Why not just ask ChatGPT to screen your resumes?

For a handful of resumes, you can. A chat window will read them, summarise them and tell you which look strongest, and for one role with a few applicants that is often enough. It starts to fail when the pile grows, because nothing in a chat holds your criteria fixed from one resume to the next, and nothing keeps the reasoning where you can find it later. This article sets out where that line sits, what a spreadsheet and a job portal’s built-in AI do instead, what HireSalt does differently, and who should not use it.

We build HireSalt, so weigh what follows accordingly. We have tried to give the other options their due.

On this page

• What a chat window does well
• Where it falls short for screening
• The spreadsheet
• Portal built-in AI
• What HireSalt does differently
• Honest limits
• Where each option stands
• Who should not use HireSalt

What a chat window does well

A general-purpose assistant is good at reading a resume and saying something sensible about it. You can ask it to summarise a career, flag gaps, draft screening questions or explain an unfamiliar job title. You can paste the job description and ask how a candidate fits it. There is no setup, no new tool to learn and no cost beyond the subscription you may already hold.

It is also flexible in a way fixed software is not. If the right question only occurs to you halfway through, you can simply ask it. For a one-off role or a small pile, that flexibility is worth more than structure.

Where it falls short for screening

Screening is a different job from reading. It means applying the same standard to every resume in a pile, then being able to show someone how you did it. A chat window has no mechanism for that. These are mechanics, not measured results; we have not tested how any particular chat product behaves across a large pile.

The rubric drifts. Your criteria exist only in the prompt you wrote and whatever memory the tool keeps. If you re-type the instructions for the thirtieth resume slightly differently from the first, you have changed the standard without noticing. Nothing in a chat holds criteria fixed for you. Consistency depends entirely on you supplying the same instructions every time.

Nothing is fixed per job. There is no object called “the criteria for this role” that every candidate is scored against and that you can open later. The criteria are whatever was in the conversation when you asked. A colleague picking up the search starts from scratch, or from your pasted prompt and a hope that it matches.

Notes only appear if you ask. A bare ranking or a score is easy to get. A short written reason for each part of the score is something you have to request, in the same words, every time. When you do not, you are left with a number and no way to check it.

Re-ranking means re-reading. Suppose you decide, after twenty candidates, that depth in a particular skill matters more than years of experience. In a chat, the way to apply that is to run the resumes through again with new instructions. The earlier answers do not recalculate; you start another round.

None of this makes the answers wrong. It makes them hard to compare and hard to explain to a hiring manager who asks why candidate nine ranked below candidate four.

The spreadsheet

Many small teams settle on a sheet, and it deserves more credit than it usually gets. It holds a fixed set of columns, so every candidate is scored on the same fields. The weighting formula is visible and you can change it. It costs nothing, everyone knows how to use it, and it keeps a record of who entered what.

What it cannot do is read. A person has to open each resume and type the scores in, so consistency depends on who is reading, how tired they are and what order the resumes arrived in. It will not pull fields out of a PDF or a Word file on its own. A reason attached to a score exists only if someone writes it into a cell. And it cannot judge implicit evidence, such as the scope a job title implies or how a career has progressed, unless the reader does that work. When the formula changes, the sheet re-ranks the numbers, but nobody re-reads the resumes.

For a short list of candidates scored by one careful person, a spreadsheet may well be the right tool. Its limit is the reading time of whoever fills it in.

Portal built-in AI

Job portals have started to put AI ranking inside their own products. We can describe only what we have read on their public pages, so the following is per their public documentation, as of October 2026.

LinkedIn’s job-post assistant sorts applicants into Top fit, Maybe and Not a fit. Top fit means a candidate meets all must-have qualifications and a set share of the preferred ones. Not a fit means at least one must-have is missing. Maybe means all must-haves are met but the preferred threshold is not. The must-haves and preferred qualifications are generated from the job description, screening questions and listed skills. Recruiters can edit them and adjust the preferred-percentage threshold, and a question mark appears when the AI cannot tell whether a qualification is met. The page says recruiters remain responsible for final decisions.

LinkedIn’s Hiring Assistant page says it ranks using LinkedIn Recruiter’s standard ranking methods, does not make autonomous hiring decisions, and does not automate stage changes or rejections unless a customer turns those settings on.

The pages we read do not publish a weighting formula or a per-criterion explanation of how a ranking was reached. That is an observation about those two pages, not a claim that none exists elsewhere. We leave other portals out because we could not confirm their documentation directly.

What HireSalt does differently

Each point below corresponds to something the product does today.

Criteria are set per job and locked once scoring starts. You define the criteria and weights for a role before candidates are scored. Once candidates have been scored, the criteria and weights are locked. Changing them means clearing the scores and scoring again, or duplicating the job to try different weights. The locking is deliberate: it stops the standard shifting halfway through a pile, which is the drift a chat window cannot prevent.

Seven categories, 33 sub-criteria. Resumes are scored across seven categories with 33 sub-criteria underneath. Five of the categories carry weight by default. Practicalities and engagement profile are scored and shown but default to a weight of zero, so they do not move the number unless you change that. Category weights are set before scoring.

A note on each sub-criterion. The scoring instruction requires a one- to two-sentence note for each sub-criterion, citing the evidence in the resume and marking where something was inferred rather than stated. We say “requires” because it is an instruction to the model, not a guarantee the software enforces on every output. For the engagement profile category, where a resume gives no explicit evidence, the model is told to score conservatively and state that evidence was limited.

A preview of reweighting without re-reading. On candidates who are already scored, you can preview a different weighting and see the ranking reorder immediately, computed from the stored sub-scores rather than by reading resumes again. The sliders move within 10 points of the baseline. It is a simulator: making a new weighting permanent means scoring again. We cover this in what happens if you weight skills higher.

Files. Resumes can be uploaded as PDF or DOCX, individually or in bulk, up to 10 MB per upload request with all files combined. Plain-text resume files are not accepted. Scoring uses credits.

The full scoring process is described in how HireSalt scores a resume.

Honest limits

A score from HireSalt is decision support. A person decides.

Scores can differ between runs of the same resume. In our latest repeatability test (October 2026), we scored 32 resume and job pairs three times each. The grade held for 26 of the 32, and the score stayed within 5 points for 21. That is below the 80% bar we set before running it, so we do not claim consistent scores. The grade flips were mostly at the C/D edge. It is one model, one set of data and one night. Our benchmark also showed that a resume labelled a good fit scored above an off-role resume for the same job in 63 of 79 pairs (79.7%), below the 9-in-10 bar we set, and that the dataset’s labels are not hiring outcomes. We make no claim of accuracy against who turned out to be a good hire. The longer account, including what we got wrong early on, is in how consistent is AI resume scoring.

We do not claim the scoring is free of bias, and we do not detect fake resumes.

Where each option stands

Who should not use HireSalt

If you screen a few resumes a month, a chat window or a sheet will serve you and cost less in time. If you hire from a single portal and its built-in ranking already does what you need, adding another tool buys you little. If you need resumes in a format other than PDF or DOCX, HireSalt will not take them as they are. If you want a tool that makes the hiring decision, or one that predicts who will succeed, this is not that, and we would be wary of any tool that says it is.

HireSalt is worth a look when you have a real pile, several people need to see the same standard applied, and you want to be able to open any score and read why it came out that way.