Can You Use ChatGPT to Screen Resumes? What Works and What Does Not

Yes, you can use ChatGPT to screen resumes, but only for low-risk help such as summarising a few CVs or drafting criteria. It is weak at consistent scoring, keeping candidate data private and leaving an audit trail. For real AI resume screening on a whole applicant pool, a dedicated tool is usually safer.

The short version of what a chatbot can and cannot do

A general chatbot is a flexible reader. Paste in a CV and a job description and it will write a sensible summary, point out missing skills and suggest interview questions. That is genuinely useful, and for a small business that sees five applicants a month it may be all the help you need.

The trouble starts when the job changes from "help me read this CV" to "screen my applicants". Screening means treating many people the same way, explaining why one ranked above another, protecting their personal data and being able to show your working later. A chat window was not built for any of those four jobs. This guide goes through what works, what does not, a prompt that reduces the risk, and the point where you should move to resume screening software. For the wider picture of how dedicated tools work, see /ai-resume-screening.

What works well in a chat window

Some tasks are a good fit for a chatbot because a mistake is cheap and a person checks the result straight away.

• Summarising one CV into five lines so you can decide whether to read it in full.
• Turning a messy job description into a list of must-have and nice-to-have skills that you then edit yourself.
• Drafting screening questions or interview questions for a role.
• Rewording a rejection or a request for more information so it sounds kind and clear.
• Explaining an unfamiliar job title, tool or qualification on a CV.

In each case you are using the chatbot as a drafting assistant and you stay in charge. Nothing is decided by the output, and nothing needs to be repeated identically next week. The moment a task needs the same answer every time, or needs to be defended, the limits below begin to matter.

Privacy: the first thing to settle

A CV is personal data. It usually holds a name, address, phone number, work history and sometimes a date of birth, a photo or a nationality. Before pasting any of that into a chatbot, you need to know where the text goes, who can read it, whether it is used to improve the service, and how long it is kept.

Those answers depend on the product, the plan and the settings, and they change over time, so check the provider's current terms and data controls rather than relying on this article or on memory. Consumer accounts and business accounts can be treated differently. Your own situation matters too: rules about handling candidate data differ by country and sometimes by region, and your candidates may have been told you would use their CV only for hiring. If you cannot answer "where does this text go" in one sentence, do not paste a stranger's CV into it.

A practical safeguard is to remove the name, email, phone number and address before pasting, and to paste only the work history and skills. That reduces the risk but does not remove it, because a CV can still identify a person from employers and dates.

Consistency: the same CV can get different answers

Ask a chatbot to score a CV out of ten, then ask again, and you may get a different number. Change one word in your prompt and the ranking of two candidates can flip. This is not a flaw that will be fixed by a better prompt, it is how these systems work: the output is generated fresh each time rather than calculated from fixed rules.

For screening, that matters because fairness depends on treating candidates alike. If candidate 3 was judged on Monday with one phrasing and candidate 40 on Thursday with another, you did not apply the same standard, and you may not even know it.

You can test this yourself in ten minutes. Take one CV, paste it with your job description and ask for a score and a one-line reason, five times in five fresh chats. Write down the five scores. A tool you can rely on for ranking should give you the same answer, or very close, each time. If the spread is wide, you have learned something important about using it to rank people.

Bias and what you cannot see

Every screening method can be biased, including a person reading a pile of CVs at the end of a long day. Language models add a particular difficulty: you cannot easily see what influenced the answer. A name, a school, a gap in dates or a writing style may all tilt the result without being mentioned in the reasoning it gives you.

A chatbot will usually sound confident and give a tidy reason. That reason is a plausible explanation written after the fact, not a record of how the answer was reached. So "it explained itself" is not the same as "it can be audited".

Ways to reduce the risk when you do use a chatbot: hide names and photos before pasting, ask it to judge only against a written list of criteria, never ask it to guess age, nationality, family status or health, and have a person read the CV for anyone it ranks low. Rules on automated decisions about people differ by country, and some places have specific laws, so check local law before using any AI tool to filter candidates. See /ai-resume-screening for how a dedicated approach handles this.

Audit trail: can you show your working later?

Imagine a candidate asks why they were not shortlisted, or a hiring manager asks why a strong-looking CV never reached them. With a chat window, your record is a long scrollback of conversations, possibly deleted, possibly spread over several accounts, with no link to the candidate record and no way to see who looked at what.

A usable screening record answers four questions: what criteria were applied, what each candidate scored against each one, who made the decision and when. A chat history gives you at most a loose version of the first and none of the rest.

Even if you never face a dispute, a record helps you improve. If you can see that eight of your last ten hires scored well on one skill and badly on another, you can fix the job description. A chat window cannot tell you that.

