Maintenance & ongoing optimization
ShopGuide itself doesn't degrade over time. Your store does keep changing around it — new products, updated policies, seasonal discounts, maybe a new market. Maintenance is really just making sure those changes actually reach what the AI can see and act on. A short, practical list, not an ops framework.
Keep your catalog content synced
Product search, product questions, and guided discovery all pull from a vector index of your catalog plus RAG over your shop's content. Discontinue a product, reprice it, or rewrite its description in Shopify without that change reaching the indexed content, and you'll get a stale or wrong answer — not because the AI got worse, but because it's still answering from what it could last see. After any real catalog or policy change, check that it shows up in chat answers; don't assume the sync is instant.
Actually look at Information Gaps
Because this detector never interrupts the chat, it's also easy to forget it's running at all. Pick an actual recurring time — weekly works for most stores — to open the Knowledge Gaps tab, work through open knowledge_gap and bug entries, and watch system_error specifically as a signal about the app itself, not your content. Recurring questions are already stacked by topic, so a high count there is a real prioritization signal, not noise. See Chat management.
Keep the discount generator's eligible list current
If you're using the AI Discount Code Generator, its eligible-products list doesn't update itself. A product line you launched last month isn't eligible for an AI-generated discount until someone adds it in Settings → AI Behavior. Review it whenever you refresh your catalog, and double-check the discount value, duration, and once-per-customer setting still match what you actually intend to run. See Settings panel.
Don't let A/B tests linger — or end too fast
A test running for months without a decision is wasted signal. One called off after two days is probably noise. Decide up front how long a launch-message or prompt test should run — a full business cycle, at minimum — before reading A/B test results, and archive tests once you've acted on them instead of letting old variants keep splitting traffic indefinitely.
Recheck language settings after a theme or market change
The widget's UI chrome only has translated strings for English, German, and Dutch, driven by the shop_content_language metafield. A theme migration or a new market can quietly reset that value. If it does, you end up with English buttons next to AI replies that correctly adapted to the shopper's own language — which reads as a bug, even though the AI side is working exactly as intended. See Multi-language support.
Revisit custom prompt setups periodically
If your store runs on a fully custom system prompt or a restricted tool set instead of the standard flow, remember it's maintained on its own — new intents, tools, or features added to the standard configuration don't automatically reach a custom one. Check every so often that it still reflects what you actually want, not a snapshot from whenever it was first built. See Custom system prompts.