"I Just Got Lucky": 62 Studies Put the Impostor Habit at 9–82%── The habit comes from locating success outside oneself; working with AI adds the opposite error, so the check is a record of what one actually did

In 1978 the psychologists Pauline Clance and Suzanne Imes gave a name, the "impostor phenomenon", to the experience of achieving things while feeling unintelligent and fearing that one has fooled everyone. Where does the habit of saying "I just got lucky" after a success come from? My answer is that it comes from placing the cause of success outside oneself, in luck or in other people's help. Work done with AI adds the opposite error, overrating one's own performance. So I propose checking both against a record of what one actually did.

01"I got lucky" grows from placing the cause of success outside oneself

The habit of crediting success to luck or to other people has been studied in psychology as part of the impostor phenomenon. A systematic review by Dena Bravata and colleagues gathered 62 studies and found that the share of people showing it ranged anywhere from 9% to 82%.

Working alongside AI produces a misjudgment in the other direction. In experiments at Aalto University in Finland, people who used ChatGPT to solve logic problems rated their own scores higher than they really were.

Neither error, rating oneself too low or too high, is corrected by feeling alone. What I recommend is to write down the points where one made a decision, and later compare them with the results.

02Named in 1978, measured in 62 studies of 14,161 people at 9–82%

It is worth pinning down how the phenomenon was defined and how widely it has been found.

In their 1978 paper, Clance and Imes described high-achieving women. By any outside measure these women had succeeded. Yet they felt, privately, that they were not really intelligent, that they were intellectual phonies. Clance and Imes gave that experience a name. It was the name of a phenomenon, not of an illness.

Many researchers went on to measure it. In 2019 Bravata and colleagues published a systematic review, a study that pools earlier research on one question. Sixty-two studies met their criteria, covering 14,161 participants in total. The reported prevalence varied from 9% to 82% depending on the study.

Bravata's team also summarized what the phenomenon involves. One element is attributing success not to one's own ability but to outside factors such as luck or help from others. Success brings little satisfaction, and the person worries about being found out.

With a range as wide as 9% to 82%, no single share of the population can be named. What can be said is that among people who have achieved a great deal, many still do not trust their own success.

03When a document is drafted with AI, felt ownership and claimed authorship part ways

People who achieve a lot can rate themselves low. When writing with AI, the estimate of one's own contribution slips in a different way.

In 2024 Fiona Draxler and colleagues reported experiments on how people regard text that AI has written for them, with groups of 30 and 96 participants. Participants did not feel that AI-generated text was their own. Even so, many of them did not declare AI authorship when presenting the text publicly. The researchers called this the "AI Ghostwriter Effect", a ghostwriter being someone who writes in another person's name.

The same effect can arise for anyone who asks AI for a first draft at work. Someone who drafts product materials with AI, for instance, may struggle to say which parts are really their own work.

Side by side, the three show that estimates of one's own contribution can err low or high. The impostor phenomenon pulls the estimate down. Overrating with AI pushes it up. The ghostwriter effect separates what people feel from what they claim. Adding AI multiplies the ways an estimate can go wrong.

04Weiner held that whether a cause seems lasting shapes the expectation of the next success

Estimates of one's own work drift low or high depending on the situation. What changes depending on where one places the cause of success can be seen through the theory of the psychologist Bernard Weiner.

In a 1985 paper, Weiner sorted the perceived causes of success and failure by three properties. The first is locus: whether the cause lies inside or outside the person. The second is stability: whether the cause lasts or is temporary. The third is controllability: whether the person can change it.

According to Weiner, it is the second property, stability, that shifts the expectation of future success. If the cause seems lasting, one can expect to succeed again. If it seems temporary, the expectation does not rise.

What follows is my own inference. Luck and other people's help are not guaranteed to line up the same way next time. So someone who credits success to luck or help will find it hard to expect success again. This reading is mine, drawn from Weiner's theory. It is not part of his classification.

05ChatGPT users rated themselves above even their improved scores

On my reading in section 04, placing the cause of success outside oneself keeps the expectation of future success low. When people work with AI, however, self-estimates swing high instead. Experiments at Aalto University measured this.

Fernandes and colleagues at Aalto ran two experiments, with 246 participants in the first and 452 in the second. The second experiment was designed to check the first. Participants solved logic problems with ChatGPT and then estimated how well they had done.

Using ChatGPT raised their scores. According to the paper's abstract, performance improved by three points compared with a norm population. Yet participants overestimated their performance by four points. Their scores went up, but their ability to judge their own performance did not. The gap may matter in practice. This is my inference, not a finding of the study: someone who believes they did better than they did may be less likely to go back and check the answer the AI helped produce.

The Aalto experiments produced one more finding. Usually, people with low scores overrate themselves and people with high scores underrate themselves, a pattern known as the Dunning-Kruger effect. The Aalto researchers report that with AI, this pattern vanished.

06Misjudgment runs both low and high, and a record of decisions checks both

Working with AI pushes self-estimates upward. Meanwhile high achievers can rate themselves too low. Setting the two errors side by side suggests a way to check both.

Rating oneself low happens even to high achievers

Clance and Imes found the phenomenon among accomplished women. Bravata's 62 studies found it in 9% to 82% of participants. So when an accomplished person says "I got lucky", the words cannot always be taken at face value.

Working with AI produces the error of rating oneself high

In the Aalto experiments, participants using ChatGPT rated their scores above what they had actually achieved, and the more AI-literate rated themselves higher still. Confidence in work produced with AI is no substitute for checking it.

Recording one's decisions makes the contribution checkable

Draxler and colleagues found that the more participants shaped a text, the more they felt it was their own. Since that feeling shifts with how much one has edited, the edited passages themselves are better evidence of one's contribution than the feeling. They also called for ways to make it easier to show who contributed what. I therefore suggest writing down one's own decisions separately from the AI output. For example, note where you changed an AI draft and why.

I have not found an experiment showing that keeping such a record makes self-estimates more accurate. This is a proposed method, and its effect has not yet been measured.

07As AI does more of the work, the reasons to place success outside oneself may multiply

A record of decisions lets one describe one's contribution in facts. But as AI takes on more of the work, what that record contains may change.

When AI output makes up a large part of a piece of work, there may be more reasons to place the cause of success outside oneself, because the part of the success carried by the AI output really does lie outside oneself. This is my inference.

I found no study that directly measures the relationship between the impostor phenomenon and AI use. Two questions remain unmeasured.

I have felt this myself. After reworking an AI draft, I sometimes wonder whether I can honestly call the result my own work. The only thing that has helped at those moments is a note of what I changed and why. Without it, I am left with two feelings that pull in opposite directions: that the AI did most of it, and that the good parts were surely mine. Neither feeling is evidence.

Once those two are measured, it will be possible to say more concretely how people who work with AI should judge their own contribution. Until then, keeping a note of what I decided is the check I can actually carry out.

The habit of saying "I just got lucky" is a way of placing the cause of success outside oneself, in luck or help that may not last. Seen that way, a success gives little ground to expect the next one. Working with AI adds the opposite error of rating oneself too high.

Neither error is fixed by feeling. Each time I edit an AI draft, I now leave a line saying what I changed and why.