How Do You Screen Entry-Level Candidates When Every Resume Looks the Same?
Every claim on an entry-level resume needs someone else to confirm it. One signal does not. Here is the cheapest check that is less wrong than sorting by school.
By Howard Davner
Quick Answer: Ask for one finished thing scoped to the role. Entry-level resumes converge because the experience genuinely is similar and the writing is AI-assisted on both sides, so reading harder recovers nothing. An artifact you open in one click, and ask one question about, is the cheapest signal that is less wrong than sorting by school.
Listen to this page (about 1 minute).
You have four hundred applications for two openings. None of the candidates has held the job before, because it is an entry-level job. The resumes are the same length, use the same verbs, and make the same unverifiable claims.
So you sort by school, or by GPA, or by whether you recognise the internship. Not because you think those predict performance, but because you have twenty minutes and they are the only fields that vary.
That is the actual problem worth solving. Not "how do I find the best candidate," but "what cheap signal is less wrong than the ones I am defaulting to."
Why the resumes converge
Entry-level candidates have genuinely similar inputs, so the documents converge for structural reasons rather than lazy ones.
The same intro courses, the same three or four clubs, the same campus jobs, the same one summer of retail. When the underlying material is comparable, differences in the writing are differences in polish, not substance.
And polish has collapsed as a differentiator. Every candidate now has access to a tool that turns a mediocre bullet into a competent one. The floor has risen, the ceiling has not moved, and the distribution has compressed. Reading harder does not recover information that was never in the document.
The defaults are weaker than they feel
School name, GPA and prior internship brand are all proxies, and each has a failure mode that maps onto background rather than ability.
- School name carries real information about admissions selectivity four years ago. For a role starting next summer it is a lagging indicator about the candidate's high school performance, filtered through their family's ability to pay. It also systematically excludes transfer students, community college students and anyone whose constraint was geography rather than ability.
- GPA measures performance under conditions the job does not replicate: explicit instructions, fixed scope, a known evaluator, and a deadline someone else set. It correlates with conscientiousness, which genuinely matters. But it is noisy across institutions and confounded by how many hours a student had to work while enrolled.
- Prior internship brand is the most circular of the three. You are using the outcome of someone else's entry-level screen as your entry-level screen, which compounds whatever bias was in theirs and rewards whoever got there first.
None of these is useless. All of them are answering a question you did not ask.
The part nobody prices: what each signal costs you to confirm
Every claim on an entry-level application needs a third party to confirm it, except one. That asymmetry, not accuracy, is why the cheap signal is the live one.
Walk down a typical application and count who you would have to contact to verify each line:
- A transcript or GPA needs the registrar, or a service the candidate pays for.
- A reference needs a phone call, scheduled, with someone who has an incentive to be positive.
- A prior internship needs a former employer who may not respond, and whose confirmation tells you the dates and not the work.
- A bullet about a class project needs trust. There is nothing to check. You are reading a description of an artifact you will never see.
- A GitHub link needs you to clone or read code before it shows you anything, which is why a repository underperforms as a screening object. That is the argument in does GitHub count as a portfolio.
- A live URL needs one click, from you, alone, in seconds, with nobody's cooperation and no scheduling.
That last line is the whole case. It is not that a deployed project is a more impressive credential than a transcript. It is that it is the only item in the stack whose verification cost to you is approximately zero, and the only one where the thing being verified is the candidate's own work rather than an institution's record of them.
Page one of this query is mostly lists of traits to look for. Attitude, coachability, curiosity, culture fit. Every one of those is real and none of them is checkable in the twenty minutes you actually have, which is why funnels keep falling back on school. The question is not which trait matters most. It is which signal you can confirm yourself before lunch.
What to ask for
Ask for one thing the candidate made, and make the request specific enough that a generic submission does not satisfy it.
- Scoped to the role. "Send a portfolio" gets you a link farm of unrelated work. "Send one thing you made that relates to what this role does, and two sentences on why you built it that way" gets you a decision you can evaluate.
- Openable. If it takes more than one click it will not get looked at, including by you. A live page beats a repo. A repo beats an attachment. An attachment beats a promise.
- Discussable. This is the part that solves the AI problem. Anything written can be generated. Almost nothing survives ten minutes of specific follow-up questions from someone who understands the work. "Why did you choose that approach over the obvious one?" separates the candidate who made the thing from the candidate who submitted it, faster and more reliably than any detection tool.
One practical warning on the openable part: free hosting plans idle. Render's documentation states that a free web service spins down after fifteen minutes without inbound traffic and takes about a minute to spin back up, as of October 2026. You will almost always be the first visitor in hours, so you will get the slow path every time. A link that loads slowly is asleep, not broken, and it is worth knowing the difference before you discard a candidate over it. What a deployed project needs for a reviewer to trust it in thirty seconds is the candidate-side version of that same check.
The fairness objection, taken seriously
The objection is correct for expensive artifacts and much weaker for cheap ones.
The strongest argument against portfolio screening is that it advantages candidates with time, money and equipment, which would make it another proxy for privilege wearing a meritocratic costume. If producing the thing takes an afternoon and a browser, though, it is available to a student working thirty hours a week in a way that an unpaid summer in another city is not.
The honest comparison is not portfolio screening versus a perfectly fair process. It is portfolio screening versus what you are doing now, which is very likely sorting by institutional prestige and prior employer brand, both of which are far more tightly coupled to family background than "did you finish something."
