Human Versus AI: Who Styled It Better?
- A reporter tested AI stylists and a human stylist to find an outfit for a gala, with the human stylist providing a more personalized and successful experience.
If you’ve read my work before, you may know I cover technology —these days, that involves a lot of artificial intelligence .
But as AI continues to proliferate, it’s coming closer and closer to the consumer. While I’m a reporter, I’m also a consumer. I know a lot about technology, but sometimes I feel I can hardly work my iPhone, despite being a Gen Zer.
More from Sourcing Journal
And now that we’re hearing more about AI stylists and the rise of agentic AI —which, for consumers, comes along with the promise that an autonomous agent could, in the near future, shop for you—I wanted to test out a few tools to understand the hype through a different lens.
That in mind, I decided to work with a personal stylist and compared her work to that of two AI “stylists.” For the test I enlisted ChatGPT and ThredUp’s Style Chat function. I do want to caveat my choice to test ChatGPT; while I know that, today, its primary function isn’t shopping, nor styling, it remains one of the most well-known AI tools on the market, and it’s the system that brought generative AI into the the zeitgeist in a very real way.
Let’s start from the beginning. My friends, Max and Will Grove, who are involved with the Lesbian, Gay, Bisexual & Transgender Community Center of New York, invited me to join them at their table at the organization’s annual fundraising gala. I happily accepted, with just one issue—I had nothing to wear.
I needed one formal look, complete with a bit of playfulness, but still sleek. Rather than doing all the leg work on selecting an outfit by myself, I enlisted some help. That’s where my stylists—AI and human—came in.
Working with a human stylist
When I met Phylicia B. Alexander, the (human!) stylist I worked with for four of the six looks included in this piece, I recall telling her, “I’m just a regular person,” feeling a bit out of place. I had never worked with a personal stylist before. After all, I’m a 25-year-old, New York City-based reporter; personal styling doesn’t exactly fit into my typical monthly budget (much to my chagrin, after this experiment).
Prior to our first meeting, I had shared the link to further details about the event, and we spent our first face-to-face session (via Zoom) discussing what I envisioned myself wearing and the frustrations I’ve previously had with shopping for clothing. At one point, she asked me how I felt about myself, on a scale from one to 10. I answered honestly, and told her that I have, at times, struggled with my self image, particularly because of the proportions of my body. Telling someone I just met about that, in granular detail, felt daunting.
Alexander handled it with grace and deep professionalism.
“I want you to know you’re absolutely not alone on that,” she told me. “I’ve worked with women of all different body types, sizes, all of that—and I honestly hear that exact same thing.”
After our first meeting, Alexander, who specializes in styling for events and occasions, asked me to fill out a questionnaire, which asked about my budget for the look ($250), descriptors of my personal style, preferences around color and more.
The form also asked, “If you get a compliment at your event, what are three to five words you’d want them to use to describe your outfit?” My response was, “sleek, unique, timeless/classy.”
To finish out this part of the process, I needed to submit my measurements: bust, waist, hip and shoulders.
After I filled out that form, I scheduled another call with Alexander, at which point she showed me inspiration photos from several different Pinterest boards; I told her specific details I liked about certain photos she had selected, which helped her better understand my preferences. On that call, I remember mentioning how much I like mesh sleeves on a dress…More to come on that.
A few days after our call, Alexander shared a lookbook with me via email, comprising about two dozen dresses—some paired with shoes or accessories—and asked me to “like” the looks I could see myself wearing. I chose a few styles on the first go-around, and once I had selected my favorites, Alexander and I scheduled a third call to discuss why I liked certain items or what I might want to see added to the lookbook.
I told her that, while I loved a few of the looks, others felt a little bit older than I might want to wear. She welcomed the feedback and said she would provide a few additional looks for me to check out based on what I’d told her; within hours, she had done just that. We selected the final looks I would order to try on for our in-person fitting, and—importantly—Alexander recommended which sizes I should order in each item to set me up for success, based on the measurements I had previously submitted to her.
She and I texted a fair bit throughout this process, and communicating with her felt like asking a (very skilled) friend for their opinions on what to wear.
We settled on ordering four dresses. At the outset of our work together, Alexander reminded me that she might pull some looks I wouldn’t necessarily choose for myself, but that I should embrace that and try a few new things. The brown dress and the sparkly green dress in the below photos are indicative of the fact that I took her advice—when Alexander originally asked me to describe my existing style, I used words like “expressive, easy going, casual and classic.”
