Why ChatGPT sucks at lighting design - and what to do instead.

Ask ChatGPT to help place lights in a room and it will tell you something that sounds entirely reasonable.

"Place recessed lights evenly across the ceiling, 120 to 150cm apart."

That answer isn't wrong in the way that bad advice is usually wrong. It doesn't tell you to wire anything incorrectly or choose a fixture that won't fit. It sounds like something a competent person might say.

And that's the problem.

It ignores ceiling height. It ignores beam angle. It ignores furniture layout, wall finishes, glare position, and how the space actually gets used.

The result can be a room that is technically lit and genuinely uncomfortable: flat ambient light with no depth, no hierarchy, no sense of where to look. The confidence of its answer is its most dangerous feature.

Where ChatGPT can be genuinely useful

This isn't an argument that ChatGPT has no place in a lighting workflow; It does.

It can explain the difference between lux and lumens. It can compare fixture types, organize a design brief, summarize product specifications, and help non-specialists understand basic concepts.

If you need to explain to a client why 2700K reads warmer than 4000K, ChatGPT will write a decent explanation in seconds. It is also reasonable for style-matching: whether a given fixture belongs in a Japandi interior versus an industrial one, for example.

The risk is in what happens next. ChatGPT gives you a fixture recommendation or a spacing rule with the same tone it uses when explaining a concept.

It does not signal the difference between describing and deciding. And incomplete lighting decisions can look, on paper, exactly like confident ones.

I’ve talked about how to tell if a lighting proposal is good in a different article.

The gap ChatGPT cannot close

There's a line in Good Will Hunting about the Sistine Chapel. Reading about it, you can describe the ceiling: the iconography, the dimensions, the history. But you cannot describe the experience of standing under it.

Lighting is structurally similar.

ChatGPT can tell you about lux levels, beam angles, and color temperatures. What it cannot do is understand how glare feels when someone is sitting on a specific sofa in a specific room, how afternoon light through a west-facing window reads against a matte plaster wall, or how the same space should feel different when you enter it at night.

These are not preferences.

They are spatial judgments, informed by room geometry, material finishes, furniture position, and how all of those things interact with the physics of light. Experienced lighting designers don't apply rules. They read a space, anticipate conflicts, and resolve them in sequence. They know that downlight output affects the case for a wall wash, which affects the decorative spec, which shifts the budget allocation across layers.

General-purpose language models are not equipped for this. Not because they lack data, but they have an enormous amount of it. But because the judgment involved is spatial, physical, and site-specific in ways that language alone cannot contain.

My recommendation: Use it for questions, not decisions

ChatGPT is useful at the research stage, not the design stage.

It is a reasonable tool for explaining concepts, drafting a checklist, comparing generic fixture categories, or helping a client understand what a layered lighting approach involves. Use it to frame the problem. Do not use it to solve it.

Before deciding on fixture locations, output levels, beam angles, or color temperature, you need the actual plan, the ceiling condition, the furniture layout, and the photometric data for the specific products you are specifying. A general answer built on general principles cannot replace that.

The better workflow, with or without any software, is to start with the space, define the lighting intent, then choose fixtures based on real constraints. Not the other way around.

This is the problem CHUBIC was built to address: lighting-specific AI that works from the plan, not from generic rules, and that understands the spatial, physical, and product-level relationships that general-purpose models cannot.

If you're planning a project and want lighting recommendations based on your actual floor plan - not generic rules, start a project in CHUBIC or get in touch with me to learn more.

- Jimmy Chu is a lighting designer with 15 years of experience leading AECOLIGHT, an award-winning lighting design agency. He is the founder of CHUBIC, which helps architects and interior designers make confident lighting decisions that better match their original design intent.