There is a second move that costs you a sentence and widens the pool rather than narrowing it: say in the posting what you will accept, and point candidates at a way to produce one. The gap most students stall in is not building, it is getting from a posting to something worth building. That is the gap Provieo's project builder closes: it reads a specific job description, generates a project matched to that role, coaches the candidate through building it, and deploys it to a live URL. Free and unlimited to use, first build and deploy free, then one single payment for unlimited deployed projects and no subscription. A posting that tells candidates how to meet the bar is screening on finishing rather than on who already knew the rules.
What a reviewer should actually check in thirty seconds
Three questions, in this order, and each one has to pass before the next is worth asking.
- Does it open? Cold, in a private window, on a phone. No login wall, no registration form, no request for an API key.
- Does it show a finished result on its own? A blank input box and a Submit button is homework, not a demo. If you have to invent a plausible input and guess the format, you cannot evaluate it quickly and neither can anyone else.
- Does it read as one person's work? Something scoped to a real posting looks nothing like tutorial output. "Flags contract clauses that deviate from a standard template" survives this check. "Predicts house prices" does not.
Then ask one question about a choice inside it. What employers look for in a CS portfolio covers how reviewers assess these once they are clicking, and why a deployed project changes the reading covers why a link does work a description cannot.
What this changes in practice
Nothing about your funnel volume, and everything about where your twenty minutes go.
You will still get four hundred applications. What changes is that a subset of them arrive with something checkable, and thirty seconds of checking tells you more than five minutes of reading claims.
You are not looking for brilliance in the artifact. You are looking for evidence that this person can take an ambiguous goal, cut it down to something achievable, and finish it, because that is what the first year of the job actually consists of, and it is the one thing the resume format cannot tell you.
Bottom line
Every other signal on an entry-level application costs you somebody else's cooperation to confirm. A link costs you a click.
That is the reason to ask for one, and it is a better reason than anything on a trait checklist. Sort the stack by what you can verify yourself, then spend the conversation on why they built it that way.
Frequently Asked Questions
How do you screen entry-level candidates when every resume looks the same?+
Ask for one finished thing scoped to the role, and screen on that instead of on school or GPA. Entry-level resumes converge because entry-level experience genuinely is similar and the writing is now AI-assisted on both sides, so reading harder cannot recover information that was never in the document. A finished artifact you can open in one click tells you in thirty seconds what a page of claims cannot, because finishing is rare and very hard to fake.
Why do entry-level resumes all look the same?+
Because entry-level candidates have the same inputs. Similar coursework, similar clubs, similar part-time jobs, and increasingly the same AI assistance shaping the same bullet structure. When the underlying experience is genuinely comparable, no amount of writing skill will differentiate it, so the documents converge. The floor on polish has risen and the ceiling has not moved.
Is GPA a good predictor for entry-level hires?+
It is a weak one. GPA measures performance in an environment with clear instructions, defined scope and a known grader, three conditions rarely present in a job. It correlates with conscientiousness, which matters, but it is noisy across institutions and heavily confounded by how much a student had to work while studying, meaning it partly measures how little a student needed the job.
What should I look for instead of school and GPA?+
Evidence the candidate did something resembling the work, unsupervised, and finished it. Finishing is the scarce trait: most people can start a project, far fewer scope it down, work through the boring middle and get it to a state where a stranger can use it. A completed artifact scoped to the role shows scoping, persistence and closure in one object.
How long does it take to screen a candidate's project?+
About thirty seconds for the artifact itself, and ten minutes in a conversation to confirm the candidate made it. The thirty seconds is spent on three questions in sequence: does the link open, does it show a finished result without you working out what to type, and does it read as the work of a specific person. The ten-minute version is the follow-up questions, and that is the part no tool replaces.
How do I word the request in a job posting?+
Be specific enough that a generic submission does not satisfy it. 'Send a portfolio' returns a link farm of unrelated work. 'Send one link to something you made that relates to what this role does, plus two sentences on why you built it that way' returns a decision you can evaluate. Ask for a live URL rather than an attachment, and say plainly that the two sentences matter more than polish.
How do I stop AI-written applications from flattening my funnel?+
Ask for something a language model cannot hand over. A written answer can be generated. A working artifact the candidate can be questioned about in a live conversation cannot be, because specific follow-up questions expose whether they understand what they submitted. The screen is not the artifact on its own, it is the artifact plus one question about a choice inside it.
Does screening for projects favour privileged candidates?+
It can, if the artifact requires expensive tools, unpaid internship time or a personal network. It is less biased than the alternative when the thing takes an afternoon and a browser. The honest comparison is not portfolio screening against a perfectly fair process, it is portfolio screening against what most entry-level funnels default to now, which is school prestige and prior employer brand, both far more tightly coupled to family background.
What does it cost a candidate to produce one?+
Hosting can be nothing. Vercel, Render and Hugging Face Spaces all have free plans, as of October 2026, though a free plan can idle and show a slow first load, so candidates should test a link cold before sending it. Check current terms before relying on any of them. Provieo is free and unlimited to use, a candidate's first project build and deploy is free, and after that it is one single payment for unlimited deployed projects, with no subscription.
Should I reject candidates who have no project at all?+
No, and treating a missing artifact as disqualifying would just invent a new proxy for who had spare time. Use it as a positive signal where it exists rather than a filter where it does not. If you want the whole cohort to have one, say in the posting what you will accept and give candidates a route to build something, which costs you a sentence and widens the pool rather than narrowing it.
Put a working link in your next application.
Paste the job you're applying to. Provieo designs a project against that exact listing and walks you through deploying it, so what a recruiter clicks is the work itself, not a description of it.
Free to use, and your first deployed project is free. Or tailor your resume to the same posting.