AI Can Polish an Application. What Makes Your Skills Believable?

A smooth paragraph is only the beginning

An application can sound confident while leaving the most useful question unanswered: what could this person actually do in the role? An AI tool can help organise a draft, but the reader still needs a reason to connect its claims with your experience. Replacing a few adjectives with more impressive ones does not create that connection.

For a remote candidate, the practical task is to make relevant work understandable without relying on a long explanation, a recognisable employer name or private client material. This article offers an evidence check you can use before submitting. It is a way to improve the clarity of your application, not a promise of passing a screening system.

What the new paper studies — and what it does not prove

A September 2026 arXiv preprint by Itai Ashlagi, Ramesh Johari, Jon Kleinberg and Anushka Murthy models a hiring market in which AI can make application materials less informative about fit. In the model, firms respond by relying more on visible experience. Candidates with little experience but a good underlying match can lose screening opportunities. An intermediate assessment can provide additional evidence before more expensive screening.

This is a theoretical result under the paper's assumptions, not a field experiment measuring your chances on a particular platform. It does not establish that every employer behaves this way, or that an assessment is automatically fair. Our practical inference is narrower: make your claims easier to examine, and ask what evidence a hiring step is intended to collect.

Build an evidence card, not another promotional claim

Choose one responsibility from the advert and make a short private note with five fields: context, your contribution, method, check and limit. Context describes the problem. Contribution separates what you did from what the team did. Method explains your approach. The check shows how you evaluated the result. The limit makes clear what the example cannot demonstrate.

For example, a hypothetical developer might describe a personal tool that validates imported rows. Their contribution could be the validation rules and tests, while a library handles file parsing. A useful check would explain what happens with missing fields and duplicate records. A sensible limit would acknowledge that the tool was tested on sample files rather than a production customer workload.

A hypothetical finance candidate could explain a reconciliation exercise using invented transactions clearly labelled as sample data. They can describe how discrepancies were found and reviewed without inventing client results. The evidence is the reasoning and the boundaries, not a made-up revenue figure. Select examples you can discuss comfortably if a reviewer asks how a decision was made.

Make the example easy to inspect safely

Match the amount of material to the request. A short explanation and one relevant work sample can be more usable than an unexplained folder of files. Give links descriptive names, check that the intended viewer can open them and remove personal contact details or confidential information that the example does not require.

If you cannot share the original, describe the method at a safe level or create a small independent demonstration with synthetic data. Label it as a demonstration. Do not imply that a recreated example is the original client deliverable, and do not upload an employer's files to an AI service just to obtain a cleaner explanation.

Before submitting, try explaining one trade-off aloud without reading the application. If you cannot explain a term or defend a number, revise it. AI assistance should not leave you responsible for claims that you do not understand. Follow any stated application policy on AI use, and answer questions about your contribution accurately.

An assessment still deserves questions

A test may give an employer evidence that a written application cannot. That does not mean every test is proportionate or suitable. Ask what skill is being assessed, how long the task is expected to take, what tools are permitted and how the output will be used. Clarify accessibility needs and technical requirements before committing substantial time.

Separate a contained demonstration from work that appears to solve a live business problem. You can ask for a smaller example or clarification without claiming that the research paper proves a particular task is improper. Keep your decision grounded in the actual request and the terms you receive.

Check the application before sending

Keep the evidence note for your own preparation. Use only the relevant parts in the application, and respect the requested format. The objective is a clearer basis for evaluation, not a larger volume of material.

• Choose one essential responsibility and one relevant example you can truthfully explain.
• Separate your contribution from team, template, library or AI assistance.
• Describe the method and how you checked the work; do not manufacture a success metric.
• Remove confidential material and label synthetic demonstrations clearly.
• Check access to the work sample and follow the employer's AI-use rules.
• Clarify the purpose, effort and permitted tools for any assessment.

Review current remote roles and their requirements →Check role fit before preparing an application →Protect confidential work when preparing a portfolio →