The brown dress, in particular, is something I would never pull for myself because of fit concerns in the bust. But Alexander knew it’d be flattering. In the end, I almost chose to wear it (the reason I didn’t was because I knew I’d be unlikely to have a reason to wear it repeatedly).
The sparkly green dress was the only dress that didn’t fit at all; it was about two sizes too big, and we had to pin back handfuls of the material for the photos. In the end, I think it would have been too difficult to alter, and even after it was pinned, I didn’t love the shape against my silhouette.
Ultimately, Alexander agreed with that sentiment, so we decided to focus on the other three looks—from the beginning, she told me that ordering multiple looks would safeguard me from any fit issues and from any apprehension about wearing a look I didn’t love.
These kinds of discussions added a personal touch that I simply didn’t find in my experience working with AI-based stylists; Alexander came to my office for the final try-on and styling experience, where the photos you see throughout this story were taken. She brought with her a variety of extra jewelry and accessories, which made the experience feel even more exciting.
You’ve already seen two of the looks Alexander put together for me—here are the other two (with a photo of her helping me style the black dress for best photo results!). Remember how I said I loved mesh sleeves?
ThredUp Style Chat
ThredUp has a tool it calls Style Chat, which allows consumers— in this case, me—to enter a natural-language prompt about an upcoming event or style need. From there, ThredUp’s AI tool offers up categorized suggestions on what to wear, as well as an AI-generated preview of what such an outfit would look like on a person.
I told Style Chat I needed an outfit for a New York City-based gala in April, and that I wanted the final look to be sleek and youthful—the same adjectives I had used with Alexander when I filled out the questionnaire. Based on my prompt, ThredUp recommended I wear an emerald-green dress, tan patent leather heels, a black hat, a silver belt and a silver clutch. Here’s what Style Chat’s vision of those pieces looks like all together:
And here’s how I looked wearing Style Chat’s selections:
I didn’t love the resulting outfit—mostly, the difference in colors felt stark, particularly in comparison to Alexander’s color-matching skills. Tan, with black, with green, with silver made the look feel a little disjointed. It felt clunky, rather than sleek, like I had requested.
I asked three other stylists to rank the looks (without knowing who or what styled them), one through six (one being their favorite, six being their least favorite). Each of them ranked this look fourth or fifth.
Nikki Venus, a business style coach, shared my sentiment about too much mixing and matching.
“While the dress has potential, the styling choices made it feel disconnected. The oversized floppy hat, which leans more vacation-chic, didn’t complement the formal vibe of the gown—it felt out of place,” she said. “On top of that, the metallic belt didn’t do the silhouette any favors; instead of elevating the look, it created a distraction. With more intentional accessories and a cohesive style direction, this dress could really shine.”
Though the mismatched nature of this look wasn’t a favorite, what I’d say about this tool is that it’s a helpful starting point and one of the more accessible tools available to the average, mass-market consumer. That’s because it provides multiple options for each individual item it recommends (and only in sizes that will fit, if you provide details about your preferred sizes, which I did). Here’s an example of what the recommendations looked like:
As compared with ChatGPT, which provided lots of broken or misdirected links (and didn’t do any favors with trying to help out with sizing), the way the options Style Chat gave were set up felt helpful—even if the options themselves didn’t suit my own style.
In stacking ThredUp’s tool against Alexander’s expertise, the most noticeable gaps were the fact that Style Chat had no sense of my preferences, while Alexander did because of the time she spent gathering information from me. She also had a discerning eye for the way colors should pair with one another, which, in this case, the AI stylist lacked.
ChatGPT
Working with ChatGPT was, candidly, one of the more frustrating pieces of working on this story (other than using Amazon’s Rufus, but I’ll get to that later).
I gave it the following prompt: “Act as if you are my stylist. I am attending a formal gala in New York City on April 10 and need a look that revolves around a long dress. I want it to be sleek, youthful and fun. My waist is much smaller than my hips, and I have a large bust, so I don’t like wearing strapless very much. I want to buy only items I can return.”
My main problem with ChatGPT was that, when it offered up a specific style name, it couldn’t source a direct link to that item for me. And, in many cases, the brand no longer sold the style it had recommended. That led to a lot of Google searching, trying to find the closest match to what it had suggested.
