Shopping AI hasn’t replaced search. It has joined the journey.
Ctrip’s study follows shoppers between chat and search. Our analysis turns that evidence into a practical product-information audit.
What the study observed
- 31.1 million Ctrip users logged in during July 10–24, 2025.
- 6.1% had used Wendao by July 10: prior adoption, not current chat activity.
- 53% of journeys using both chat and search alternated between them—not 53% of all purchases.
This descriptive Chinese travel-platform study uses 2025 activity. It establishes neither sales causation nor a US retail benchmark. Longer journeys also provide more opportunities for alternation. Read the original paper.
Why it matters for commerce
A product page often assumes the shopper already knows what to ask for. The title names the category. The specification table lists dimensions and materials. But someone choosing a product may still be deciding which constraint matters most.
“Which carry-on should I buy?” can become a question about airline limits, carrying weight, stairs, storage or packing for a particular trip. A keyword, a comparison and a conversation can serve different parts of that decision.
Our practical inference is that the content team should examine the questions between the specification and the purchase. Better answers can be useful even when a shopper never uses an AI assistant. That is a stronger starting point than guessing which phrase will make a model recommend the brand.
WIZIUM perspective
Our reading: explain fit, including its limits
For a luggage brand, a product description could connect exterior dimensions to the space a traveler needs, explain whether the published weight is empty weight, and distinguish a larger capacity from easier carrying. Where airline requirements vary, the page should identify the limit of its claim rather than promise universal compatibility.
This is an illustrative content exercise, not a result from the Ctrip study or a WIZIUM customer case. Its purpose is to help a buyer evaluate suitability. It does not guarantee visibility in AI recommendations.
The useful editorial unit is a question with a verifiable answer: who the product fits, what evidence supports that answer, and when another choice may make more sense. A promotional adjective cannot do all three jobs.
Build a small answer set from existing evidence
Choose a product you know well. Review its current listing, specifications, customer questions and return reasons that your team can legitimately use. Look for repeated uncertainty rather than inventing a new persona for an imagined AI audience.
- Intended use: what job does the buyer need this product to do?
- Constraints: which dimensions, materials, compatibility requirements or maintenance needs can rule it out?
- Trade-offs: what does the buyer gain, and what do they give up compared with a relevant alternative?
- Evidence: which current specification, instruction or documented test supports the answer?
- Ownership: who updates the answer when the product or its terms change?
Keep uncertainty visible. If a claim depends on a specific model, configuration or test condition, carry that condition with the claim. Do not let a shortened listing or support reply turn a limited statement into a blanket promise.
Check the same question across touchpoints
Follow one buyer question through the product page, marketplace listing, comparison content and support material. Does each answer refer to the same version of the product? Are the dimensions, included items and exclusions consistent? Can the reader find the source of an important claim?
If the answers disagree, assign a source and an owner before producing more content. A larger volume of polished copy can make an inconsistency harder to fix if nobody knows which version is authoritative.
For an initial content trial, watch for repeated questions, avoidable clarification and complaints about misunderstood features. If you also measure conversion, define the period and comparison carefully. A change after publication alone cannot establish what caused it.
Try this with your team
Audit five questions about one product
- Select five questions that customers or the sales team already ask.
- Write an answer to each using current evidence and clear limits.
- Check those answers against every active listing and support source.
- Ask someone unfamiliar with the product to explain who it is suitable for and why.
Use what they misunderstand to choose the next edit. Keep the first exercise small enough to maintain.