WIRED shows how ChatGPT distorts its editors' picks for the best laptops, TVs and headphones

A WIRED journalist tested how accurately ChatGPT can recap the magazine's tech recommendations, and found that the bot regularly gets them wrong. Even when the model links to the right guide, it may replace the editors' real pick with its own version.

What WIRED tested

The trigger was OpenAI's new push into AI shopping. The company promises that ChatGPT will help people choose products faster and spare them endless tab-hopping and "best of" lists. WIRED chose to test not the abstract quality of the advice, but a narrower, measurable case: can the bot accurately recap which devices the editorial team has tested itself and recommends right now? It is a convenient target for such a test, because the publication has regularly updated guides built on real reviews and hours of hands-on testing.

The questions were deliberately simple: which TVs, wireless headphones and laptops does WIRED currently consider the best? The key condition was that the answer must reflect WIRED reviewers' picks specifically, not generic market equivalents or "similar-class" models. In other words, ChatGPT was not asked to invent a handy summary from memory, but to carefully recap material that was already published. It was at this basic level that the model began to fail, even though its answers looked convincing and came with correct links.

Where the answers broke down

The most telling failure came in the TV category. ChatGPT opened the correct WIRED guide but named the LG QNED Evo Mini-LED as the best option for most people, a model that was not on the editorial list at all. WIRED's actual pick was the TCL QM6K.

The error is especially nasty because a reader can easily miss the swap: the link goes where it should, product photos sit alongside, the tone is confident, and it feels like a simple summary of the article. In fact, the bot had mixed its own interpretation into someone else's editorial work.

A similar story repeated in other categories. In the headphones roundup, ChatGPT presented the AirPods Max 2 as WIRED's recommendation for Apple ecosystem users, even though at the time of the test the editors had not yet tested them and had not added them to the guide. For laptops, the bot stubbornly named last year's MacBook Air M4 as the best choice, although the guide's current top pick was already the 2026 MacBook Air M5. When the journalist pointed out the inconsistencies, the model itself admitted that it had replaced the exact recommendations with more general and, in places, outdated options.

• TVs: the bot replaced WIRED's top pick with a model that was not in the guide.
• Headphones: it credited the editors with advice to buy a device they have not yet tested.
• Laptops: it gave the outdated MacBook Air M4 instead of the current MacBook Air M5.
• In every case: a correct link sat next to an incorrect recap.

"I took WIRED's actual pick and replaced it with a more general, similar option,"

ChatGPT admitted after being challenged.

Why it matters

The problem is not only the factual error itself. When someone asks "what does WIRED recommend," they want not the model's averaged answer but the outcome of a specific editorial process: tests, comparisons, updates and the author's final choice. When ChatGPT swaps one product for another, the user may take it as the publication's position and spend money relying on an authority that, in this particular recommendation, is not actually there. For media, this is no longer a question of interface convenience but of attribution accuracy and trust in the brand.

There is also a business side. WIRED states plainly that the magazine earns money from affiliate links, and that revenue helps fund reviews and long-term gadget testing. ChatGPT answers built on the publication's material carry no such links, even though the model relies on the editors' work while reducing the incentive to visit the site. The situation looks especially ironic given the Condé Nast deal with OpenAI, which is meant to help surface links to the publisher's material inside the chatbot. Formally the link is there, but it is of little practical use if an incorrect conclusion sits next to it.

What this means

WIRED's experience shows that even the updated AI shopping still handles the task of accurately recapping other outlets' recommendations poorly. If you need the conclusion of a specific editorial team, lab or testing group, it is still better to go to the primary source. For OpenAI and other players, this is a signal: in commerce scenarios, a convenient output is not enough; literal fidelity to the data matters too.