AI-Written Applications and Assignments: What Hiring Teams Should Verify

Every marketer’s application reads well. The ad performance reports submitted as take-home assignments look just as polished: clean tables, crisp prose. You suspect AI helped. But you have no basis to penalize the candidate, and no reason to take the analysis at face value.

Two questions matter when an application or assignment appears to be AI-written: Did the candidate follow the AI rules you shared in advance? And can they make the judgments the role requires? Polished writing answers neither. Ask what supports their conclusion, which alternatives they rejected, and why they revised their work. Then change a condition and see how they adjust their reasoning.

A detection tool cannot answer those questions for you. NHS Employers, which supports HR and employment practice across the UK’s NHS, stated in its August 2025 guidance on AI use in candidate applications that no official commercially available technology currently detects AI use. It advises against automatically rejecting applications on suspicion alone. If a candidate otherwise looks suitable, it recommends a verbal conversation to establish whether they have the understanding and knowledge required.

Our guide to AI-free assessment and AI use rules covers where to draw the line on AI assistance. Once those rules are set, the next step is to verify the work candidates submit.

Consider a performance marketer assignment for a skincare brand that sells through its own online store. Candidates receive two weeks of performance data from two ad channels and must allocate KRW 10 million for the next two weeks. One submits a polished report recommending a budget increase for Channel B. What should the hiring team verify?

Separate AI Use From Job Competency

Writing an application without help does not prove someone can interpret ad performance data. Using AI to polish sentences does not prove they lack analytical ability. Keep the process for investigating suspected rule violations separate from the assessment of job judgment.

Apply that distinction when reviewing detection results, too. In their 2023 study of AI text detection tools, published in the International Journal for Educational Integrity, Debora Weber-Wulff and colleagues tested 12 publicly available tools and two commercial systems, Turnitin and PlagiarismCheck. They concluded that the tools were neither accurate nor reliable and tended to classify AI-generated text as human-written. Those tests used educational writing, so the findings do not transfer directly to every current tool or hiring application. They do, however, support a practical principle: weigh the candidate’s explanation and supporting evidence alongside any AI detection score, rather than letting the score settle the question.

Give Every Candidate the Same Rules Up Front

Your instructions should reflect what you want to assess: data interpretation, basic judgment without tools, or the ability to complete work with AI. Anthropic’s candidate AI guidance sets rules for each stage. Candidates may use AI to refine an application after writing the first draft themselves. Assignments and live interviews must be completed without AI unless the company explicitly permits it. Anthropic also explains why: it wants to understand each candidate’s own abilities and thinking. NHS Employers likewise suggests adding language about monitoring AI use to job descriptions and application forms, and creating candidate guidance that explains permitted uses.

Specify which tools candidates may use, when they may use them, what process records to submit, what information must stay out of tools for security reasons, and how to ask questions. Do not introduce new rules after assigning the task.

Request only the process records needed for the assessment. There is no reason to ask for a candidate’s entire personal account history or private materials. Have candidates retain records that explain their decisions within the assigned work, using the same materials provided to everyone.

Our review of AI detectors for application essays explains how detection works and its use during resume screening. Beyond that stage, focus on gathering evidence of job judgment instead of repeatedly checking AI detection scores.

Ask What Supports the Recommendation

Return to the marketer’s report. Before judging the prose, ask which numbers the candidate treated as evidence of performance, what denominator they used, which alternatives they rejected, and what else they would investigate. In this assignment, the channel with the higher ROAS is A at 300%, while B brings in more first-time customers at 140. Ad-attributed revenue includes repeat purchases by existing members, and coupon discount costs differ: KRW 1.5 million for A and KRW 500,000 for B. The recommendation depends on which figures the candidate uses.

Change a Condition and Watch the Reasoning

Ask the same candidate, “If you discovered duplicates in Channel B’s first-time customer count, what would you check first?” This is not a trick to catch a memorized answer. It reveals which calculations and conclusions they revisit when the counting criteria change. When assessing the same competency, give every candidate the same question and the same opportunity to respond.

Suppose Candidate C says only, “B is better, so I would still increase its budget.” Candidate D says, “I would remove the duplicates, recalculate ad spend per first-time customer, and compare both channels’ ad-attributed revenue after subtracting coupon discount costs.” Record these answers as evidence of how candidates interpret data and revise their judgment. They do not establish whether either candidate used AI.

Assess the AI Workflow When the Role Requires It

For roles that involve working with AI, examine how candidates frame the problem and constraints, what they use to check the output, and why they correct errors. Polished work can still carry risks if no one has checked its basis. An imperfect first output can also lead to a sound result through careful review and revision. Give those behaviors credit in the assessment.

The Telta AI Literacy Assessment covers the process of delegating work to AI, reviewing its output, and completing the task. Interview Pro helps organize interview questions and evidence from candidates’ answers. To observe performance more directly, consider an Assessment suited to your goal. The purpose of these two products is to establish the evidence behind a candidate’s judgment, rather than determine who wrote the submission.

Record What You Verified and What Still Needs an Answer

Replace “It looks like they used AI” with a record of what the candidate understood and what they could not explain. If you suspect a rule violation, follow the procedure established in advance and give the candidate a chance to respond. When evidence of judgment is thin, follow NHS Employers’ recommendation to explore understanding in conversation, or assign a short task that applies the same principle to new data. Record assessment findings and rule compliance decisions with their respective evidence so you can later see exactly what each decision rested on.

See our work sample task example for shared task materials and behavioral criteria, and our structured interview guide for follow-up questions and documentation. When reviewing finalists, use an evidence comparison table to distinguish what you have verified from what still needs validation.

Frequently Asked Questions

Should a High AI Detection Score Disqualify a Candidate?

A score alone is not a sound basis for rejection. Review what it means and its limitations alongside the rules you shared, the candidate’s explanation, and supporting evidence. NHS Employers also advises against automatic rejection based on suspected AI use and recommends a verbal conversation to assess understanding.

Can You Assess Ability When AI Is Allowed?

Yes. Define where AI is permitted and which behaviors you want to observe, then assess how candidates prompt, review, revise, and apply the output. If part of the assessment requires work without tools, design a separate segment for that purpose.

Related Reading

• How to Reduce Bad Hires: A Hiring Improvement Guide for Growing Companies
• Work Sample Task Example: From Brief to Answers, Rubric, and Evaluation Notes
• Hiring Assessment Methods Compared: Interviews, Work Samples, and SJTs
• What Is a Structured Interview? Definition, BEI Comparison, and 4 Design Steps

Sources

• NHS Employers · Guidance on the Use of Artificial Intelligence in Candidate Applications (2025) — AI detection technology, conversations to assess understanding instead of automatic rejection, and candidate notices and guidance.
• Anthropic · Guidance on Candidates’ AI Usage — AI use rules for applications, assignments, and live interviews, and the reasoning behind them.
• Weber-Wulff et al. · Testing of Detection Tools for AI-Generated Text (2023) — Tests of 14 detection tools, accuracy and reliability limitations, and a tendency to classify text as human-written.
• Telta · AI Literacy Assessment — Assessment of the process of delegating work to AI and reviewing its output.