
I bought my ASUS ROG Ally X to play games. Only after I owned it did I start using it as a regular Windows computer. I plug in a keyboard and mouse, answer email, work on a spreadsheet, and go back to my game. If I spent most of my day in spreadsheets, I’d buy a laptop. I spend enough of it gaming that I’d make the same purchase again.
That left me with an odd question. If I’d described my needs to a shopping assistant before I bought it, would it have suggested the computer I now own?
Plenty of buyers now start exactly that way. In a 2026 survey of more than 1,000 U.S. consumers by Semrush’s Exploding Topics, 77.6% said they had used AI to help with a shopping or purchasing decision in the previous six months. See the survey results. So on November 15, 2026, we tried 15 versions of my question in fresh, signed-out ChatGPT sessions. The shopper had about $1,000, put gaming first, and also needed Windows for email and spreadsheets. We varied the gaming priorities and whether the shopper asked for a laptop, a desktop, or a PC.
None of the 15 first answers named an Ally. Two brought up handhelds. When we followed up in those two chats, both suggested the newer ROG Xbox Ally X, and only one also offered my 2024 model as a lower-priced option.
I can defend a laptop recommendation. It puts a larger screen and a keyboard in the price. What interests me as a marketer is something else. Once I name the Ally X, an assistant can explain its merits perfectly well. The harder question is whether it ever comes up when I don’t. An assistant can give sensible advice while leaving out a product the buyer might have preferred, and the buyer walks away with an answer and very little reason to suspect there was another question worth asking.
The printer that turned up when we asked for proof
A 3D printer gave us a cleaner way to test this. When a printer pushes several colors through one nozzle, it can discard filament every time it switches colors. On some jobs, you finish with a very nice model and an even more impressive pile of trash. Our shopper wanted a home printer for around $1,000 that could print in several colors without wasting much filament. They were open to a newer brand and willing to wait.
In six fresh chats, the Snapmaker U1 was mentioned in all six. The Anycubic Kobra X was mentioned once. The Co Print Quadro, a new four-head printer designed to address exactly this problem, never came up. We then ran six more chats with the same question and one extra instruction: look for published comparisons in which somebody weighed the discarded filament from the same print.
Snapmaker still appeared in all six. Anycubic jumped from one mention to five. Four answers cited an actual test by a reviewer: on the same multicolor dragon, the Kobra X discarded 76 grams of filament and the Bambu Lab A1 discarded 230. Being mentioned is not the same as being recommended, and only one of those answers went on to recommend buying the Anycubic on the strength of that comparison. Still, somebody had put the scraps on a scale, and that work made it all the way into the buying advice. See the reviewer’s measurements.
The Quadro still got zero mentions. Three days before our test, a trade publication had reported Co Print’s own measurements on a different four-color model: 135 grams of waste from its printer and 1,079 grams from a single-nozzle machine printing the same object. The Quadro was still being crowdfunded, and the company hadn’t tested it against the Snapmaker. Those are good reasons to qualify a recommendation. They also make it impossible to say whether an independent review would have changed the result. Read the launch coverage and manufacturer’s figures.
| Printer | Shopping question alone | Asked for measured waste |
|---|---|---|
| Snapmaker U1 | 6 of 6 | 6 of 6 |
| Anycubic Kobra X | 1 of 6 | 5 of 6 |
| Co Print Quadro | 0 of 6 | 0 of 6 |
Then we did what most brand teams do. We named the product. We handed the assistant notes on three printers from Snapmaker, Bambu Lab, and Co Print, in all six possible orders. In one set of six chats, the Quadro notes included extra technical details. In the other six, we swapped those details for the company’s waste measurement.
Once named, the Quadro made the top two in all 12 answers, and all six chats that received the measurement repeated it. Yet swapping in the numbers never changed the first choice for a given order. The Quadro ranked first all four times it appeared first in our notes and none of the eight times it appeared later. I wouldn’t build a theory of AI rankings on a small, one-day experiment, and nothing here tells us why the model behaved this way or whether anyone bought a printer differently. I would, however, think twice about an impressive demo that always puts my brand at the top of the prompt.
What a brand name tells you about a broken espresso machine
A machine that matches products to specifications can make brands look interchangeable. But a brand often matters most after two products have both met the specifications. Consider two $400 espresso machines, a Breville and a De’Longhi. We told ChatGPT the shopper had tried both, liked the coffee and milk drinks equally, and had room for either one. In two fresh conversations, with the names in opposite orders, it leaned toward Breville both times.
Then we introduced service terms we made up for the test. One machine came with three years of parts and labor at a local repair shop. The other had one year of coverage, and the buyer had to pay to ship it away. When we gave De’Longhi the better terms, ChatGPT switched to De’Longhi. When we gave them to Breville, it stayed with Breville. The cup of coffee was the same. What changed was the prospect of boxing up a broken machine, paying for postage, and making coffee some other way until it came back.
That is the kind of detail a friend passes along when they recommend a brand. Trust can mean the price was fair, the product did what you thought it would, or somebody picked up the phone when it didn’t. The happiest owner may never have written a review, and the person who did write one because something broke shouldn’t be dismissed for it. I’d want to know what broke, how recently, and what the company did about it. A star average can hide the part of the story that would actually settle my decision.
I see a version of this in my work leading branding at CyberPowerPC. The U.S. retail tracking reports I’ve worked with show CyberPowerPC leading gaming desktops in unit and dollar sales over multiple years. Yet in one of our chats, an answer about the best-selling gaming PC pointed to an iBUYPOWER model at the top of a Best Buy list. Both can be true: one model can lead a retailer’s list while another brand sells more desktops across its lineup. If someone asks what sells most, the period, channel, and unit of comparison belong in the answer. If they ask what gamers prefer, even a complete sales report won’t tell us who bought on a friend’s advice.
When an alternative is worth bringing up
If I were marketing the Ally X, I’d keep gaming at the center of the product page. I’d also give the person wondering whether it could double as their travel computer a short, direct answer. Show someone doing a day’s work on it. Say what they plugged in, what the accessories cost, and when they wished they’d brought a laptop. A reviewer who reaches for a laptop halfway through the test has told me something valuable, and I can decide whether I would have reached for it, too.
Then I’d test the recommendations the way we tested the printers, in three passes:
- Ask without your name. Describe the customer’s problem in the words a customer would use, in several fresh sessions. This tells you whether buyers encounter you at all.
- Ask for proof, still without your name. Repeat the question and ask the assistant to look for published tests or measurements. This tells you whether independent evidence about you exists in a form it can find.
- Name yourself, then shuffle the order. Put your product alongside plausible alternatives and rotate which one comes first. This tells you whether your evidence gives people a reason to choose you, or whether you’re just winning the top of the prompt.
A brand team that only runs the third test is skipping the hardest part of discovery, which customers still have to do on their own.
The people designing these assistants have a job here, too. All 12 unbranded printer answers ended with follow-up questions about what the buyer wanted to print and how many colors they needed. A more useful question might have been whether being willing to wait also meant being willing to back a printer that hadn’t shipped yet. Those are different decisions, and a shopper can answer that question without ever having heard of Co Print.
That’s the standard I’d use for whether a machine understands a brand. I don’t expect it to recommend my handheld to everyone who asks for a laptop. I’d like it to notice when the balance between gaming and work makes a handheld worth discussing, explain the inconvenience along with the appeal, and leave the buyer room to choose. I had to own the Ally X before I knew to ask that question. The next buyer should get to ask it before they spend the money.