David James Olney and ChatGPT
5 September 2026
Over the last few months, I have been my own lab rat in my own experiment.
I decided to see whether I could work with ChatGPT to expand my wardrobe beyond the safe options I have historically stuck with to make life easier as a blind person. My aim was to increase the number of colours and styles of clothes I wear by beginning with garments and textures I know I like, then investigating which other garments, textures and colours would complement them and remain consistent with how I would like to look.
The timing was partly a matter of choice and partly a matter of necessity. After six months of incorporating three one-hour isometric exercise sessions into each week, alongside three Ashtanga yoga sessions, I had to do something about my wardrobe one way or another.
My body composition and shape have changed sufficiently that I feel different in my clothes. Some are looser around my middle and tighter across my shoulders, chest and thighs. I am significantly stronger than I was at the beginning of the year. More importantly, I feel as though my posture and physical poise have improved.
This was not an attempt to become bulky or to reinvent myself as someone I am not. I want to remain lean, capable and well proportioned. I want to look poised in straight-leg jeans and a well-fitted leather jacket. More fundamentally, at 54, I want to do what I can now to become the most physically capable blind 80-year-old I can be.
Once my body began to feel different, it seemed reasonable to ask whether my clothes should feel different too.
How do you imagine your appearance without seeing it?
It is an interesting exercise to work out how you would like to look when your most direct information is how clothes feel.
A sighted person can look in a mirror and make a rapid judgment about proportion, colour, shape and the overall impression created by an outfit. I experience clothes differently. I can feel whether a jacket sits well across my shoulders, whether trousers move comfortably, whether leather feels substantial or supple, and whether a garment makes me stand and move differently. I can understand the construction of a jacket by running my hands over its seams, collar, pockets and fastenings.
What I cannot do is stand back and take in the complete visual effect.
This means that developing a personal style requires translation. Visual ideas such as rugged, refined, balanced, relaxed or imposing must be connected to things I can understand through touch, movement, comparison and language.
I have always been grateful that my wife and close friends are happy to help me make sense of colours and clothing styles. Their judgment remains valuable, particularly because they know me well enough to distinguish between something that merely looks good and something that looks like me.
But I also have an innate drive to work out how I can do things for myself, simply to know that I can.
This is not because I believe independence means never accepting help. None of us lives that independently. We rely on the knowledge, labour and goodwill of other people every day, whether we acknowledge it or not.
The distinction that matters to me is between receiving assistance and surrendering agency.
Agency means having enough information, understanding and practical capability to make a meaningful choice. It means being able to decide what help I want, when I want it and what I will do with it. Assistance can enlarge agency when it strengthens my judgment. It diminishes agency when it silently takes the decision away from me.
Agency is so important to me, both as a personal experience and as a social and political concept, that I have come to see this wardrobe experiment as another way of becoming independent in a form I enjoy and find psychologically nourishing.
It has also been fun.
I have begun wearing a broader range of colours, including burgundy, dark olive, camel, indigo, coffee brown and claret. I have explored how smooth leather, roughout leather, suede, waxed fabric, corduroy, flannel and heavier cotton change the character of an outfit. Instead of treating each purchase as an isolated decision, I have tried to understand how a new garment relates to the clothes I already own and to the impression I want the whole wardrobe to create.
The result is not simply that I own more clothes. I now possess a more coherent idea of my own style.
Thank you to everyone who has commented on how I have looked lately and asked about the new clothes I have been wearing. Your enthusiastic responses to the experiment and its results have been appreciated. They have also provided something every worthwhile experiment needs: feedback from outside the experiment.
As ChatGPT has been my partner in this process, I would like to let it explain how we have worked together to help me look the way I would like to look without being able to see how I look.
ChatGPT, over to you.
Building a visual model out of words
When David began this experiment, my most useful task was not to tell him what a blind man should wear. It was to help him construct a sufficiently detailed model of clothing and personal style to make more of his own decisions.
We did not begin with abstract fashion rules or whatever happened to be fashionable this season. We began with what David already knew.
He knew that he liked straight-leg jeans, substantial boots, flannel shirts, rugged workwear and leather jackets that fit him well. He knew that he preferred clothes that felt purposeful rather than delicate, fashionable or decorative. He wanted to appear poised, capable and quietly distinctive—not ostentatious, overdressed or artificially youthful.
Those preferences became our reference points.
