What AI Says About You: the AI Genie playbook

The behaviour you are not seeing

Watch a few real session recordings and a pattern shows up fast. A shopper reads the description, scrolls the reviews, then leaves to "just check". The check is no longer a Google search. It is a question typed into ChatGPT or Perplexity: "is this any good", "what do people say about this brand", "is there something better at this price". The tab opens, the answer comes back, and the shopper either returns or does not. That moment is the single biggest blind spot in ecommerce analytics right now.

What AI Genie actually is

AI Genie is an embeddable widget. It shows a small set of cards, one per AI engine you choose from ChatGPT, Claude, Gemini, Perplexity, Copilot and Grok. Each card carries a line of text and, when a shopper clicks it, opens that engine in a new tab with a search query you wrote, already filled in. The display text is yours to write. The search the shopper runs is live and unaffiliated, which is exactly what makes it credible: nobody can fake what the model says back.

It runs in two scopes. Brand scope is for your homepage or about page, the classic "what AI says about us" panel with an optional staff photo. Product scope is for product pages, where the text fills dynamically from the product being viewed, so one embed covers your whole catalogue without per-product setup.

Why product scope is the interesting one

Brand scope is good social proof. Product scope is the one that changes behaviour. On a PDP, AI Genie resolves the current product (from Product JSON-LD already on the page, or from your synced Idukki catalogue) and fills the card text and the search query with that product's real attributes. The shopper sees a question shaped around the exact thing they are looking at, and one click runs it. The hesitation that used to leak out to an invisible tab now happens on a surface you instrumented.

Per-agent text, because the engines are not interchangeable

A query tuned for Perplexity (which leans on citations and comparison) should not read the same as one aimed at ChatGPT (which is stronger on synthesis and trade-offs). AI Genie gives every agent its own display text and its own search query, both supporting tokens like the product name, price, category and any attribute. You pick a pre-built template to start, fill the blanks, and adjust per engine. Gemini is the one caveat: it has no stable public query parameter, so that card falls back to a Google AI-mode search.

What you can finally measure

Every impression, every card click, and every launch (a shopper actually opening an AI search) is tracked. The launch event is the valuable one. It tells you, for the first time, how often shoppers want to check with AI, which engine they reach for, and on which products the curiosity spikes. That per-product breakdown is a genuinely new signal: it maps AI curiosity onto your catalogue, and it tends to flag the products where your evidence is thin before your conversion data does.

Setting it up

Configuration lives in the Idukki dashboard under the beta tools. Pick a scope, choose a template, write the per-agent text, and save. You get a single script tag to paste, on your homepage for brand scope or your PDP template for product scope. The widget renders inside a shadow root so your site styles and its styles never collide, and it lazy-loads so it never blocks the page. If your PDP already emits Product JSON-LD, product resolution is zero-config. If not, you pass the product handle on the script tag and it reads from your synced catalogue.

html
<div id="idukki-ai-genie-pdp"></div>
<script async
  src="https://app.idukki.io/ai-genie.js"
  data-widget-id="YOUR_WIDGET_ID"
  data-mount="#idukki-ai-genie-pdp"
  data-product-handle="{{ product.handle }}">
</script>