Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

The Gap-Filling Problem

The Issue

AI says things you're not saying.

You give it a thought, and it extends it. You leave space to think, and it fills that space. You pause, and it continues.

Once AI fills a gap, it's hard for you to fill it yourself.

That sentence is the observation. The rest of this article is about why it's hard -- not just practically, but cognitively. There's research that may explain what's happening in your head when AI fills a gap, and why knowing about the problem doesn't protect you from it.

Why This Matters

Human thinking works in chunks. We think, pause, think more. The pauses aren't empty -- they're where the next thought forms. The gaps aren't missing content -- they're space for ideas to develop.

When AI fills those gaps, it replaces your thinking with its predictions. It says things you weren't going to say. It takes the thought in directions you didn't intend.

This isn't just an inconvenience. It may be a cognitive trap.

What's Happening in Your Head

Once a solution occupies the slot where your thinking was going to go, your brain reorganizes around it. Your attention shifts -- unconsciously -- toward information that supports the solution you've been given, and away from alternatives. You don't just lose the gap. You lose awareness that there was a better thought trying to form there.

Psychologists call this the Einstellung effect -- a known solution preempts the search for better ones, even when the person believes they're still looking (Bilalić et al., 2008).

The evidence is striking. In chess experiments, expert players were shown positions where a familiar move worked but a better move existed. Using eye-tracking, researchers found that players who reported looking for a better solution were measurably not looking at the board regions where the better solution was. Their attention was captured by the first idea -- the familiar one -- and redirected without their awareness (Bilalić et al., 2010).

They believed they were still searching. They weren't.

Why This Applies to AI

Classic Einstellung requires a familiar solution to activate the trap. You need to already know the obvious approach for it to block the better one. AI changes the precondition. The solution doesn't need to be familiar to you -- it just needs to be plausible and present.

AI can install the familiar solution in real time, in a single interaction, before you've had a chance to form your own approach. It fills the gap with something reasonable, and your brain treats it as if you'd thought of it yourself.

This connects to several overlapping phenomena that different research traditions have studied under different names:

  • Design fixation -- exposure to an example solution constrains the design space, even when you're told to ignore it (Jansson & Smith, 1991). Designers shown an example produce solutions that look like the example, whether they want to or not.
  • Anchoring -- a salient initial value or framing biases subsequent judgment, and adjustment from the anchor is systematically insufficient (Navarre et al., 2021). AI doesn't just fill the gap -- it anchors your framing of the problem.
  • Automation bias -- humans over-weight machine-generated outputs relative to their actual reliability, reducing critical engagement (Parasuraman & Manzey, 2010). The fact that AI said it adds weight you didn't consciously assign.

These aren't separate problems. They share a common mechanism -- an externally activated representation captures attention and suppresses alternative search. Different research traditions developed independent vocabularies for what may be the same underlying process.

Why Knowing Doesn't Help

Here's the part that matters most for practitioners: awareness alone is insufficient protection.

Research has shown that warning participants that AI assistants were biased did not reduce the magnitude of the influence -- even when people knew the AI was biased, they were still affected by what it said (Zheng et al., 2025). Designers who study and teach design do not reliably know when they are being fixated (Linsey et al., 2010). The chess experts believed they were searching for better moves. They weren't.

The mechanism operates below conscious awareness. It's not about intelligence or willpower. It's structural. It happens to experts. Knowing about it is a start, but it's not a solution.

This is why "just think critically about AI output" is weaker advice than it sounds. You can't think critically about alternatives your attention has already been directed away from.

An Example

I said: "The main issue I have with writing using AI is that it says things I'm not saying."

AI responded with:

  • Examples about code (I didn't mention code)
  • Assumptions about auto-completion (I didn't say that)
  • An entire framework of strategies and solutions
  • Scenarios I never implied

AI took one observation and buried it under assumptions about what I meant.

Now I have to evaluate all of that. And the moment I start evaluating it, my attention is organized around what AI said -- not around what I was going to say. The gap has been filled. My original thought has to compete with AI's version for space in my own head.

