You get an email that bothers you. Nothing in it is rude, but something feels off, and before you've decided what to make of it you paste it into an AI tool and ask how it reads the tone. A few seconds later you have an answer. The writer sounds defensive, it says, and it points to a sentence you had barely noticed. So you read the email again, and now you can't read it any other way.
The tool may be right. It may have caught something you would have missed. But something else happened too. Before you had made up your own mind, you were handed an opinion, and more and more, the first opinion we're handed comes from AI.
A Small Change in the Order of Things
I use AI all the time, and one of the things I like most about it is how fast it can give me another perspective. When I've been staring at a problem too long, it can point out what I'm not seeing, question an assumption, or offer an option I hadn't considered. There are plenty of times I ask for its view precisely because I know mine is incomplete.
What I've started noticing is when I ask. "Here's what I think. What am I missing?" is a different request from "What should I think about this?" In the first, AI joins a conversation I've already started. In the second, it gets the first word, and going first matters more than it seems.
Proverbs 18:17 says, "The one who states his case first seems right, until the other comes and examines him." I've usually read that as a reminder to hear both sides, and it is. But it also captures something recognizable about how judgment works. The first account we hear can have an advantage. It directs our attention toward certain things and away from others. Once a problem has been framed for us, stepping outside that frame takes effort.
The First Answer Does More Than Answer
Picture a student who opens a business case and immediately asks AI what the main problem is. The answer might be excellent. Maybe the company really does have a pricing problem, or maybe the CEO really is underestimating a competitor. But from that moment on, the student isn't looking at an open question anymore. The pricing data look more important than they did a minute ago. The competitor gets more attention. The evidence for one recommendation starts to stand out.
She hasn't followed the AI blindly. She may disagree with half of what it said. Still, the AI helped decide what she's disagreeing about. We usually think of AI influencing us when we accept its answer. But it can influence us earlier than that, by shaping what we think the question is. The same thing happens to a manager writing a performance review, to an entrepreneur weighing a new market, and to any of us trying to make sense of a hard decision with a chat window open.
Maybe AI Should Sometimes Go Second
My coauthor, John Delano, and I have been working on a framework for helping students keep their judgment while using AI, and one idea we keep coming back to is simple: form some kind of initial judgment before asking AI what it thinks. It doesn't have to be polished. "I think this is the main problem, but I'm not sure." "These two things seem to matter most." "My instinct is option B, and here's the assumption I might be making." Then bring in AI.
That's when the interesting part starts. AI can tell me I missed an important stakeholder, or show me that my assumption doesn't hold, or make the case for option C. It can tell me the thing I called the problem is only a symptom. I want that kind of help. And because I started somewhere, I can see my own thinking move. I know where I began, I know what changed my mind, and I can ask whether it should have. Without that first thought, it's surprisingly hard to know how much of the final judgment was ever mine.
Of course, going first has its own danger. We can get anchored to ourselves. Anyone who has defended a bad idea longer than necessary knows that people don't need AI to fall in love with their first opinion. So I don't think first in order to keep AI from influencing me. If it has better information, catches my blind spot, or makes a stronger argument, I want to change my mind. Thinking first gives me something to compare with what comes back. If AI changes my mind, I can see what changed it. That may be part of why the proverb resonates here. The first account is not necessarily rejected; it is examined. Our first thought should be open to the same treatment.
What the Research Suggests
A 2026 experiment with 273 business professionals found something more complicated than simply “AI makes us more biased” or “AI makes us less biased.” People who used AI at an intermediate level showed stronger cognitive processing and less anchoring, confirmation bias, and overconfidence. Outcomes were weaker at both low and high levels of use. The amount of AI use mattered, but so did what people actually did with what AI gave them.
Everett et al. (2026) looked at the order question directly in a randomized trial with physicians working through difficult diagnostic cases. One group saw AI's assessment first; another formed an initial assessment before consulting AI. Both AI-supported groups performed better than physicians using conventional resources alone, and overall performance did not differ significantly depending on who went first. In fact, AI-first performed better on some clinically actionable decisions. The researchers also found signs of anchoring in both directions: clinicians could be influenced by AI when it went first, while AI could be influenced by the clinician's earlier judgment when it went second. That makes the question more interesting. There may not be one right order for every task. But order changes the interaction, and it is worth noticing what each sequence makes easier to see and easier to miss.
Taken together, the studies point in a less tidy direction than “AI first” or “human first.” AI can improve performance in either position. For me, the takeaway is that there still needs to be room to form a judgment rather than simply inherit one.
A Small Habit Worth Keeping
I suspect one of the most useful AI habits will turn out to be almost embarrassingly simple. Before I ask AI about something that matters, I try to decide what I already think about it, even if all I can say is which way I'm leaning and what I'm unsure of. It rarely takes long. Then I ask. What comes back may confirm what I thought, sharpen it, or show me I was wrong, and any of those is useful, because now I can compare what came back with something I had already thought through for myself.
References
Cristofaro, M., Giardino, P. L., & Augier, M.-S. (2026). Generative AI and cognitive biases in strategic decision-making. Strategic Organization. https://doi.org/10.1177/14761270261457350
Everett, S. S., Bunning, B. J., Jain, P., Lopez, I., Agarwal, A., Desai, M., Gallo, R., Goh, E., Kadiyala, V. B., Kanjee, Z., Koshy, J. M., Olson, A., Rodman, A., Schulman, K., Strong, E., Chen, J. H., & Horvitz, E. (2026). From tool to teammate in a randomized controlled trial of clinician-AI collaborative workflows for diagnosis. npj Digital Medicine, 9, 409. https://doi.org/10.1038/s41746-026-02545-1