At work, it is common to pass a draft written by generative AI to the next stage without changing a single word. Is staying the same also a choice? It is. Whenever change is possible, keeping things as they are is something you chose. And that choice is easily pushed by defaults and by our habit of treating harm from inaction more lightly.
01If change is possible, keeping things as they are is a choice
In a lecture given in 1945, the French philosopher Jean-Paul Sartre said that a person can always choose, but must understand that not choosing is still a choice. The line rejects the idea that you can escape responsibility by declining to decide.
Behavioral economics has since measured a related pattern in numbers. In 1988 the economists William Samuelson and Richard Zeckhauser showed that people pick the "leave it as it is" option more often than its merits justify. Later studies of pension plans and organ donation measured the same pull as large gaps in real outcomes.
Sending an AI draft onward without edits is one of these choices. Because editing was possible, the draft that went through is a draft someone chose. So when you decide to leave something unchanged, check your reasons once, just as you would when you change it.
02Samuelson and Zeckhauser named the pull toward the status quo in 1988
In their 1988 paper Samuelson and Zeckhauser pointed out a gap between textbook problems and real decisions. Most real decisions, they wrote, include a status quo alternative. Through experiments they showed that people choose that alternative disproportionately, and they called the tendency status quo bias.
Three related tendencies carry similar names. This essay is mainly about the first, but the other two return in later sections.
Status quo bias
Choosing the option that keeps things as they are more often than the other options would warrant.
Default effect
A setting that applies when nobody acts tends to become the final outcome.
Omission bias
Judging harm caused by not acting more lightly than harm caused by acting.
At the third box in the chart, a person decides whether to change anything. Most people lean toward the status quo more than the merits of the options would predict. Even someone who believes they made no choice has chosen the status quo option.
The finding was retested in 2021. A team led by Qinyu Xiao rebuilt the 1988 scenarios and found strong support for the bias in three of four. In the remaining scenario, the support was not strong. The pull toward the status quo does not appear with the same force everywhere.
03Default effects show up in pension enrollment and organ-donor consent
Inside an institution, the status quo option usually takes the form of a default: the setting that applies if nobody does anything. One study measured how strong a default can be in a real workplace.
In a 2000 paper, Brigitte Madrian and Dennis Shea examined the 401(k) retirement plan at a US company. Among employees hired when they had to sign up themselves, participation at three to fifteen months of tenure was 37%. Among employees hired after the company enrolled people automatically, with the option to leave, it was 86%. The comparison is between two groups hired at different times, not the same people before and after.
Organ donation showed a gap in the same direction. In 2003 Eric Johnson and Daniel Goldstein reported an online experiment in which consent was about twice as high when donation was the default as when it was not. They measured stated consent, not the number of organs actually donated.
| What to compare | Opt in to join | Joined by default, opt out to leave |
|---|---|---|
| 401(k) participation (3–15 months of tenure) | 37% | 86% |
| Organ-donor consent (online experiment) | Lower | About twice as high |
| Outcome for someone who does nothing | Not enrolled | Enrolled |
Put side by side, the two studies show that the outcome for people who do nothing is decided by whoever set the default. The pension plan and the donation scheme kept their content. The main thing that changed was which side the inactive person landed on. Because the pension comparison involves groups hired at different times, it cannot show that this was the only difference.
Passing an AI draft unchanged also has a status quo option: the path where the text the AI produced becomes the final version. One example is a promotional document drafted with generative AI that clears review without a single edit. I have not, however, found a study that directly measures the default effect for AI drafts. What can be said is that the 1988 definition applies to this situation as well.
04People judge harm from inaction more lightly than harm from action
Employees hired before and after the default changed showed pension participation of 37% and 86%. Why do people stay with defaults so firmly? Research in psychology and economics points to three reasons.
The first is a habit that Ilana Ritov and Jonathan Baron documented in 1990. They asked participants whether they would vaccinate a hypothetical child. Even when the vaccine's own risk of death was smaller than the disease's, many hesitated. A death from the vaccine would follow from something the parent did. A death from the disease would follow from something the parent did not do. People weigh the second kind of harm more lightly.
The second and third reasons come from the pension study. Madrian and Shea pointed to inertia, the wish to avoid the effort of changing anything. They also noted that employees seemed to read the default as advice from the company. That is why their paper is called "The Power of Suggestion."
The three forces come from separate studies. Which one matters most will differ from case to case, and I do not rank them. What they share is that each is a reason people stay where they are.
05Workers who trusted generative AI more reported less critical thinking
People treat harm from inaction lightly and tend to stay with defaults. How much, then, do people check what a generative AI tool gives them? One survey has put numbers on that.
In 2025 Hao-Ping Lee and colleagues at Carnegie Mellon University and Microsoft Research surveyed 319 knowledge workers about 936 cases in which they had used generative AI at work. The results ran in two directions. The more people trusted generative AI, the less critical thinking they reported. The more confident they were in their own abilities, the more critical thinking they reported.
