We tried to buy sweaters with AI. Here’s how it went

What I wanted to buy: A white sweater suitable for work or a formal dinner for under $100. Bonus points if it was made in Canada and from natural fibres.

What I bought on AI’s recommendation: Dynamite Balloon Sleeve Sweater in Java Brown – $59.95 + tax.

What the interaction with AI was like: In the bot’s defence, sweaters are tricky. If you don’t end up with scratchy, straw-like fabric, it’s pilling after one wash or shrinking into a Merino wool tea cozy. I wanted something durable, appropriate for work and a prewedding dinner with family. The sweater needed to be white (but not sheer); preferably cotton or another natural fibre.

I started by sizing up ChatGPT against Gemini. My first prompt: “Help me find a white sweater to wear for work and a formal dinner under $100. Preferably made in Canada.” This was enough for an elimination round. Gemini’s results were clogged with exhaustive comparison charts without a product image in sight. A breakdown of this sort might be helpful when choosing a kitchen appliance or TV, not so much for apparel.

ChatGPT served up photos but also indulged in product descriptions heavy on marketing jargon rather than consumer-led reviews. A Heritage Knit Crewneck Sweater from Aritzia was described as “refined” with a “timeless silhouette.” ChatGPT said little about fit or comfort.

I settled for “Canadian retailer” over “Made in Canada” as options within my budget shrunk the more I refined my search by size (small) and fit (cinched waist).

The bubble-sleeve knit from Quebec-based Dynamite fit the bill. But in the chaos of jumping from bot to browser, I somehow ended up ordering it in Java Brown instead of white. Human error or AI? You decide.

My personal finance take: Was I happy with my purchase using ChatGPT? Yes. Did ordering through AI overcomplicate the shopping experience? Absolutely.

The tool is well-suited for comparing tech, toys and anything where quality isn’t highly subjective. But as things stand, it doesn’t lend itself to clothes shopping.

Interrogating the bot about sizing turned up its signature generic one-liners (“Dynamite clothing is true to size but items may vary”). And yet, clicking the link to purchase on the store website (the tool can’t make the full transaction just yet), a disclaimer about the item running small was readily available. My prompt about natural fibres was also overlooked somewhere along the way.

What I wanted to buy: A sweater I could wear before and after yoga or indoor cycling classes. I was looking for a wrap-style top that was soft and easy to slip on and off, and under $100. Ideally, it would also be as chemical-free as possible and made by a Canadian company.

What I bought on AI’s recommendation: Lululemon BeCalm Wrap-Front Long-Sleeve Shirt in Walnut Crunch – $88 + tax.

What the interaction with AI was like: I also used the free version of ChatGPT to shop. I started by trying to describe exactly what I wanted, with one fatal flaw. In my initial prompt, I said I was looking for a “cross-shawl vibe,” when I absolutely did not want a shawl.

The AI, naturally, took that word and ran with it. Every suggestion leaned scarf-like or drapey in ways that were impractical for working out. It also flagged that Canadian-made, natural-fabric wrap tops under $100 are rare, which was helpful context. I appreciated how clearly it presented options, complete with hyperlinks, but the recommendations themselves missed the mark.

“Maybe the word shawl threw you off. I meant a wrap top,” I clarified. ChatGPT’s results improved dramatically. The chatbot suggested options that looked exactly like what I had in mind, this time including photos. However, it forgot two of my requirements: natural materials and Canadian brands.

After prompting again with those criteria, it suggested several Lululemon options and explained how each material would function during workouts. For example, it recommended the BeCalm wrap top, the one I ultimately bought, noting that it works well for warming up and can be easily removed during more intense parts of a workout.

When I asked for similar suggestions, the AI started recommending items over $100, so I had to re-establish my budget. But those more costly options were tempting.

I asked for a deeper breakdown of the BeCalm top. It described the fabric as plant-based, which is partly true. When I checked the information on Lululemon’s website, the shirt contained modal (derived from wood pulp) but also elastane, a synthetic stretch fibre. To its credit, the chatbot did note the product was not strictly organic or fully natural.

My personal finance take: Over all, the chatbot helped me stay close to budget and narrow down exactly what I wanted. But seeing higher-priced alternatives almost had me overspending, similar to being in a store and having something shiny catch your eye.

Where the tool was most valuable was in helping me compare products and evaluate whether a specific item would work for my needs. I recommend using AI as a shopping assistant, especially for research and comparisons, but be extremely precise with your prompts. Small wording choices can dramatically change results.

