AI Design Fixation: Look Beyond Variations of One Idea

Suppose a creator receives twelve AI-generated poster proposals for a pottery workshop. Each shows a bowl in the centre, with different colours, lighting and decorative details. The images may be attractive. Yet the review contains only one basic proposition: look at this beautiful object. Twelve files have become twelve versions of the same invitation.

That fictional situation raises a useful psychological question about AI-assisted creativity. How much conceptual ground have we explored before choosing a direction? A large batch of images can make that question harder to notice. For creators and business owners, judging the range of ideas deserves a separate moment from judging their finish.

What design fixation means here

The term predates generative AI. David Jansson and Steven Smith’s 1991 paper on design fixation investigated how restrictive adherence to existing ideas can limit conceptual design. The useful distinction is between a deliberate constraint and a possibility that has gone unexamined.

A pottery poster may need the studio’s established colours and a readable workshop title. Those requirements do not automatically require a solitary bowl, a particular viewpoint or a luxury-product mood. In the fictional brief, the bowl has quietly become a rule even though nobody agreed it must be.

What an AI-image experiment actually tested

In a study published at CHI on 11 May 2024, Samangi Wadinambiarachchi and colleagues randomly assigned 60 participants with visual-design experience to three groups. Everyone received an example chatbot avatar and had 20 minutes to sketch ideas. Additional inspiration support was either absent, Google Images or Midjourney V4. The experiment took place in May 2023.

The AI-supported group showed more fixation and lower fluency, variety and originality than the baseline. Fixation was measured through features repeated from the example; fluency counted participants’ sketches, not AI-generated pictures. The authors note the short task duration, limited professional industry experience in the sample and the time spent seeking inspiration. These findings concern that particular workflow and setting; they do not establish a permanent loss of creativity or evaluate every current image model.

A better piece and a broader collection are separate outcomes

A different result helps keep the interpretation proportionate. In research published in Science Advances on 12 July 2024, Anil Doshi and Oliver Hauser studied eight-sentence stories. Writers had no AI idea assistance, access to one idea or access to up to five ideas. Access to AI ideas improved readers’ evaluations of the stories, while AI-assisted stories were more similar to one another.

That study concerns short written fiction, so it cannot settle a visual-design question. It does illustrate why reviewing one finished piece and reviewing the diversity of a collection can produce different assessments. For the pottery creator, the strongest individual bowl image might be useful even while the overall set leaves other possible messages unexplored.

Name the idea beneath each image

Before discussing shadows, texture or colour grading, describe each proposal in one plain sentence. What does it invite the audience to imagine doing? For the fictional workshop, three directions could be:

• Make something together: the invitation centres on shared participation and learning with other people.
• See a transformation: the invitation follows clay becoming a usable object through stages of making.
• Discover the tools: the invitation introduces unfamiliar materials and tools as something beginners can explore.

These are proposed directions, not research-tested remedies for fixation. They also need checking against the real offer: a social message would mislead if participants work separately without that shared experience. A wide-ranging concept still needs to describe what the workshop actually provides.

Keep the underlying message visible during review. A warm portrait, a dramatic close-up and a flat illustration could all serve the same participation concept. Conversely, visually restrained images might communicate very different reasons to attend. The concept sentence lets the team discuss meaning alongside appearance.

Separate fixed requirements from inherited choices

Return to the brief and mark which decisions are genuinely fixed. For this example, the audience, session format and essential booking information belong in that group. The central object, visual metaphor and framing remain available for exploration unless the creator has a reason to constrain them.

MaryChuks.com’s guide to turning a mood board into a creative brief explains how to translate taste into usable direction. Here, the practical extension is to ask what each reference contributes. Perhaps the team wants its generous space or tactile materials, while leaving the subject and arrangement open.

When requesting further AI exploration, specify the conceptual dimension to change. For example: develop the transformation direction around the stages of making, retaining the agreed information and palette. That request gives the next output a clear purpose. Afterwards, judge whether the result actually changed the invitation’s idea, rather than counting the new file as evidence that it did.

Review the set before committing

Place the concept sentences beside the proposals. Group together those that express the same proposition, then look at which directions have serious representation. The purpose is to reveal choices still available. It is not to invent a precise creativity score from a small commercial project.

Give an unfinished direction a fair description before rejecting it. A rough process sketch can communicate a promising invitation even when a polished bowl image is more immediately pleasing. You could bring shortlisted concepts to a similar level of finish before making the final comparison, while keeping the actual audience and offer as the decision criteria.

Then commit to the chosen direction and record why it fits. Exploration does not need to continue indefinitely. A note such as “selected the making-process concept because the workshop introduces beginners to each stage” preserves a reason that can guide later revisions.

This is an editorial review routine suggested by the questions the research raises, not a proven intervention tested in those experiments. Its value can be assessed in your own work: did the review expose a meaningful option, improve the brief or help explain the final choice? AI-assisted creativity becomes easier to direct when the team can name what it explored and why it stopped.

Sources

• Jansson and Smith: Design fixation — Design Studies, January 1991.
• Wadinambiarachchi and colleagues: The Effects of Generative AI on Design Fixation and Divergent Thinking — CHI, 11 May 2024; author manuscript.
• Doshi and Hauser: Generative AI enhances individual creativity but reduces the collective diversity of novel content — Science Advances, 12 July 2024.

Sources checked on 25 October 2026. The pottery workshop, poster proposals and review examples are hypothetical.