You Can Show AI Skill With No Job That Used It

Pick One Piece of Real Work

Choose one thing you did, not a course you sat through. A class project, a volunteer task, a personal build, a portfolio piece: any of these count, as long as you used a model on part of it and can describe what happened.

The piece you want is not the polished final output. It is the moment somewhere in the middle where the model gave you something wrong, thin, or off for what you needed, and you noticed and fixed it. That moment is the entire credential. A finished deliverable proves you can produce a finished deliverable. A caught correction proves you can tell a good output from a bad one, which is harder to assert without having done it and easy for a reviewer to question you about.

Size matters less than specificity. A three-line spreadsheet check counts as much as a semester-long project, provided you can say what was wrong and how you knew. Keep two of these ready rather than one, since an interviewer who likes the first will often ask for a second. Write it down the same day, before the details blur together.

Write It as an Accomplishment, Not a Skill

Turn the piece of work into one line built on three parts: the task, what the model produced that was wrong or incomplete, and what you changed. "Used AI to build a marketing plan for a class project, caught a target audience assumption that did not match the assigned company's actual customer base, and rebuilt the segment from the case data" tells a reviewer something a Skills line naming Excel, ChatGPT, and Canva never can.

Anyone can write "proficient with AI tools" on a page, and a reviewer has no way to test it. A named wrong output and a named fix invite a question instead: what was the assumption, how did you check it, what did you do next. Answering that in detail is what separates the line from the list it replaces.

Write two or three of these before you need them, not the night before an interview. Memory fades on the exact detail that makes a story credible, the number, the assumption you caught, and a story told from memory six months later tends to soften into something generic. A note taken close to the moment keeps the details an interviewer will ask about.

Why This Now Competes Equally With a Job

Unpaid work competes with a job here because the employer-side advice now says it should. Handshake, analysing the resumes graduating seniors uploaded to its own platform, found Class of 2026 resumes mention AI skills more than nine times as often as Class of 2022 resumes, with two-thirds of those mentions coming from majors outside computer science. ¹ That is US four-year students on one career platform rather than a national resume corpus.

The same analysis found the share of AI mentions tied to coursework fell from 65% in 2022 to 51%. ¹ Where the rest now comes from is not published, so nobody can tell you personal and volunteer projects specifically are the growth. What it does establish is that AI on a resume has stopped being a computer science signal, and that a reviewer is no longer surprised to see it with no employer attached.

What is left to screen on when every new-grad resume is AI-built? is the employer side of the same ground, and it is blunter than most advice written for you. It tells reviewers to screen only lines that have an address, an artifact someone can open, an affiliation with a named supervisor, an accomplishment a reference can confirm, and to take a coursework or internship product as readily as a public repository, because the alternative screens on who could afford an unpaid summer. Then it moves the decision off the page entirely, onto a first-round case of about 45 minutes run the same way for everyone.

What Employers Actually Ask For

What gets asked for is one real instance, described in detail: what you asked the model for, what came back, and what you did with it. That description carries across whatever kind of use you had, and it is worth writing now rather than assembling it cold in a room.

Most of that use will not look like what people picture. In a July 2025 Handshake survey of rising seniors, the most common uses of generative AI were brainstorming and self-teaching, reported by more than two-thirds, against less than a quarter who said they used the tools to directly generate content. ² Those are self-reported uses across study and work generally, not a breakdown of what anyone does inside a job application. If your own use looks like that, thinking with a model and learning from it rather than having it write things wholesale, it is common and usable as evidence.

Write your example in whichever register actually happened. If it was mostly brainstorming, say that plainly, then name the one output you checked and what you found. Stretching a brainstorming session into a claim of building something end to end is the kind of overreach a follow-up question exposes fast.

Publish the Work, Not Just the Line

Once you have the accomplishment written down, put it somewhere a reviewer can open it, not only in a private document. A randomized trial run with over 800,000 online certificate earners in developing countries who lacked college degrees found that prompting people to post a certificate they had already earned raised new employment by 6%, a gain of one percentage point, with the largest gains going to learners whose baseline employability was weakest. ³

That population is not yours and the paper is a working paper, so read it as a direction rather than a rate you can expect. What it tests cleanly is the cheap half of the advice, since it measures displaying a credential already held and not earning a new one. Nobody has run the same test on portfolio projects, but the reasoning transfers without much strain: a link, a public repository, a shared document is worth more where a reviewer can reach it than in a drawer.

If the project produced something you can point at directly, link it. How do you tell who did the work on an AI-built portfolio? is the honest account of how a reviewer reads a project like yours: not by assuming AI wrote it, but by asking what you decided, where the model was wrong, and what is still unfinished, which is exactly the account this article has been asking you to prepare.

See a sample report >