A safe prompt and its limits

If you decide to use a chatbot for a small, low-stakes batch, use a prompt that forces it to work from your criteria and to show evidence. Here is a word-for-word template you can copy.

"You are helping me review one CV for a hiring decision that a person will make. Do not guess age, gender, nationality, health, family status or any protected characteristic, and ignore the candidate's name, photo, address and school names. Judge the CV only against the criteria below. For each must-have, answer Found, Partly found or Not found, and quote the exact words from the CV that support your answer. If there is no quote, answer Not found. Then list anything unclear that I should ask the candidate. Do not give an overall score and do not recommend rejecting anyone. Must-haves: [paste 4 to 6 must-haves]. Nice-to-haves: [paste 2 to 3]. CV: [paste CV with name and contact details removed]."

This prompt works because it asks for quotes, bans an overall score, and leaves the decision with you. It still has limits. It will not give you the same answer every time, it cannot keep a record for you, and it still depends on where the text is sent. It also treats a skill the CV never mentions as missing, which can be wrong, so read the CV before you reject anyone.

Chatbot or dedicated tool: a decision table

Use this table to decide. It describes how the two approaches generally differ, not any one product, so check each tool's current features.

The pattern is simple. The more candidates, the more the decisions matter and the more you need to explain them, the more a purpose-built tool earns its cost.

A scoring checklist to decide for yourself

Answer each line with 0 for no, 1 for partly and 2 for yes about the way you plan to screen. Add up the points.

A total of 14 to 16 means your process is sound whatever tool you use. From 8 to 13, fix the gaps before you scale up. Below 8, do not use AI to filter candidates yet. Notice that a chatbot alone will struggle to score well on lines 3, 4 and 6, which is exactly where a dedicated tool is meant to help.

Worked example with example numbers

These numbers are examples only. A small agency has one customer-support vacancy and receives 60 CVs. The recruiter tries a chatbot first.

She pastes 10 CVs with the safe prompt. Nine take about two minutes each to paste, clean and read, so around 20 minutes for ten, and the answers are helpful. She then runs one borderline CV three more times and gets "Found", "Partly found" and "Found" for the same must-have. For the remaining 50 CVs she estimates 100 more minutes of pasting, with no combined ranking and nothing saved against the candidates.

She switches to a dedicated tool for the rest. She enters the must-haves once, uploads the 50 CVs together, and gets a ranked list with quoted evidence in minutes, then spends her time reading the top 15 and checking a few low scorers. The time saved is the reading of 35 CVs she can safely deprioritise, not the decision itself. She also has the notes and statuses stored against each candidate. The lesson is not that the chatbot was bad, it is that it suited 10 CVs and did not suit 60.

Common mistakes, and when not to use each approach

Mistake one: pasting full CVs with contact details into a free consumer account without reading the data terms. Mistake two: asking for a score out of ten and treating the number as a fact. Mistake three: letting the output reject people with nobody reading the CV. Mistake four: changing the prompt between candidates, so they are not judged on the same basis. Mistake five: forgetting that a CV which does not mention a skill reads as lacking it, even if the person has it. Mistake six: keeping no record, then being unable to explain a decision. Mistake seven: assuming the tool is neutral because it sounds polite.

Skip the dedicated tool if you hire only a few people a year and read every CV yourself, or if your problem is too few applicants rather than too many. A spreadsheet or the options at /free-applicant-tracking-system may be enough.

Skip the chatbot for screening if candidate data is sensitive, if you will rank more than a handful of people, or if you may need to explain a decision to a manager or a candidate. And skip both if nobody on your team has time to read the CVs the tool puts at the top, because a ranking nobody checks is just a faster way to make unreviewed decisions.

Where Rezumz fits

Rezumz is an AI hiring workspace for small and mid-size hiring teams. It scores each CV against the must-have and nice-to-have skills you set for a job, and shows the matched and missing requirements with the CV lines it relied on quoted, so a person can check the reasoning. The score is calculated from the skills found in the CV, with 80% from the share of must-haves and 20% from nice-to-haves, and the AI adds extra signals alongside rather than replacing it. Scoring does not need AI, and the score never rejects, emails or moves anyone on its own. It also covers the apply page with up to five screening questions, notes on CV lines, shortlisting, interview scorecards, offers and an audit log of who did what for the account owner.

What it does not do: it does not decide for you, it does not post to job boards in one click, and it is not an enterprise suite with deep HRIS or payroll links, so large enterprises needing those should look elsewhere. It suits small and mid-size hiring teams, and the Team plan is meant for four or more recruiters.

Prices: a free plan with 2 free candidate analyses; $29 a month, $199 a year, or $499 lifetime (limited); Team is $79 a month or $699 a year for up to 10 teammates. See /pricing and try the live demo at /demo. Compare options at /compare, and see /resume-parser and /recruitment-software for the related tools.