AI Claims to 'Understand You' But Can't Sell a Ready-to-Use Quilt

The Incident: Three Days Trying to Buy a Quilt

When the weather turned cold, the author needed a thicker quilt for their child. On JD.com (referred to as "a certain East"), they found a brand they liked but discovered the quilt core and cover were sold as separate links. Customer service repeatedly replied: "It's just the core; the cover must be bought separately." Even after asking again, the answer was the same: "Both need separate orders." The frustration was immediate — buying a quilt shouldn't require this much back-and-forth.

"宝儿，是单被芯的，被套需要单拍的呢。"

"都是需要单拍呢。"

The author then found a bundled "quilt + cover" set, but reviews were dominated by complaints of strong, persistent odors: "Has odor, still smells after two days in sun," "Quilt stinks, dare not use," "Smell very heavy, hasn't dissipated." The choice became either assemble pieces yourself or gamble on a potential defect.

"有异味，在太阳下面晒了两天还是有异味。" "被子都是臭的，都不敢用。" "但是味道非常重！好久都没消。"

Finally, on Taobao ("a certain Treasure"), the experience was smooth: a single link listed AC quilt, spring/autumn quilt, winter quilt, and thick winter quilt, each labeled with suitable room temperature — e.g., "Spring/autumn quilt 150×200cm ~4kg, suitable for 15-25°C." The author's home has heating in winter, so the spring/autumn option was perfect. No assembly, no guesswork.

Where the Problem Lies

E-commerce pages are still organized by warehouse logic: cover, core, and set are separate SKUs, each with parameters, sales volume, and ratings. The user's mental model is a single task — "I want a quilt I can directly use" — while the system stores hundreds of SKUs and expects the user to bridge the gap.

Platform organization: Separate links for cover, core, set; parameters, sales, ratings; scripted service replies ("need separate orders"); negative reviews buried for user to find.

User need: A ready-to-use quilt; context like room temperature, user, budget; a pre-assembled set; upfront suitability assessment.

If the user bridges the gap well, it's luck; if not, they waste money.

What AI Should Do If It Truly Understands

Current AI e-commerce focuses on "guess what you like" — recommending products, not understanding the user's task. The real divide: first clarify what the user wants to do, then decide what to show.

If AI actually did this, the author's purchase would have been far simpler:

Bundle components automatically. When the user says "I want a spring/autumn quilt with a cover," the system should combine core and cover into a single order entry. The "need separate orders" reply is a one-sentence fix.

Translate parameters into human language. "150×200cm, ~4kg" is less useful than "suitable for 15-25°C room, single bed." Taobao has taken the first step; the next step is to combine season and delivery location's heating status to default the right choice.

Surface negative reviews proactively. If "odor" appears repeatedly in reviews, the page should show a banner: "Recent users reported odor; issue resolved / returns supported." Problems aren't scary; hiding them so the next buyer steps in the same trap is.

After-sales that recognizes intent. When a user says "odor makes it unusable," the system should classify it as a quality complaint, initiate the return process, and feed the issue back to the merchant — not wait for a negative review to count.

With all the user data platforms hold, doing more to improve experience and efficiency is the only right path — why hasn't shopping improved in the AI era?

Further Thoughts: The Future Shopping Interface

Search boxes and category trees will matter less. The entry point becomes a natural-language scenario: "Home has heating, need a single spring/autumn quilt with cover" — and the system returns a solution.

Product pages will no longer be static; they'll rearrange per user — parameters, descriptions, Q&A all centered on the user's specific use case. "Thousand faces" shifts from recommending products to giving answers .

Scattered signals — reviews, service chats, return reasons — will be aggregated and fed back to merchants to improve products. Negative reviews then become not just user pain but direct product feedback.

Final Word

The "need separate orders" line isn't the customer service agent's fault; they're reading a script. The real problem: systems (JD, Taobao, etc.) never treated "user wants a ready-to-use quilt" as a goal — only as a category keyword to match.

AI buzzwords keep piling up — intent recognition, generative recommendation, shopping agents — but they all boil down to one thing: don't make the user do the system's work.

I just want to buy a quilt. Let the AI handle the rest while I wait for delivery. Is that too much to ask in this era?