The survey relies on self-reports. It shows that two things appeared together, not that trust caused people to check less. The arrow could also run the other way: people who find checking tedious may be quicker to trust AI.
Even so, the survey points in the same direction as the vaccination experiment and the pension study. The survey did not, however, measure whether people edited the text. Whether those who trust AI more are more likely to pass drafts unchanged remains an inference from these results.
06Keeping a draft unchanged needs as much justification as changing it
People who trusted generative AI more reported less critical thinking. Taken together, the studies support three points about keeping things as they are and about setting defaults.
Not changing is a choice
Be able to explain passing a precedent or an AI draft as your own decision.
Whoever sets the default decides
People who prepare templates and precedents should assume these will become the final form for many others.
More trust, less checking
Before passing an AI draft, add one step that compares it with a changed version.
First, not changing something remains a decision you made. Sartre said that not choosing is a choice, and Samuelson and Zeckhauser measured how often the status quo wins. Anyone who passes a precedent or an AI draft should be able to say, in their own words, why they let it through.
Second, whoever sets a default decides the outcome for many people in advance. In the 401(k) study the gap was 37% against 86%; in the donation experiment, consent roughly doubled. People who prepare templates or precedents should build them on the assumption that many users will keep them exactly as they are.
Third, people who trusted generative AI more reported thinking less critically. The survey of 319 workers shows only an association, but its direction was clear. Before letting an AI draft through, add a step that compares it with a version where at least one thing is changed.
The point of this sequence is to treat the status quo as one option among others. If you decide not to change anything, write the reason in one sentence and record that you chose it.
07Whether writing reasons weakens the pull of defaults is still unmeasured
The step of writing down why you are not changing something is my own inference from these studies. Nobody has yet measured whether it works.
The research I checked shows that defaults are powerful and that people discount harm from inaction. I found no study that tests whether writing out reasons weakens a default's pull. Even the 1988 finding was not strongly replicated in one of the four scenarios in 2021. The strength of the pull toward the status quo varies with the situation.
Situations where an AI draft acts as the default will multiply to the extent that more people use generative AI at work. Some workplaces may soon have more people reviewing AI drafts than writing documents from scratch. How strongly the default effect operates on AI drafts is something nobody has measured.
Even so, I do not think we need to wait for that measurement. Writing one sentence of reasons costs almost nothing. Its effect is unknown, but the sentence stays behind as a record you can read later.
- At one US firm, 401(k) participation was 37% when workers had to opt in and 86% under automatic enrollment. For people who do nothing, whoever sets the default decides.
- In Ritov and Baron's experiments, many people hesitated to vaccinate a child even when the vaccine's risk of death was smaller than the disease's. Discounting harm from inaction makes the status quo easier to pick.
- In a survey of 319 knowledge workers, those who trusted generative AI more reported less critical thinking. Passing an AI draft unchanged is a decision whose reasons deserve checking.
Staying the same is a choice. If change is possible, keeping things as they are is something you chose. Research points to defaults and our lighter view of inaction as reasons it is chosen so often. Passing an AI draft unchanged also has a status quo option, so there is reason to expect the same tendencies there, though no study has measured this directly. So when you decide not to change something, check the reason once.
- Jean-Paul Sartre (trans. Philip Mairet, Marxists Internet Archive). Existentialism Is a Humanism. 1946. https://www.marxists.org/reference/archive/sartre/works/exist/sartre.htm (not choosing is still a choice)
- William Samuelson, Richard Zeckhauser. Status Quo Bias in Decision Making. Journal of Risk and Uncertainty, 1988-03. https://ideas.repec.org/a/kap/jrisku/v1y1988i1p7-59.html (most real decisions include a status quo alternative, which people favor)
- Qinyu Xiao, Choi Shan Lam, Muhrajan Piara, Gilad Feldman. Revisiting status quo bias: Replication of Samuelson and Zeckhauser (1988). Meta-Psychology, 2021-05-24. https://open.lnu.se/index.php/metapsychology/article/view/2470 (strong support in three of four scenarios)
- Brigitte C. Madrian, Dennis F. Shea. The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior. NBER Working Paper 7682, 2000-05. https://www.nber.org/papers/w7682 (37% versus 86% participation; inertia and defaults read as advice)
- Eric J. Johnson, Daniel Goldstein. Do Defaults Save Lives? Science, 2003-11-21. https://www.dangoldstein.com/papers/DefaultsScience.pdf (donor consent about twice as high under opt-out)
- Ilana Ritov, Jonathan Baron. Reluctance to vaccinate: Omission bias and ambiguity. Journal of Behavioral Decision Making, 1990. https://cris.huji.ac.il/en/publications/reluctance-to-vaccinate-omission-bias-and-ambiguity/ (reluctance to vaccinate when the vaccine itself can cause death)
- Hao-Ping Lee, Advait Sarkar, et al. (Carnegie Mellon University, Microsoft Research). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. CHI 2025, 2025-04. https://advait.org/files/lee_2025_ai_critical_thinking_survey.pdf (319 workers, 936 cases; higher trust in GenAI associated with less critical thinking)
