Skip to main content

Common ShopGuide mistakes to avoid

Most ShopGuide problems trace back to a handful of specific, avoidable configuration mistakes rather than the AI itself. This guide covers the ones that actually show up in practice.

Turning on the AI Discount Code Generator without eligible products​

The AI Discount Code Generator lets the AI create a personalized, time-limited Shopify discount code mid-conversation (for example SARAH-LASHES-A3X) — but it only works for products or collections you've explicitly marked eligible in Settings → AI Behavior. Merchants sometimes enable the toggle, set a discount value and duration, and stop there, without selecting eligible products or collections.

The result isn't a broken feature — the AI simply has nothing it's allowed to discount, so it either avoids offering a code or the tool call has no scope to act on. Since the code is scoped in Shopify to your eligible list (not just AI-controlled), an empty eligible list means an empty code. Revisit this list whenever your catalog changes materially — a product you added last month isn't eligible for AI-generated discounts until you add it to the setting. See Settings panel.

Not checking Information Gaps​

Information Gaps Detection runs automatically after every chat response and classifies anything unresolved into knowledge_gap, bug, live_data, system_error, or other, stacking recurring questions into a single row by topic. It's the single most direct signal for what your AI is failing at — and it's easy to forget it exists because it works silently in the background and never interrupts the chat.

The mistake isn't ignoring one flagged conversation — it's never checking the Knowledge Gaps tab at all, so a knowledge_gap cluster that represents dozens of customers asking the same unanswered question (e.g., missing packaging or sizing detail) sits invisible until a merchant happens to read a transcript. Treat system_error entries differently from the rest: a spike there points at the app misbehaving (failed tool calls, wrong-language replies), not at a content gap you can fix by writing more FAQ copy.

Declaring an A/B test winner too early​

ShopGuide's A/B testing framework attributes both chat engagement and completed Shopify orders to each variant, which makes it tempting to check results after a day or two and ship the "winning" variant. Order attribution takes time to accumulate, and traffic patterns vary by day of week — a variant that looks ahead on day two can flip by day seven. Let a test run a full business cycle before reading A/B test results and acting on them, and treat small early leads as noise rather than a signal.

Misconfiguring language for a market outside en/de/nl​

The chat widget's static UI chrome — buttons, guided-discovery overlay labels, footer attribution, disclaimer text — only has translated strings for English, German, and Dutch. If your store's shop_content_language metafield is set to a fourth language, or left inconsistent with your actual storefront language, you'll get a visibly mismatched experience: UI chrome in the wrong language sitting next to AI prose that correctly adapts to the shopper's language (the AI's replies aren't limited to en/de/nl — only the static labels are). Confirm the shop content language metafield actually matches one of the three supported values, or plan for English chrome as the fallback. See Multi-language support.

Treating every chat session as a new customer​

ShopGuide reports unique users, sessions, and interactions as separate numbers for a reason — a single shopper can open multiple sessions, and a session can include multiple messages without ever becoming an "interaction" if they never actually type anything. Reading session count as customer count, or interaction count as conversion count, inflates adoption figures and produces confusing month-over-month comparisons. Use the metric definitions in Success metrics rather than assuming one number stands in for another.

Letting catalog content lag behind the live store​

Because product search and product-question answers come from a vector index of your catalog (Qdrant) plus RAG over your shop knowledge, a product that's been updated, discontinued, or repriced in Shopify but not reflected in the indexed content will produce stale or wrong answers — not because the AI is unreliable, but because it's answering from what it can see. This is the most common root cause behind knowledge_gap and bug-classified entries in Information Gaps; check catalog sync status before assuming a prompt issue.

Next steps​

Most of these mistakes are configuration gaps, not AI limitations — they're fixed in Settings and the transcript viewer, not by rewriting prompts.