Can a Staffing Agency Just Use ChatGPT to Screen Candidates?

Yes. A staffing agency can use ChatGPT to screen candidates, and for a handful of CVs on one role it works well enough. It stops being enough when you need the same rubric applied the same way every time, claims checked against the public record, and a trail a client can read.

That is the short answer for an agency owner deciding whether a screening tool is worth buying. The rest of this page says where the line sits, what the evidence says about ranking CVs with a general chat model, and what the alternative costs in rupees as of October 2026. AgentR is one such alternative, so read its section with that in mind.

When is ChatGPT enough for screening?

ChatGPT is enough when one consultant is reading a few CVs for one role against criteria already written down. In that setting it is a fast second reader: it summarises a CV, lines it up against the job description and flags gaps for the consultant to look at.

Check the plan before you paste anything. OpenAI says that by default its business offerings (ChatGPT Business, Enterprise, Edu and the API Platform) are not used to train models, while on personal plans data sharing is on by default and has to be switched off under Data Controls (OpenAI help centre, as of October 2026). A consultant pasting client candidates into a personal account has made a data decision on the agency's behalf.

Where does a chat window break for an agency?

It breaks in four places: consistency, claim checking, evidence for the client, and candidate data. None of them shows up on the first CV. All of them show up when a client asks how you shortlisted.

Can you trust it to score the same way twice?

Not without testing it yourself. Bloomberg ranked near-identical resumes with GPT-3.5 and GPT-4 across hundreds of names and four job types, and found that at least one demographic group was adversely affected for every job listing except retail workers ranked by GPT-4 (Bloomberg, 2024). For the senior software engineer role, names distinct to Black women were ranked top only 11% of the time.

A University of Washington study of more than three million resume-to-job comparisons reached a similar result with three other language models, from Mistral AI, Salesforce and Contextual AI: white-associated names were favoured 85% of the time and Black-associated names 9% (University of Washington, October 2024). Both studies used US names and older models, and neither tested Indian names or current ChatGPT. That is the point: an agency using a chat window has no way to know how it behaves on its own candidates. Our piece on what a bias audit actually checks explains what testing would involve.

Does it check claims, or just read them?

A chat model reads the CV it is handed. You can ask it to search the web, but then you own the checking: which sources, what counts as a mismatch, and whether every candidate got the same treatment. Overlapping tenures, a job title that no longer fits the company's size and a degree that does not fit the dates are checkable facts, and doing them by prompt, one chat at a time, is the work a screening tool exists to take over. What to look at before a client's own checks is covered in what to check before background verification in India.

Can you hand the chat to a client?

A chat log is only a transcript, with no criteria, scores or checks laid out per candidate. Clients judge the agency when a placement turns out to be inflated, so they want to see why a candidate was submitted: the criteria, the score against each, and what was checked. Rebuilding that from chat history for every submission is consultant time the agency does not bill for.

Where do the CVs go?

Candidate CVs are personal data, and India's DPDP Act applies whichever tool reads them. The duties that matter most to an agency, consent and notice, are laid out in our guide to AI screening under the DPDP Act. The practical question for a chat tool is simple: can you say, for each CV, which account it went into and under what terms.

What does AgentR do differently, and what does it leave alone?

AgentR reads every application in full against criteria the recruiter sets, checks what the candidate claims for consistency against public and organisational signals, and runs a structured AI interview scored against a rubric. The agency gets a ranked shortlist with the evidence attached, in a form that can go to a client. A person makes every decision, and a recruiter can invite any candidate to interview manually, whatever the cutoff says.

The limits matter as much. AgentR is in private beta and has no customer record to point to. Its verification checks claims for consistency. It does not contact employers or institutions and does not replace a background verification vendor, so a client who requires BGV still needs one (see verification and what it covers). It is not an applicant tracking system: there is no pipeline view and no public API as of October 2026. If your whole process lives in an ATS and you only want a quick read of a few CVs, ChatGPT or your ATS may be the better fit.

What does it cost for one role?

AgentR's plan is ₹11,999 a month, which converts to ₹12,000 of usage, then ₹25 per applicant reviewed and ₹10 per interview minute, billed only for time used. Verification and proctoring are included. Prices exclude GST (AgentR pricing, as of October 2026).

As an illustration, not a customer: an agency reviews 300 applicants for one role and interviews 20 of them for 30 minutes each. That is 300 × ₹25 = ₹7,500 plus 600 minutes × ₹10 = ₹6,000, so ₹13,500 of usage. The base plan covers ₹12,000 and the remaining ₹1,500 comes from credits, about ₹13,500 in all before GST. Change the assumptions and the total moves in proportion. ChatGPT's price depends on the plan and is on OpenAI's own pricing page; this page does not quote it, because the real cost of the chat route is the consultant time spent re-checking and re-formatting.

Which should you pick?

You read a few CVs a week for one or two roles. Use ChatGPT on a business plan, with the criteria written first and a consultant making every call.

A client asks how you checked a candidate. You need a record of criteria, scores and checks per candidate, which a dedicated tool produces and a chat history does not.

You screen hundreds of applicants per role. Volume is where consistency and cost per applicant start to matter, and where a tool pays back consultant hours.

You are not sure. Take a role you have already filled and run its applicants through ChatGPT three times with the same prompt. If the three rankings disagree, you have your answer about repeatability. Then run the same pool through the tool you are considering and compare both against who actually got placed.

Run one closed role through AgentR and compare it with your current process. The first role is free.

See what the trial covers on the pricing page.