Why we built the Conversational PDP

That gap is what the Conversational PDP closes. It puts a real conversation on the page, so the question gets answered in the moment it forms instead of becoming an exit.

The silent question

Watch session recordings of a product page for an afternoon and a pattern emerges. People read the description, scroll to the reviews, hunt for one specific fact, and when they cannot find it fast, they go. The leaving is quiet. No support ticket, no survey, no signal in your analytics beyond a slightly worse conversion rate you cannot easily explain.

The uncomfortable truth is that your product copy can never pre-empt every doubt. Write more and the page becomes a wall nobody reads. Write less and you leave gaps. The format is the problem: a static page answering dynamic, personal questions.

Why not just bolt on a chatbot?

The obvious move in 2026 is to drop a generic AI chatbot on the page and call it done. We deliberately did not build that, for two reasons that matter on a real storefront.

A wrong answer on a PDP is expensive. A generic LLM will confidently invent a spec, a material, a compatibility claim. On an enterprise dashboard a hallucination is annoying. On a product page it is a return, a chargeback, a one-star review. The cost of being wrong is asymmetric, so the safe default has to be an answer the merchant approved.

An open-web answer is off-brand and off-catalogue. Ask a generic assistant about your product and it might recommend a competitor, quote an outdated price, or describe a feature you discontinued. The answer has to be scoped to your evidence: your catalogue, your verified reviews, your UGC. Nothing else.

Curated first, AI second

So the design is two layers. The first is a set of curated question-and-answer pairs the merchant writes (or drafts with AI and approves). When a shopper asks something close to one of them, they get that approved answer back, verbatim, instantly, at zero AI cost and zero risk of a wrong answer.

The second layer is the concierge. When a shopper asks something the curated set does not cover, the question falls through to an assistant grounded in the store's own data, speaking in the brand's voice, inside guardrails the merchant sets. It will not answer from the open web and it will not name a competitor. If it genuinely cannot help, it says so, and can hand the shopper to an external AI on their terms rather than bluff.

What the merchant actually gets

Conversion. The pre-sale doubt gets resolved in the moment, on the page, next to the buy button. The pinned product sits in the chat and recommended products surface inline, so the answer flows into add-to-cart instead of into a new tab.

Lower support load. The same five questions get asked before every purchase. Answer them once as curated pairs and they are answered forever. Fewer "is this in stock / will it fit" emails, more self-served confidence.

Research you cannot buy. Every question a shopper types is a gap in your product page, ranked by how often it is asked. That is the most honest CRO backlog you will ever get, generated by the people you are trying to convert.

Compounding AI-search value. Structured Q&A bound to a SKU is exactly the shape AI assistants quote when a shopper asks them about your category. The same answers that convert on-site help you get cited off-site. The work pays twice.

Why this came out of Idukki

We did not start from "let's add a chatbot". We started from the evidence we already hold for brands: verified reviews, UGC, tagged products, a synced catalogue. A grounded conversation needs exactly that substrate, and most stores do not have it cleanly assembled. We did, so the concierge had something real to stand on from day one. The conversation is only as good as the evidence behind it, and the evidence was the part we had already spent years getting right.

It also fits a belief we keep returning to: the buy decision is becoming a conversation, whether or not the merchant is in the room. Shoppers already ask AI about products before they buy. You can pretend that is not happening, or you can host the conversation on your own page, on your own terms, from your own evidence. We would rather host it.