Answer from the product information you maintain.
Visitors may ask whether a feature fits their workflow, what a setup step involves, or where to find a published plan detail. Give the assistant approved, current sources for those questions.
Separate released functionality from a roadmap or proposal. If a page discusses a future capability, make its status explicit and exclude it from the answer source where it could create a false expectation.
Example visitor question
“Can different clients see only their own project?” The answer should reflect the product’s actual access controls and any stated plan conditions, with a route to the relevant documentation.
Keep the public assistant’s scope clear.
A website assistant using indexed public content does not automatically know the visitor’s subscription, account permissions, or private project data. A link to the product’s sign-in page does not grant access to those records.
Direct account-specific issues to the established authenticated support process. The assistant should explain that boundary instead of presenting a guess about someone’s plan or account state.
Collect context for a useful product conversation.
Where the assistant’s goal enables contact collection, it can save a lead with the visitor’s request. Use that context to understand the intended use case before the team follows up.
Ask for information that helps route the enquiry, such as the problem being solved or the number of sites involved. Avoid requesting passwords, tokens, or private customer datasets in a general website chat.
Use recurring questions to improve documentation.
An unanswered question may reveal a missing setup step, ambiguous feature description, or a pricing condition that is hard to find. Review the conversation and identify the public page that should answer it.
Update that source, refresh the indexed content, and retest the original question. Keep a small set of product-critical questions for reviews after a release, plan change, or documentation reorganisation.
Use Ask Sairo to review a misunderstood feature or an outdated public answer, then provide the approved correction. A Claude MCP connection can bring permitted Sairo conversations and knowledge into your team’s AI workflow without exposing private product accounts.
Treat the marketing site as a real integration.
Install the script in the intended page scope and check it across navigation, documentation pages, and narrow layouts. Review sticky calls to action, cookie controls, and any existing support launcher for interference.
When live chat is enabled, assign the product specialists who cover that site and test both the online and offline states. A specialist can join from the Sairo workspace and reply inside the visitor’s existing widget conversation.