How People Perceive AI Use: What Happens When Others Find Out You Used ChatGPT

By /

I use AI every day. I write about it, I advocate for it, and I encourage educators to embrace it. But I also know that how we use AI affects how others see us. And a new study by Claessens, Veitch, and Everett (2026) offers some of the clearest evidence yet that outsourcing tasks to AI carries a social cost, especially when those tasks carry emotional weight.

Their paper, “Negative Perceptions of Outsourcing to Artificial Intelligence,” published in Computers in Human Behavior, reports findings from six preregistered studies with British participants. And the research is already playing out in real life. A New York Times article from February 2026 reported that a judge in Christchurch, New Zealand, discovered that a defendant’s apology letters in an arson case had been generated with AI.

The judge ran the letters through AI tools himself and recognized them immediately. He granted only a 5 percent reduction in sentencing for remorse, half of what the defense had requested. Everett, one of the study’s co-authors, was quoted in the article commenting on the case: “A.I. is a tool for efficiency, and it can be helpful, but it also typically involves, and signals, reduced effort” (Livni, 2026). The question the study asks is straightforward: how do people judge someone who uses AI to complete a task? The answer is consistent across every study, and it’s worth taking seriously.

When observers learn that someone used AI to complete a task, they rate that person as lazier and less competent. Instrumental tasks like writing code or planning a schedule carry a moderate cost. Socio-relational tasks like writing wedding vows or an apology letter hit much harder. Observers downgrade warmth, morality, and trustworthiness.

Claessens et al. describe the pattern clearly: “We highlight a tradeoff between efficiency and inferred moral character, authenticity, and value: outsourcing makes us think more negatively about not only the person and their motivations, but also the outsourced work itself” (p. 1).

And the penalty extends beyond the person to the product. In one experiment, participants read identical text. Some were told it was human-written. Others were told it was AI-assisted. Quality ratings were the same. But the AI-labeled text was judged as less meaningful, less authentic, and less deserving of reward: “text purportedly generated using AI was perceived to be less meaningful, less authentic, and less reward-worthy compared to the same text described as human-generated” (p. 9).

The writing didn’t change. The perception did.

Effort Helps, but Only So Much

Claessens et al. tested whether the manner of AI use affects judgment. It does, partially. Carefully crafting prompts and revising AI output improves perceptions compared to copying output verbatim. Deception about AI use damages morality and trustworthiness the most.

But even transparent, high-effort AI use on relational tasks draws more criticism than completing the task independently. A personalized AI model trained on someone’s prior writing didn’t reliably reduce the reputational cost either. The authors put it memorably: “An apology written by an AI, even in someone’s own style and fully ‘personalized’, arguably is still not an actual apology” (p. 11).

How People Perceive AI Us

Telling people that the AI user cared deeply about the task and wanted to get it right didn’t fix things. For relational tasks, observers infer that if something truly matters, you should do it yourself.

According to the authors, the mechanism involves second-order inferences: “Our results suggest that reduced effort is important not only because people value time and energy spent, but because expending less effort through outsourcing signals second-order perceptions that people are being less authentic and care less about the task” (p. 1).

Now here’s where I want to push the conversation into territory this study doesn’t cover but that I think is directly relevant.

If people judge AI-outsourced work as less authentic and less meaningful even when the quality is identical, then education faces a perception problem alongside a learning problem. Students who use AI in their writing may produce technically competent work that their teachers, peers, and future employers perceive as less valuable. The process behind it signals less effort and less care, regardless of actual quality.

Kalantzis and Cope (2025) argued that literacy in the age of AI should be understood as design agency: the active, intentional work of making meaning with purpose and voice. Claessens et al.’s findings reinforce that argument from a social psychology angle. When writing is perceived as outsourced, it loses meaning in the eyes of others. The work becomes product without process.

I’ve been thinking about this since covering Cleland et al.’s (2025) AMEE guide on AI disclosure. Transparency about AI use matters. But Claessens et al. show that transparency alone doesn’t eliminate negative judgment, especially for tasks that feel personal or consequential. Disclosing AI use is the right thing to do, and I’ve argued that strongly. We should also prepare students for the reality that disclosure may come with a reputational cost.

Niloy et al. (2024) found in their creative writing experiment that ChatGPT improved elaboration and presentability but reduced originality and accuracy. Claessens et al. add another layer: even if quality holds up, perceived authenticity drops. Two different studies, two different angles, same conclusion. Effort, voice, and care matter to readers.

What Educators Should Take From This

I don’t think this study should make anyone afraid of using AI. I use it constantly and advocate for it strongly. But we need to teach students something beyond prompt engineering and output evaluation. They need to understand the social dynamics of AI use.

A job application letter carries different expectations than a data analysis report. Wedding vows carry different weight than a meeting agenda. Context shapes how AI use is interpreted, and students need to develop judgment about those contexts. When should they use AI? When should they disclose it? When might AI involvement undermine the purpose of the task itself?

Perkins, Roe, and Furze’s (2024) AI Assessment Scale offers a useful framework. Different levels of AI involvement suit different tasks. Level 1 (no AI) might be the right call for deeply personal or relational writing. Levels 2 through 4 work well for analytical, instrumental, or collaborative tasks. The scale helps students make intentional decisions about when and how much AI to use.

Claessens et al. have given us strong evidence that perceptions of AI use are nuanced, context-dependent, and consequential. We should be preparing students for that reality, not just teaching them to operate the tools.

Reference

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top