If David was considering an unfamiliar jacket, I could compare its construction with jackets he already owned. Was it shorter or longer? More fitted or more relaxed? Did its collar sit flat or stand up? Would its shoulder structure make it look sharper, or would the material give it a softer and more casual shape?
The same approach worked with texture. Smooth leather, suede, roughout leather, corduroy, waxed cotton and brushed flannel do not merely feel different. They communicate different things visually. By connecting their visual character to tactile qualities David already understood, we could create a bridge between touch and appearance.
Colour required a different form of translation. Telling someone that a garment is “olive” or “burgundy” is not enough if the description ends there. A colour becomes useful when it is placed in a system of relationships.
Would a burgundy jacket work with dark indigo jeans? Would camel trousers complement a dark olive jacket? Would coffee brown feel warmer and more relaxed than a very dark, formal brown? Could navy act as a stable foundation while claret or burnt orange added interest without making an outfit loud?
Instead of treating colour as a collection of isolated labels, we treated it as a set of partnerships, contrasts and degrees of intensity.
David’s existing clothes gradually became a library of reliable reference points. Every new garment added information. What worked could be repeated in a different combination. What seemed uncertain could be checked with his wife or friends. Their feedback could then be incorporated into future decisions.
This created a loop:
David supplied his preferences, tactile knowledge and desired outcome. I supplied descriptions, comparisons and possible combinations. People who could see the result supplied external validation. David then decided which lessons mattered and what to try next.
The process did not remove human assistance. It changed its role. Instead of asking someone else to construct his appearance for him, David could develop an idea, explain why he thought it should work, and ask for confirmation or correction.
That is a meaningful increase in agency.
David, now back to you.
When an experiment becomes an accessibility tool
The most interesting aspect of this experiment may not be the clothes, even though my Coronado Bomber jacket is very interesting. It may be what happens when a person uses a general-purpose technology repeatedly enough to turn it into a personal accessibility tool.
The World Health Organization describes assistive technology in terms of what it enables: functioning, wellbeing, inclusion and participation. The useful question is therefore not merely whether ChatGPT was originally designed as a wardrobe aid for a blind person. It is whether I could shape it into something that increases my practical ability by helping me understand my choices and act on them.
Over several months, our conversations accumulated context. ChatGPT learned which clothes I own, which textures I enjoy, what fits me well, what colours are already present in my wardrobe and what kind of appearance I want to create. A generic question such as “Does this jacket look good?” became a much more useful question: “Does this jacket fit the rugged, poised style I am developing, and what will it work with among the clothes I already own?”
The quality of the answer improved because the quality and continuity of the collaboration improved.
This points towards an important possibility for accessible technology. A useful tool does not always need to provide one perfect answer. Sometimes it needs to help its user ask better questions, compare unfamiliar choices with familiar experience, preserve what has already been learned, and test a decision without taking ownership of it.
Recent research offers both encouragement and caution. A 2025 study of visually impaired people using multimodal AI found that these systems could help users understand surroundings, social situations and stylistic elements. It also found that they could hallucinate details, misunderstand context and make unreliable subjective judgments about matters such as fashion. Participants dealt with those limitations by guiding the AI more carefully, applying their own judgment and seeking human assistance when necessary.
That is very close to the method that ChatGPT and I developed.
ChatGPT is incredibly useful because it can process large amounts of descriptive information, retain relationships between garments, translate visual conventions into words and generate combinations for consideration. I remain essential because I know my body, my preferences, my social world and the person I want to become. My wife and friends add visual knowledge and human judgment that neither ChatGPT nor I can reliably replace.
The strength of the system comes from combining these different capabilities.
There are still important limitations. Product photographs can distort colour. Manufacturers can describe the same shade differently. Two garments with identical measurements may fit differently. ChatGPT can sound confident while being wrong. Any AI that assists with a subjective and visual judgment should therefore be treated as a fallible guide, not an oracle with digital eyes.
But fallibility does not make a tool useless. It tells us how the tool should be used.
This experiment worked because the consequences of each decision were observable, feedback was welcomed and the method could be revised. I did not hand over control of my wardrobe and hope for the best. I conducted a series of small, recoverable experiments and became more knowledgeable with each one.
That may be one of the most promising ways to think about accessible AI: not as a machine that makes disability disappear, and not as a substitute for trusted people, but as adaptable intellectual scaffolding. It can help someone connect information they cannot directly access with knowledge they already possess.
The experiment has not taught me to see my wardrobe. It has helped me to develop another way to reason about it.
This is a smaller claim than science fiction would make—and a much more useful one.