Yes, this is very achievable, and Proactive Messaging is probably the right starting point.
You could configure a proactive message to appear only on selected product pages, potentially after the visitor has spent a certain amount of time there. If they engage, the response can be handed to your published AI agent, which then runs a product-recommendation dialogue or procedure. Zendesk specifically supports URL-based conditions for proactive messages and allows the follow-up conversation to be managed by an AI agent.
I would design it something like this:
Proactive message:
“Not sure which option is right for you? I can help you choose.”
Product recommendation use case:
The agent asks a short series of questions about needs, preferences, budget, environment, or whatever factors genuinely differentiate the products.
Recommendation:
The agent recommends one or two products, explains why, and provides direct links to the appropriate product pages.
Fallback:
If the answers do not produce a confident recommendation, offer a handoff or another way to contact the sales team rather than forcing a result.
One clarification: adding a conversation tag such as product_recommendation would be useful for reporting, routing, and giving an eventual human agent context, but I would not depend on the tag itself to trigger the use case. AI agent use cases are normally selected by matching what the customer is asking to the relevant dialogue or procedure.
Depending on the version of the AI agent experience in your account, the proactive message configuration may allow you to reference the specific published answer that begins this experience. Otherwise, you will want the opening message and use-case description to make the intended path unmistakable. Interestingly, Zendesk has also documented product recommendation as an e-commerce AI agent pattern, so you are very much within the intended shape of the platform.
I would also consider whether the recommendation experience should live inside the same agent as post-purchase support. It certainly can, but the use cases, instructions, knowledge sources, escalation rules, and reporting should be clearly separated. “Help me choose” and “help me with what I bought” sound similar conversationally but represent very different customer journeys.