What I wanted to buy: A warm crewneck sweatshirt to change up my work-from-home uniform.

What I bought on AI’s recommendation: Gap Heavyweight Oversized French Terry Sweatshirt in Mauve Pink Wash – $44 + tax.

What the interaction with AI was like: I used the free version of ChatGPT and was impressed at first. I liked that it presented options with pictures, prices and links to retailers that opened in separate tabs, so I wouldn’t lose my AI thread.

ChatGPT’s first selection of sweatshirts included the one I eventually bought. While Gap wasn’t a brand I had in mind, the cut and colour of that sweater was exactly what I’d been looking for. This was by chance, though, since I hadn’t included those details in my search prompt.

But ChatGPT also warned me that French terry, the stretchy fabric of the Gap sweater, would not be as warm as a fleece lining. It was a useful tip that kept me searching. Could the bot recommend warmer options with the same cut as the Gap sweatshirt?

The next few suggestions were nothing like what I wanted, and I noticed ChatGPT kept resurfacing sweatshirts from the same handful of retailers, including Amazon.

I decided to feed the bot an old-fashioned keyword search. Gap described the pink sweatshirt as having a “boxy cut.” Could ChatGPT please find me fleece-lined, crewneck sweatshirts with a boxy cut?

The next attempts were better but didn’t include a line of sweatshirts I had eyed at Simons that I knew fit those criteria. So, I asked the bot to search the Canadian retailer specifically.

I was pleasantly surprised that ChatGPT found not only the shirts I had in mind but other, similar ones I hadn’t come across before.

Still, ultimately, I really liked Gap’s washed-out pink. I figured I could layer up with a warmer undershirt and decided for colour over warmth.

My personal finance take: When it comes to apparel shopping, ChatGPT is not ready for prime time yet.

In fairness, I’m sure its ability to anticipate what I like would get a lot better if I kept using it. And I might rely again on ChatGPT to help me find products from a particular retailer.

In general, though, a simple web search seems the faster and better way to find what I want for now.

What I wanted to buy: A V-neck sweater, maybe cotton, maybe wool, that I could wear to the office with (mostly blue) dress shirts. Under $100.

What I bought on AI’s recommendation: La Maison Simons Le 31 V-neck cardigan in ivory white – $39.95 + tax.

What the interaction with AI was like: I rarely make any clothing decisions without significant input from my family, because a roll of the eyes from my partner or daughter can be devastating. So, this was a big step.

I’m probably a stereotypical middle-aged guy, in that I don’t like to shop. Too many options, and I basically shut down – or I start bragging about a fraying shirt I’ve owned for 10 years.

So, a guiding hand could be just the thing I need if I want a sartorial refresh.

I turned to Google AI Mode, which relies on Gemini’s core technology. I rattled off a few general questions about sweaters, then fine-tuned my search: I’m a guy. Show me options in medium. I prefer to buy Canadian.

A cardigan from La Maison Simons popped up among a few other options. And what’s this … $39.95, reduced from $89? But the colour looked a bit bright for me.

Hey Gemini, what colour do you recommend for a guy who has a lot of blue dress shirts? The response was definitive: navy blue. “This option checks all your boxes and offers a classic, professional look.”

I couldn’t argue with that. But a few snags emerged.

I would have preferred more hand-holding than Google AI Mode could offer. I had to cut-and-paste the Simons URL into a web browser, which took me to – oh no – dozens of other sweaters, which forced me to scan through all the options. (That was in early February. I tried a similar approach on Google AI mode about a month later, and already the shopping experience had improved, with direct links to sweaters.)

Then, I couldn’t find the navy blue sweater that Google AI Mode recommended. I returned to AI and asked if the off-white version would work.

Its response: It is perfectly okay “and can look very sharp and intentional, especially with your blue shirts.”

My personal finance take: I can’t argue with the results. I found an inexpensive sweater, in a Canadian store, in about 10 minutes.

But will AI replace my family’s input? I doubt that. My partner initially didn’t love the colour or the fit, and thought the sweater looked cheap.

Despite that withering criticism, I would likely use AI again – especially for those more complex shopping experiences, mostly involving gear and gadgets, that tend to take me down rabbit holes for days and leave me with a headache.

AI is terrific at providing quick summaries and cutting through a lot of jargon and nonsense you’ll find online. For sweaters, I’ll stick with my family. Buying my next computer? AI.