Imagine a creator comparing three ways to teach a subject: a live workshop, recorded lessons and a written guide. A fictional AI planning tool insists on the workshop. The creator finds it bossy. Another version offers a warm recommendation while quietly hiding the other two routes.
The second experience might feel smoother. Yet the creator has less opportunity to compare. This hypothetical contrast raises a useful question for human–AI relationships: how can we distinguish our reaction to a recommendation from the choices actually available?
Reactance concerns a perceived threat to choice
Psychological reactance describes motivation to recover a freedom that seems threatened. It offers one way to understand resistance to advice, rather than a label for every disagreement.
In Dillard and Shen’s 2005 research, two studies examined persuasive health messages: one concerned flossing, with 196 participants; the other concerned limiting alcohol intake, with 200. The researchers found support for measuring reactance through a combination of reported anger and negative thoughts. These were communication studies, not tests of modern conversational AI.
That distinction matters. A reader can object to an AI answer because its facts are wrong, its assumptions are unsuitable or its manner feels controlling. Several objections can coexist. Describing the person as resistant does little to identify which problem needs attention.
Advice can provoke an opposite choice
A 2004 paper by Fitzsimons and Lehmann reported four studies of recommendations that contradicted consumers’ initial impressions. Under the conditions they examined, unsolicited recommendations could prompt people to choose against the advice, rather than simply disregard it. The experiments involved product choices and recommendations attributed to experts or intelligent agents; they did not study today’s language-model conversations.
This suggests a question to ask yourself when rejecting an AI suggestion: would you still prefer the alternative if the recommendation had never appeared? The answer may be yes, for good reasons. The purpose is to recover those reasons, so that opposition to the adviser does not become the whole basis for the decision.
Hidden restrictions may attract less resistance
Görlitz and Rosenthal-von der Pütten’s 2026 technology-paternalism study, available online on 31 March 2026, examined 137 participants using shopping prototypes. Five conditions ranged from no AI influence through visible nudges and restrictions to undisclosed filtering. Some restrictions could be reversed; others could not.
Initially, the visible-nudge condition produced greater measured reactance than the other conditions, although it left the choices available. In the covert condition, out-of-budget items were hidden without an explanation or an override. Reactance increased after participants received information explaining the restrictions and their assigned condition.
The study used a particular online shopping scenario and a predominantly Germany-based sample. It does not establish how every chatbot, customer or culture responds. Its findings nevertheless warn against treating a lack of immediate resistance as proof that people understand the boundaries of their choice.
Examine the recommendation and the reaction separately
For everyday AI-assisted work, I suggest keeping two questions apart. First, is the proposed action appropriate to the task? Second, is the recommendation being presented in a way that lets you examine and challenge it? These review questions are practical proposals; the studies did not test this routine.
A friendly answer could contain a weak argument. A blunt answer could identify a real constraint. Changing the tone may improve the conversation while leaving the underlying evidence untouched. Ask for the assumptions, relevant alternatives and reasons that would change the recommendation before deciding whether to accept it.
If a tool cannot provide a requested comparison, establish what is missing. It might lack the information or functionality needed. An unexplained omission should remain a gap to investigate, rather than become evidence that the omitted route is unsuitable. Where the decision matters, check the original catalogue or task materials too. A tool’s explanation of its own omissions is not an independent check.
Make the creator’s priorities explicit
Return to the fictional teaching project. Suppose the creator wants material people can use at different times and has limited capacity for live delivery. A workshop-only recommendation needs to account for those priorities. It should not quietly replace them with an assumed goal of maximising live attendance.
The creator could request a comparison of the three formats against preparation effort, the audience’s access needs and opportunities for questions. They could then ask which trade-offs favour the recommended format. These are proposed review steps; no sales or learning outcome is implied.
The final decision might still be a workshop, perhaps with supporting written material. It might be the guide, or a small trial before producing a larger course. What matters is that the recommendation has been assessed against the creator’s stated purpose, with alternatives available for consideration.
Give a shortlist an understandable boundary
For business owners designing an AI-assisted journey, a useful review would ask:
- Which goal and preferences produced this recommendation?
- Which options were excluded, and for what reason?
- Can the person revise their preferences or request another comparison?
- What remains unavailable, and is the reason understandable?
This does not require putting an entire catalogue on every screen. A shortlist can have a clear purpose and an accessible route to a wider comparison. The MaryChuks article on helping customers compare digital products explores how to make those comparisons meaningful without assuming that fewer choices always improve decisions.
Treat these questions as design proposals to test with the intended audience. Ask people to explain their options and the selection criteria in their own words. Observe whether they can find an alternative when the first recommendation does not fit.
Assess understanding as well as agreement
A review should leave room for a person to reject a recommendation with a clear reason. Agreement alone cannot tell you whether the person understood the trade-offs, accepted an assumption accidentally or simply followed the most visible route.
For your own work, write down the reason for the choice you make, including any unresolved uncertainty. For a customer journey, investigate where people misunderstand what they can change. A useful AI recommendation should make the decision easier to examine and the remaining choices easier to understand.
Sources
- Dillard and Shen — reactance and persuasive communication (2005), Communication Monographs, DOI: 10.1080/03637750500111815.
- Fitzsimons and Lehmann — reactance to recommendations (2004), Marketing Science, DOI: 10.1287/mksc.1030.0033.
- Görlitz and Rosenthal-von der Pütten — technology paternalism (2026), Computers in Human Behavior: Artificial Humans, DOI: 10.1016/j.chbah.2026.100290.
Discover more from Marychuks.com AI, Psychology, Business & CreativeVerse
Subscribe to get the latest posts sent to your email.