candidate notice. And in Mobley v. Workday, a federal judge let an age discrimination claim against a screening vendor proceed as a nationwide collective on 16 May 2025. The rules don't stop at the hire, either. AI for HR shows which of six US AI employment rules also reach pay, discipline and discharge.

What should a recruiter never paste into ChatGPT?

Never paste a candidate's Social Security number or other government ID, date of birth, photo, background check, credit report, medical or disability information, or accommodation request, and never paste a client's confidential pay bands or an unannounced role. A resume with the name, phone number and email removed is the useful part, and Prompt 5 needs nothing else.

The reason is where a pasted line goes. OpenAI's data controls page says consumer conversations can train its models unless you switch off "Improve the model for everyone", and that ChatGPT Business, Enterprise and Edu are not used for training by default. Anthropic's pricing page lists training as an opt-out on Claude Free, Pro and Max. What never to paste into an AI chat has the full list for every vendor, and an agency handling a client's data should check its contract before the first paste.

Will AI replace recruiters?

No, on the US government's own projection. The Bureau of Labor Statistics counts recruiters inside human resources specialists, 939,700 jobs in 2025 with a median pay of $75,940 in May 2025, and projects 6 percent growth from 2025 to 2035. Its description of the job is the part a chat app doesn't touch: recruiters "find, screen, and interview applicants for job openings."

So pay $20 for a chat app and use Prompts 1 to 10 for the writing. Pay for a suite when the bottleneck is contact data, scheduling or pipeline volume. If its scoring will touch candidates in New York City, the date on the vendor's most recent bias audit is the document the city's rule turns on.