Does GPA Still Predict Anything for Entry-Level Roles?

Does GPA predict job performance at all?

Weakly, and less than the field assumed for thirty years. The current re-estimates of selection-method validity put structured interviews at .42, job knowledge tests at .40, work samples at .33 and general mental ability at .31, and they find that measures built for a specific job beat general measures of a candidate's attributes ¹. GPA belongs to the general family, and it arrives already compressed.

Compressed is the operative word. Across 135 institutions and 1.5 million students, A's account for 43% of all letter grades, 12 percentage points more than in 1988 and 28 more than in 1960, and the researchers who counted them call it likely that GPAs at many selective schools are now saturated enough at the high end to have little use as an evaluation tool for employers ². A number most of a graduating class shares cannot separate that class. The bar stayed at 3.5 while the distribution moved under it.

Grade inflation is the recent problem. The older one is what a transcript measures at all: performance on structured, closed-book, individually graded work. That is a fair description of an accounting midterm and a poor description of most first-year jobs, which are open-book, collaborative, and judged on whether the answer survives contact with a client. GPA has always predicted the first thing and been asked to stand in for the second.

So the honest reading is not that GPA is worthless. It is that GPA is a weak general signal that gets weaker the further the work drifts from what a transcript measures, and AI has moved a great deal of entry-level work in exactly that direction. It is the same drift behind the question of what is left to screen on when every new-grad resume is AI-built.

Where does GPA still hold?

Where the curriculum and the job's knowledge base are the same body of knowledge, tested closed-book, with a licensure exam standing between school and practice. Nursing, accounting, actuarial work, structural engineering, pharmacy: in those fields a transcript is a four-year proctored knowledge test, and job knowledge tests sit second on the current validity table ¹. The overlap is what carries the signal, not the number itself.

The practical test is one question. In the first year, is the work bounded by facts that are right or wrong, or by judgment calls that are defensible or not? A new nurse's year is heavily bounded: dosages, protocols, escalation criteria. A junior strategy analyst's year is not: the arithmetic is checkable, but which number matters is a judgment, and that is the part a transcript never covered.

Two cautions apply even there. The overlap has to be real. A finance GPA earned mostly in electives is not a job-knowledge measure for a credit role. And the licensure exam usually does the job the GPA is asked to do, later and better: once someone has passed the boards or cleared two CPA sections, the transcript adds very little on top.

Where has GPA stopped telling you anything?

In work whose first year has become reviewing generated output rather than producing a first draft. Employment of 22-to-25-year-olds in AI-exposed occupations now sits 19% below where it would be had it tracked their less-exposed peers, and the declines concentrate in occupations where AI substitutes for tasks rather than complementing them ³. The tasks thinning out are the ones a strong transcript used to predict.

Read that finding at its stated strength. The authors call these descriptive indicators rather than causal estimates, and note the pattern attenuates once education is controlled for ³. It describes a labor market, not any one candidate. But the direction matches what campus leads are seeing: the marketing associate who used to write the first draft now edits one, and the junior developer who used to write the boilerplate now reviews it.

That inverts what a high GPA was proxying for. Coursework rewards producing competent output against a deadline, alone. A generated draft is competent output, produced in seconds, and it arrives with the surface confidence of a finished piece. What separates a good first year now is whether the person notices the confident draft is wrong, which is why the sharper version of this problem is how to hire for verification when AI does the production.

What is replacing the GPA cutoff?

A sample of the work, read for named behaviors. The revised estimates favor job-specific measures over general ones by a wide margin, though the same authors are candid that work samples and job knowledge tests are often unusable for entry-level hiring, where the knowledge and skills develop after the hire ¹. So the early-career version puts judgment in front of a graduate on material anyone could plausibly meet.

Three pieces show up in the loops that have dropped the bar:

• A short task with the assistant open. A realistic brief, any tool allowed, and two things handed back: the deliverable and the record (the prompts, the directions rejected, the checks actually run). The record is what gets read, not the polish.
• A structured interview built on that task. Structured interviews carry the highest mean validity in the revised estimates and work with inexperienced candidates, where a true work sample often cannot ¹.
• One claim checked outside the model. Which statement in the deliverable was verified, and what verified it. It takes two minutes and it separates a stack faster than any credential does.

None of that reads a transcript, which is the point. The loops doing it well swap a stage rather than appending one, which is how AI skills get tested without adding an hour to the loop, and the instruments they buy are judged on what an early-career assessment should actually measure.

What do you bring instead of the number?

The record of how you work, not a better adjective. A transcript reports a result; a loop built on work samples reads process: the brief, the directions you rejected, the claim you checked outside the model, and what changed because you checked. That evidence is portable across employers, and unlike a GPA it is still being made after graduation.

Three things worth having ready before an interview:

• One task you can walk through end to end. A course project, a part-time brief, something you built on your own. What the first attempt got wrong is the interesting half.
• A confident wrong answer you caught. Where an assistant produced something plausible and the error was real, name the check that found it.
• A claim and its source. One statement you verified, what you opened to verify it, and what it changed in the work.

Where the number still earns its line on the page is the overlap case above: a licensed field whose curriculum is the job's knowledge base, or a strong result in coursework the role is built on. Everywhere else it is a line someone reads after the work, and the work is what is being read. A cutoff you fall under is a filter, not a judgment about you. It runs before any evidence about you exists.

Is a GPA cutoff a legal risk?

It is a selection procedure, and it carries the obligations of one. Under Title VII, a neutral screen that disproportionately excludes people on a protected basis must be shown job-related and consistent with business necessity, and the EEOC states plainly that this means evaluating an individual's skills as related to the particular job rather than measuring skills in general ⁴. A cutoff no job analysis supports is a poor position to defend.

Two things follow. A bar inherited and never validated is exposure carried for a signal the research calls weak, which is the worst of both trades. And the fix is not to stop writing it down: it is to state what the bar is for, check whether the evidence supports it in that field, and remove it where it does not.

A bar worth keeping is narrow. A named role, a stated reason tied to the knowledge that role needs, and a date to review it. Whatever gets bought to replace it faces the same standard: setting a defensible bar for good enough at AI runs on the evidence a validated GPA cutoff would have needed.

See a sample report >