Summary
AI authorship labels can change creative comparisons and encourage attempts without proving better performance. Use that confidence alongside revision and fair evaluation.
At a glance
People may view AI as a lower creative benchmark and consequently feel more capable. That can encourage an attempt without demonstrating improved skill; assess the work through relevant feedback.
Seeing an impressive piece of work can make you question your own ability. But learning that it was generated by AI may change the comparison: instead of “I could never do that,” the response becomes “I could do better.” The work itself may be unchanged. What has shifted is the creator you imagine behind it.
That is the central question in Taly Reich and Jacob Teeny’s study, Does artificial intelligence cause artificial confidence?, published online in 2025. Their research found that creative content believed to come from generative AI could raise confidence in people’s own relevant abilities compared with content attributed to a human peer. The reported pattern covered jokes, stories, poetry and visual art, and could increase willingness to try the activity. The authors caution that this confidence may be unwarranted.
The Label Can Change the Comparison
Imagine encountering the same joke under two different descriptions: one identifies a human creator, the other an AI system. Holding the content steady helps separate a response to the joke from a response to its presumed author. If your assessment of yourself changes, quality alone cannot explain the change.
This is different from receiving useful assistance while making something. You may feel more willing to write a story after seeing an AI-labeled example even though you have not learned a new writing technique. Encouragement to begin and improvement in the finished result are separate outcomes.
Why AI Can Become a Lower Benchmark
Social comparison provides a useful explanation. A colleague’s presentation or a friend’s photography can become a standard against which you judge yourself. AI-generated work can also supply a comparison point, even though its creator is not a person.
The process can be understood in three parts: notice a creator, estimate that creator’s relevant capability, then assess your own capability in relation to it. This is an explanatory framework, rather than a claim that every viewer consciously follows three steps.
Someone who regards creativity as distinctively human may treat an AI system as a weaker creative benchmark. A polished result may still be discounted as lacking personal experience or emotional meaning. The comparison can therefore favor the viewer without providing new evidence about the viewer’s work.
That interpretation concerns perceived capability. It does not establish that all AI output is inferior, or that every person responds to the label in the same way. The relevant belief may also differ by task.
Confidence Is Different From Demonstrated Skill
A useful distinction is between predicting that you can create something good, rating your own attempt and having other people assess the result. These measures need not move together. A person can become more enthusiastic about a caption or story without producing a better one.
The supplied account uses cartoon captions to illustrate the gap between self-evaluation and outside assessment. Its precise ratings and judge details could not be independently checked, so they should not be treated as established numerical evidence here. The broader lesson is to keep confidence and performance separate when assessing whether people overvalue their own work.
Inflated confidence is not necessarily useless. Being willing to attempt an activity can create an opportunity to practice. But feeling more capable is the beginning of that opportunity, not proof that the learning has already occurred.
Using Encouragement Without Losing Judgment
The source proposes three stages for creative teams. During an initial idea session, an AI example might make starting feel less intimidating. Treat that as an invitation to generate alternatives, rather than evidence that the team’s first alternative is superior.
During refinement, bring in demanding examples, including expert human work where relevant. Ask what makes an example effective: structure, clarity, originality or sensitivity to its audience. This redirects comparison toward features that can guide revision.
During evaluation, test proposals against the same criteria. Compare human and AI suggestions in parallel when useful, with authorship concealed from evaluators where practical. The test should concern whether an idea works for its purpose. These are proposed working practices, not interventions whose effectiveness this study establishes.
The Effect Depends on the Reference Point
The source also raises factual explanations, such as how rain forms, and descriptions of AI as emotionally capable. These examples highlight important boundary questions: would a system seen as knowledgeable create a more intimidating comparison, and would attributing emotional depth change how people judge its creativity? The available original abstract does not establish all the account’s detailed results, so neither example warrants a universal rule.
People’s beliefs about AI can change. A system dismissed in one domain may be respected in another, and a creator label can matter differently across audiences. Use any burst of creative confidence to begin, then let practice, revision and relevant feedback tell you what has improved.
Frequently asked questions
Can an AI label change confidence?
The reported research found higher relevant creative confidence compared with human-peer attribution across several creative tasks.
Does confidence prove improvement?
No. Willingness, self-ratings and externally assessed performance are separate outcomes.
How can a team use this carefully?
Use examples to start, refine against demanding standards and evaluate proposals with consistent criteria.
Is the effect universal?
No. Perceptions, tasks and audiences matter; broad claims beyond tested contexts are unwarranted.
