Product page AI: answering before support ever sees it
See how an AI assistant sitting on the product pages turned unanswered questions into add-to-carts, and fed the product FAQ while it worked.
726
Conversations
Handled without a human
98
Add-to-carts
From an answered question
43
Orders attributed
Tracked to a conversation
~€4.2K
Revenue attributed
Over the measured period
Executive summary
An online store was losing the questions its visitors asked. Size, compatibility, materials: everything that decides a purchase went to a contact form or nowhere. We put an AI assistant on the product pages, answering from the catalogue, and mined every conversation for the product FAQ.
The challenge
A question asked on a product page is a purchase in progress. Answer it in seconds and it converts; route it to a contact form and it disappears. The store had no answer path at all on the page itself.
The questions were also being wasted. Nobody was reading them in aggregate, so the product pages kept failing to answer the same things week after week.
A generic chatbot would have made it worse. An assistant that hallucinates a compatibility claim on a cookware page produces a return, a bad review, and a support ticket that costs more than the sale.
What we built
An assistant on the page, grounded in the catalogue
The assistant answers from the product data rather than from general knowledge, so it says what the catalogue says or nothing at all.
Every conversation logged
The full history is exported and read, not just the first message of each thread, which is all the tool's own interface shows.
The FAQ written from the questions
The conversations were sorted by product into a per-product FAQ, including the questions the assistant could not answer, which are the most useful list in the whole export.
The stack
Claude answers, grounded in the store's own product data pulled over the storefront API. The conversation export and the FAQ analysis were done with Claude Code.
Results
726 conversations and 1,686 messages over the measured period, handled without a human.
98 add-to-carts and 43 orders traced back to a conversation, for around 4,236 EUR in attributed revenue.
The most-questioned products came out of the analysis ranked, which is a product roadmap as much as a support one: when one product generates three times the questions of any other, its page is the problem, not the customer.
What we learned
- An unanswered question on a product page is not a support ticket you avoided. It is an order you lost quietly.
- The value is in the transcript as much as in the answer. The questions tell you what the page failed to say.
- The list of questions the assistant could not answer is worth more than the list it could.
- Ground the assistant in the catalogue or do not ship it. On physical products, one confident wrong answer costs more than a hundred right ones earn.
- Read the whole export. The tool's own interface showed only the first message of each conversation, which made the volume look five times smaller than it was.
How it works
Put the answer where the question is
On the product page, not behind a form.
Ground it in the catalogue
The assistant answers from product data or says it does not know.
Log everything
Full conversations, exported, not just the opening message.
Feed the pages back
Turn the recurring questions into the product FAQ so the assistant has less to do.
See what your team
can hand to AI
A free 30-minute call. We map one workflow together and you leave with a plan you can use, with or without us. Meeting us at the AI Summit Barcelona on September 22? Grab your slot here.
Book a Free Strategy Call