Some people look at a pie chart and instantly see the story. Others stare at the same chart, squint, and quietly wish someone would just give them the numbers in a neat table. If you have ever sat in a meeting where half the room loved the colourful dashboard and the other half asked for “the raw data, please,” you have witnessed what psychologists call differences in cognitive style.
A recent research study asks how these differences play out when we bring generative AI into the decision-making process. Their central question is a timely one: does ChatGPT make us smarter decision-makers, or does it simply give us another way to switch off our brains?
Intuitive minds and analytical minds
Psychologists have long sorted thinkers into two broad styles. Intuitive thinkers make quick, holistic judgements based on gut feeling and experience. They tend to love visuals and big-picture summaries. Analytical thinkers prefer to work methodically, step by step, and are drawn to structured information such as tables and lists. Neither style is smarter than the other. They are simply different ways of processing the world, and research suggests they are fairly stable traits that do not change much with training.
Two theories help explain why this matters. The first, cognitive fit theory, says that we make better decisions when information is presented in a format that matches the way we think. Give an analytical thinker a neat table and they will thrive. Give them a messy infographic and they will burn mental energy just decoding it. The second, cognitive continuum theory, suggests that the structure of a task nudges us toward intuitive or analytical modes. Fuzzy, visual tasks invite gut reactions; clear, structured ones invite careful reasoning.
Now add a third player: an AI assistant that can explain, summarise, and advise on demand. In theory, this should help everyone. The intuitive thinker gets a structured breakdown; the analytical thinker gets a quick summary. But AI also brings a familiar risk: overreliance. History offers sobering examples, from healthcare to criminal justice, where people trusted algorithms too much and paid the price.
Spotting fake reviews with a robot sidekick
To explore this, the researchers designed an experiment around a task many of us face while shopping online: deciding whether a product review is genuine or fake. Participants were shown six reviews, three real and three deceptive, each accompanied by sentiment statistics. Here is the clever twist. Three reviews came with the statistics shown as a graph, a format suited to intuitive thinking, and three came with the same information in a table, a format suited to analytical thinking.
Participants were split into two groups. The control group worked alone. The treatment group first made their own initial judgement, then could chat with ChatGPT as much as they liked before making a final decision. The researchers recorded the conversations, measured how mentally drained participants felt, and assessed each person’s cognitive style. They could then see whether each person was working in a format that matched their thinking style, or one that clashed with it, and how that affected their reliance on AI.
In this pilot stage, 45 people took part. Analytical thinkers, perhaps unsurprisingly, took longer to finish the tasks than intuitive ones.
The surprising upside of a bad fit
Here is where things get interesting. You might expect that when information matched a person’s thinking style, they would make the best decisions. Instead, the opposite pattern appeared in how people used ChatGPT.
When the format fit their style, participants tended to overrely on the AI. They seemed comfortable, perhaps a bit too comfortable, and accepted the AI’s suggestions without enough scrutiny. When the format clashed with their style, participants relied on the AI more appropriately. They engaged more critically, using the AI as a helpful second opinion rather than an oracle.
For analytical thinkers, this made a real difference to accuracy. Those working in a mismatched format, and therefore using AI more sensibly, got about 62 percent of their decisions right, compared with about 46 percent when the format matched their style and they leaned on AI too heavily. For intuitive thinkers, accuracy stayed flat at around 50 percent regardless, which is roughly the success rate of a coin toss.
One possible interpretation is that a little friction keeps us alert. When information feels easy and familiar, we relax and let the AI do the thinking. When it feels slightly awkward, we stay on our toes and question what the AI says. It is a bit like driving: the familiar road home is exactly where your mind wanders, while an unfamiliar route keeps you paying attention.
What this means for AI at work
These are early findings from a small pilot, and the authors plan a much larger, longer-term study with working professionals solving real-world problems. Still, the implications are thought-provoking.
First, the study challenges the assumption that the goal of AI design should always be to make things as smooth and comfortable as possible. If comfort breeds complacency, a little thoughtful friction might actually improve decisions. AI tools could, for instance, prompt users to explain their reasoning or highlight uncertainty in ways that keep people engaged.
Second, it suggests that organisations should stop asking simply “Is this AI tool accurate?” and start asking “Is the combination of this person, this task, and this tool working well?” The same AI can help one employee and mislead another, depending on how they think and how information is presented to them.
Third, it reinforces a lesson that runs through much of the research on AI at work: the danger lies less in the AI itself than in how much we lean on it. Appropriate reliance means knowing when to trust the machine and when to trust yourself. Overreliance, the study suggests, may creep in precisely when we feel most at ease.
So the next time you find yourself nodding along with a chatbot’s answer because it all feels so obvious, it might be worth a second look. Comfort, it seems, is not always a sign that you are making the right call.
Reference
Das, S., & George, A. (2025). Augmented or overloaded? Exploring cognitive fit and GenAI reliance in decision making. In ICIS 2025 Proceedings
