I Handed My Shopping List to a Chatbot for 30 Days. Here's Where It Crushed Me — and Where It Got Played.

I've spent years writing about how to shop smarter, and I still burn an embarrassing amount of time doing it. Reading reviews. Cross-checking Reddit threads. Opening seven tabs to compare dish soap. So when AI shopping assistants got good enough to actually answer product questions, I wanted to know: can one of these things do my homework for me?

The rules were simple. Over 30 days, I had roughly 30 planned purchases — everything from trash bags to a desk lamp to a birthday gift for my brother-in-law. For each one, I asked an AI chatbot or comparison tool what to buy, then purchased the recommendation without second-guessing it. Alongside that, my usual human-picked equivalents served as the control group, either from past purchases or from my wife's parallel choices for the same items.

Then I graded everything after a month of living with the results.

Where the AI Absolutely Cooked Me

The bot was ruthlessly good at commodities. Paper towels. Sponges. Ziploc bags. Detergent. When I asked for "the best value paper towels for a family of four," it recommended a store brand I'd never considered, and honestly? Identical performance to the name brand I'd been buying, roughly 40% cheaper.

This tracks with everything we've found at Ellis Review over the years: for boring, spec-driven products, the reviews are mostly noise, and the specs are mostly the same. The AI cut through the noise by simply not reading the noise. It compared absorbency ratings and unit prices like a spreadsheet with a personality.

It was also weirdly good at gift-giving. When I described my brother-in-law — "42, into grilling, owns too many cast iron pans already" — it suggested a smoke-setting thermometer probe I'd never have found. He texted me a photo of his ribs two weeks later. Point, robot.

Where It Got Played

Here's where things got ugly: anything with a heavy affiliate or sponsorship ecosystem.

I asked for a recommendation on standing desk converters. The AI confidently suggested a mid-range brand I'd barely heard of, with language that sounded suspiciously like the product's own marketing copy — "ergonomic innovation," "premium build quality." When I dug in, that brand was all over listicle sites that earn commissions on clicks. The AI hadn't found the best product; it had found the product with the best SEO.

The same thing happened with a "cleaning gadget" recommendation and a luggage brand. In both cases, the AI's pick had thousands of suspiciously uniform five-star reviews and a torrent of sponsored placements. My wife's human-picked alternative — boring, well-established, found via a Wirecutter-style deep dive — beat the AI's choice in both head-to-heads.

The pattern was clear: the more money sloshing around a product category, the less you should trust an AI's first answer. Mattresses, supplements, kitchen gadgets, anything sold primarily through affiliate sites — treat the bot's suggestion as a lead, not a verdict.

The Scorecard

After grading all 30 purchases on value, performance, and whether I'd rebuy:

• AI wins clearly: 11 items — almost all commodities and utility purchases
• Tie: 9 items — products where any reasonable choice worked
• Human picks win: 10 items — mostly gifts, apparel, and anything style-adjacent, plus the sponsored-category traps

So the AI went roughly even overall. But it saved me an estimated three hours of research time, and on the items where it won, it often won big — store-brand swaps and value picks I'd have scrolled past.

My Framework: AI as First Draft, Not Final Word

Here's how I'd use it going forward, and how I'd suggest you do:

1. Use AI for commodity purchases. If the product is defined by specs and price — batteries, storage containers, printer paper — just ask and buy. This is where AI beats the review ecosystem, because there's barely anything to game.

2. Never ask "what's the best X." Ask "what should I look for in an X." The second question gets you evaluation criteria instead of a parroted marketing pitch. Then you do the final picking yourself.

3. Ask it to argue against its own pick. "What are the downsides of that recommendation?" produced some genuinely useful caveats — and on the sponsored-category items, occasionally made the bot walk back its own answer.

4. Verify anything expensive or hype-adjacent with a human source. Independent reviewers, enthusiast forums, a friend who owns the thing. AI is a research accelerator, not a research replacement.

The bottom line: the bot was right more often than I expected, and it saved real time. But on the purchases that actually required judgment — taste, fit, gifts, anything a brand is paying to promote — outsourcing the decision meant outsourcing it to whoever had the best marketing budget. Let the AI clear the boring underbrush. Save your own judgment for the decisions that deserve it.