How Caro uses your training data to answer questions

Everything in Sources — both Store data and Custom sources — becomes part of what Caro can reference when a customer asks a question.

The basic idea

When a customer sends a message, Caro searches across all your indexed sources for the most relevant matches, then uses those matches to write its reply. This means:

• The more relevant, accurate content you've added, the better and more specific Caro's answers will be.
• Caro will only answer based on what it's actually been given — it doesn't guess store-specific facts (prices, policies, stock) that aren't in your sources.
• If nothing relevant is found, Caro falls back to a general, honest response rather than inventing an answer — see Guardrails: what Caro won't say.

Why FAQ and Text block sources are worth adding even with store data connected

Store data gives Caro your products, pages, and policies verbatim — but customers often ask things that are never written down exactly that way, e.g. "can I combine two discount codes?" or "do you restock sold-out sizes?". A short, deliberately-written FAQ entry answering that exact question in your own words is often more reliable than hoping the answer is implied somewhere in a policy page.

Freshness matters

Caro answers from whatever was indexed at the last sync — not the live page in real time. If you just edited a policy or product description, either wait for the next scheduled sync or force one immediately via Sync now (see Editing and removing sources) so Caro's answers reflect the change right away.

Training data vs. Actions

Sources give Caro knowledge — facts it can talk about. They don't let Caro do anything (look up a specific order, apply a discount). That capability comes from Actions, configured separately — see Sales & Support Actions.