If shoppers are getting answers from the chat bot that don’t match your store — wrong prices, missing products, made-up specs — that’s a trust problem you want to fix today, not tomorrow. The good news: most accuracy issues trace back to a small number of root causes, and each one has a concrete fix. This article walks through the diagnostics in order, from the most common cause to the rarer ones.
How the chat bot generates answers
Before fixing accuracy, it helps to understand the source. StoreAgent Chat answers from AI Memory — a snapshot of your store data that the plugin indexes for fast lookups. When a shopper asks a question:
- The bot searches AI Memory for the most relevant indexed content (product descriptions, page text, etc).
- It writes an answer grounded in what it found.
- If nothing relevant is indexed, the bot either declines to answer or — and this is the failure mode — makes up plausible-sounding content (a “hallucination”).
So accuracy fixes generally fall into one of three buckets: fix what’s indexed, fix how the bot uses the index, or fix the source content the bot is grounding on.
Step 1: Confirm AI Memory has indexed your products
If the index is empty, the bot has nothing to answer from. Check this first.
- Go to StoreAgent → Settings → AI Chat → Post Types.
- For each post type you want the chat to reference (Products, Pages, Posts), confirm the enable toggle is on and the indexed fraction matches the total. For example, 19/19 means all 19 products are indexed. 3/19 means 16 are missing.
- If the numbers don’t match, trigger a manual sync — see How StoreAgent Chat accesses your product data for the steps.
Step 2: Improve your source content
The bot is only as accurate as what you’ve written. If product descriptions are thin, missing, or inconsistent, the bot fills the gaps with guesses. The single biggest accuracy lift on most stores is filling out product content:
- Long descriptions — at least 100 words covering features, dimensions, materials, and use cases. Short descriptions starve the index.
- Short descriptions — a 1-2 sentence summary on every product, so the bot has a canonical “what is this?” line to quote.
- Categories and tags — these help the bot answer category-level questions (“do you have any waterproof jackets?”) without parsing every individual product description.
- FAQ and policy pages — shipping, returns, sizing guides. Enable Pages indexing under Post Types so these answer non-product questions too.
If a lot of products are missing content, bulk-fix them with Product Description AI and Product Summaries AI instead of writing each by hand.
Step 3: Add custom fields the bot needs to see
If your store stores key product info in custom meta fields (size, voltage, allergens, ISBN — anything not in the standard WooCommerce fields), the bot won’t see them by default.
- Go to StoreAgent → Settings → AI Chat → Post Types.
- Expand the Products panel.
- In Metadata Keys to Include, list each meta key the bot should reference, comma-separated. For example:
_size_chart, _voltage, _allergens. - Click Save Post Types Settings, then trigger a manual sync so the new fields land in the index.
Step 4: Set a clear tone and instruction
StoreAgent → Settings → AI Chat → General has a Communication Tones selector and an instruction field. Both shape how the bot answers. Common tweaks that help accuracy:
- Add a sentence like “If you do not know the answer, say you don’t know and offer to escalate to support — never guess.” Telling the bot when not to answer reduces hallucinations.
- Add brand-specific facts the bot should always know: shipping cutoff times, return window, who you ship to.
- Pick a tone that matches your brand voice. A “Professional” tone tends to be more conservative about uncertain answers than “Casual”.
Step 5: Enable customer feedback so you spot problems early
Add thumbs-up / thumbs-down ratings to every chat reply so you can see exactly which answers shoppers found wrong. See How to set up chat feedback and escalation.
Turn on Request Reason on Negative Feedback so negative ratings come with a free-text explanation — those explanations point you straight to the product or topic that needs better content.
Step 6: Test, iterate, and review chat history
Once you’ve applied steps 1-5, open your storefront in a private browser window and ask the bot:
- Three questions you know have correct answers in your product catalog (verify the bot finds them).
- Three questions about products that don’t exist (verify the bot declines instead of inventing them).
- Three borderline questions where the answer is ambiguous (e.g. “is this product washing-machine safe?”) — these are the ones most prone to hallucination.
Review the live chat history under StoreAgent → Settings → AI Chat → Chat History to see what real shoppers are asking and where the bot is struggling. Use the feedback ratings to prioritise which content gaps to fix first.
Troubleshooting specific accuracy issues
The bot quotes the wrong price
Two common causes. If the price changed recently, sync the product manually — the index might still hold the old value. If the price is correct in the index but the bot still misquotes, check whether the customer is logged in with a wholesale role (if so, see the Wholesale Suite integration article). For currency-specific stores, the bot uses the price stored in the product, not a converted one — multi-currency stores need additional handling.
The bot makes up products that don’t exist
Classic hallucination. Add the explicit “say you don’t know” instruction from Step 4 to your chat settings. The bot is much less likely to invent products when it’s been told not to guess.
The bot gives different answers to the same question
Inherent to AI — generative models have a temperature setting that introduces variation. For factual questions about specific products, the bot should be consistent. If different sessions give wildly different answers, the underlying content is probably ambiguous — check the product description and see what’s open to interpretation, then rewrite for clarity.
The bot ignores my brand-specific FAQ pages
Confirm Pages indexing is on in Post Types and that the specific FAQ pages aren’t on the Excluded Pages list (Display tab). Trigger a manual sync after enabling. The FAQ content should appear in chat answers on the next conversation.
Need help?
If you have a question or run into any issues, we’re here to help.
- Premium users: Open a support ticket
- Free users: Visit our community forum
