Chat management
Chats is where you see what your customers and the AI actually said to each other — what worked, what didn't, and what keeps coming up that your store content doesn't answer yet.
Getting to your transcripts
Click Chats from the dashboard, browse the list, narrow it down with search and badge filters, and click into any conversation to read the full thing.
Each row in the list shows the customer identifier (or an anonymous ID if there isn't one), when it started, the page context, and badges for any notable tool calls — Discount, Guided Search, Web Search, and similar. Routine tools like basic product search don't get a badge; there'd be too many of them to be useful.
Inside a transcript
Open one and you get the full message timeline with timestamps, the customer's context (page, product, order info if any), and the detected intent — pulled from ShopGuide's 12-intent router: product search, comparison, guided discovery, order management, support, checkout, and a few more.
You can search by keyword, filter by badge (tool used, ⭐ starred, your own 👍/👎 rating), and filter by date range.
Starring flags a conversation for follow-up. Rating lets you leave a 👍 or 👎 directly on an AI response — a 👎 asks for a short note on what went wrong. This is your own internal quality review, not a satisfaction score collected from the customer.
Knowledge Gaps
The Knowledge gaps tab surfaces what ShopGuide's background detector found across conversations — things the AI couldn't fully resolve. Each row falls into one category:
| Category | Meaning |
|---|---|
| Knowledge gap | Your store is missing documented info — packaging, materials, sizing, shipping, policy |
| Bug | A customer reported something actually broken — add-to-cart, checkout, pricing |
| Live data | The AI needed real-time data it has no tool for — exact delivery date, live stock |
| System error | The chat app itself malfunctioned |
| Other | A real need outside the above — wholesale, partnerships, careers |
Different phrasings of the same underlying question get stacked into one row with an occurrence count, rather than listed separately, so what you see is what actually needs attention — not forty near-duplicate entries.
Next steps
- Set up an A/B test informed by what you see in transcripts
- Adjust AI behavior in Settings to close knowledge gaps
- Review chat analytics