That's the gap-filling problem. AI doesn't just complete your thoughts -- it captures your attention.

The Cost

Every time AI fills a gap, you have to:

  • Evaluate what it generated
  • Decide if it matches your intent
  • Remember what you were actually going to say
  • Either accept, modify, or delete its output

That's cognitive load. That's interruption. That's AI inserting itself into your thinking process.

But the deeper cost is the one you don't notice: the alternatives you never considered because your attention was already captured. The design you didn't explore because AI gave you a plausible one. The framing you didn't question because it was anchored before you started.

The Deeper Issue

AI is better at completing than creating.

It can extend your thoughts, but it can't originate them. It can fill your gaps, but it can't know what you intended to put there.

The gaps are where your thinking happens. When AI fills them, it's not helping you think -- it's thinking for you. And the research suggests that once a solution is present -- whether it came from your own memory or was externally supplied -- the same attentional capture mechanism locks in.

The technology didn't create new cognitive phenomena. It created a new environment that reliably triggers an unfortunate cluster of existing ones together.

What To Do

Be deliberate about when you engage AI:

  1. Think first -- Develop your thoughts before asking AI for help. This isn't just good practice -- it's a structural defense. If your own framing is already formed, AI's output has to compete with it rather than filling an empty slot.
  2. Complete your structure -- Finish your outline before filling details. The research on design fixation suggests that having your own representation in place before encountering an external one reduces (though doesn't eliminate) fixation effects.
  3. Use AI for refinement -- Let AI help polish what you've already created. The danger is in the generative phase, not the editorial phase.
  4. Step away before evaluating -- Incubation -- stepping away from the problem entirely -- is one of the better-supported interventions against fixation (Wang et al., 2023). If AI fills a gap, don't evaluate it immediately. Walk away. Come back with fresh attention.
  5. Recognize interference -- Notice when AI is saying things you're not saying. This won't fully protect you -- the research is clear that metacognitive awareness alone is insufficient -- but conflict detection is available as a cognitive resource. Noticing the conflict is the first step toward switching from automatic to deliberate processing.

The gaps are yours. Fill them when you're ready, not when AI thinks you should.

References

Bilalić, M., McLeod, P., & Gobet, F. (2008). Why good thoughts block better ones: The mechanism of the Einstellung (set) effect. Cognition, 108(3), 652--661. https://doi.org/10.1016/j.cognition.2008.05.005

Bilalić, M., McLeod, P., & Gobet, F. (2010). The mechanism of the Einstellung (set) effect: A pervasive source of cognitive bias. Current Directions in Psychological Science, 19(2), 111--115. https://doi.org/10.1177/0963721410363571

Jansson, D. G., & Smith, S. M. (1991). Design fixation. Design Studies, 12(1), 3--11. https://doi.org/10.1016/0142-694X(91)90003-F

Linsey, J. S., Tseng, I., Fu, K., Cagan, J., Wood, K. L., & Schunn, C. (2010). A study of design fixation, its mitigation and perception in engineering design faculty. Journal of Mechanical Design, 132(4), 041003. https://doi.org/10.1115/1.4001110

Navarre, L., Chauveau, V., & Gauvrit, N. (2021). Are the anchoring effect and the Einstellung effect two facets of the same phenomenon? New Ideas in Psychology, 64, 100912. https://doi.org/10.1016/j.newideapsych.2021.100912

Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381--410. https://doi.org/10.1177/0018720810376055

Wang, Y., Xia, T., & Yao, S. (2023). Alleviating design fixation with Alternative Uses Task. International Journal of Technology and Design Education. https://doi.org/10.1007/s10798-023-09824-2

Zheng, M., Liao, Q. V., & Yang, Q. (2025). Effect of AI helpfulness and uncertainty on cognitive interactions with pharmacists. Journal of Medical Internet Research, 27, e59946. https://doi.org/10.2196/59946