You can see here that, when I asked my “stylist” to share the sources for the dresses it had recommended, it struggled to point me to a single product page, instead sending me to sites’ general dress landings or admitting it couldn’t share where its suggestion came from.
The Abercrombie & Fitch dress ChatGPT selected was available, and in response to my original query, the system recommended I wear the black or green iteration of the dress. I couldn’t find the dress in black, so I ordered it in green. Here’s how it looked:
Though I have had some success using ChatGPT for other, non-fashion-related queries, its efforts in styling me for this event fell flat among the three stylists who ranked the final looks. They each ranked this look as their least favorite.
While I think ChatGPT’s final look (which included gold jewelry and gold shoes) felt more cohesive than ThredUp’s, it didn’t fit my body well—and also felt a bit too informal for the occasion. As style curator Aleisha Bradley put it, “it’s a trendy, go-to dress that doesn’t represent her style and personality. It’s not flattering on her body and lacks luster.”
One thing I did appreciate about ChatGPT was that—like Alexander—it recommended specific hairstyles I could’ve worn with the pieces I selected.
Other experiments
In the process of finding an outfit that would work well for the gala, I tried several other AI-powered tools, with varying degrees of success.
Alta, which just inked a partnership with the Council of Fashion Designers of America ( CFDA ), felt like a strong contender for an AI-based personal stylist. The outfits it suggested to me felt cohesive, even if, at times, not quite my style. The trouble I ran into was that the tool had a difficult time respecting my budget.
Amazon Rufus, meanwhile, was extremely difficult to work with in any cohesive way. It suggested items that weren’t in stock, had a hard time adhering to budgetary requirements and suggested dresses that were less “sleek” and more “mother of the bride.”
I felt I got more out of the “styling ideas” tab on individual Amazon product pages than from Rufus itself.
I would note that this experiment took place prior to the release of Amazon’s Nova Act, which supposedly has more agentic capabilities; Rufus is a chatbot meant to field natural-language questions. It’s certainly better suited to handle a question like, “What should I purchase for a Super Bowl party?” than it is to style someone for an important event.
Conclusions
I think it’s clear that Alexander’s expertise far supersedes that of the current AI stylists I worked with. Each time she and I ran into an issue (out-of-stock undergarments, pieces I wasn’t fond of, a dress that didn’t fit correctly), she had a solution or suggestion.
She had a keen understanding of the colors that would suit me best, and had a vision for how certain looks would compliment my figure. A perfect example of that: I thought the dress ChatGPT selected would look great, but it ultimately wasn’t a match for my body type; meanwhile, I thought the brown dress Alexander selected might be a tough look for me, but it ended up fitting my silhouette quite nicely.
In addition to that, Alexander sourced items from several sites or brands I’d never heard of (and thus would have been less likely to come across on my own). The dress I ultimately chose came from a brand I’d never shopped with before. ChatGPT, meanwhile, selected items from mass-market brands I’ve previously shopped at or viewed—it felt like I could have come to a similar conclusion myself with less frustration involved.
As much as I thought AI styling could be a “quick fix,” neither system selected a look that was viable or appropriate for the formality level of the event I attended. And, unlike working with a human, asking AI to alter its preliminary results yielded seemingly random suggestions, not tailored to my own style.
What I’d also note is that the process of working with Alexander positively influenced other pieces of this experiment; I had never properly measured myself, so while she used those measurements to help me select sizes for the pieces she recommended, I also used those measurements to ensure I picked the best size for the garments the AI systems recommended. Similarly, Alexander asking about my goals for the look helped me better define what I sought when working with the AI models.
What I wore
In the days following my final styling session with Alexander, it came time to make the final choice, which proved a difficult decision because I loved three of the four looks she had styled. Alexander checked in with me to ask what I’d selected and offer up any last-minute accessory ideas I might need.
I ultimately selected the look I felt most confident in—a look I knew I’d wear more than once: the black, mesh-sleeve dress.
Coincidentally, that dress was also the favorite among the stylists who ranked the looks; it came in first with all three. Tami Harrigan, a style coach who helped rank the looks, explained why the black dress was her favorite—and I agree.
“The dress fits her body perfectly, it’s flattering, and she looks comfortable and confident in it,” Harrigan said. “The style of the dress is classic, while incorporating modern/trending elements with the sheer over the arms. It also compliments her body and style.”
