AI on a site should know when to stop

A site chatbot is $750. The rule is simple: it answers from your pages, FAQs, and policies. If it cannot answer, it offers the contact form instead of guessing. Invented prices and made-up policies destroy trust faster than a missing widget ever earned it.

Bound the job before the model

Before anything is public, we write down what the model should answer and what it must refuse. Sales has to trust the handoff. Support has to trust the refusals. If those two lists are fuzzy, the chatbot is not ready.

Personalization without different prices

Personalization is the same idea with a different face. Returning visitors can see a relevant case study or industry first. They should not see a different price than a new visitor. Relevance is allowed. Quiet price discrimination is not.

Drafts are fine. Publishing is human.

Content tools can draft outlines and meta. A person still ships. Voice is part of the product. Raw model output published as brand copy is how sites get ignored by the people you actually want.

Where it sits in the packages

Presence, Starter, and Business can add a chatbot for $750. We plan the source set, the refusals, and the handoff to a human before anything is public.

Missed questions are product input

Unknown questions become the contact form, then get reviewed with the care plan. The source set tightens over time. That loop is the difference between a demo widget and a tool the studio will put its name on.

If a widget cannot name its source, it does not belong on a client-facing page. That filter has saved more projects than it has blocked.