The field guide to conversational AI that actually converses. Luxury concierges, travel designers, retail clienteling, sommelier and art advisors, character companions doing billions of turns — and the regulated-conversation engines underneath. The spine tension: at the top end the winning design is AI-drafted, human-fronted. At the regulated end it is machine-fronted, disclosure-first. Both are right. Neither works in the other's context.
The field's front door. New launches, funding rounds, and platform moves land here first — each scored P/Q/S (productivity · quality · sellable, out of 9) with a plain verdict: act, watch, or skip. What proves out graduates into the Stack or the Voices; the rest decays. The freshest section on the site.
The highest-value, highest-design-sophistication conversational AI is running at the HNW and luxury end of the market — and almost nobody is documenting it. Concierge operators, premium banking, luxury retail clienteling, auction-house lot-matching. Where AI status is inferred from competitive dynamics rather than confirmed by the operator, it is marked. Flag-don't-invent is the rule.
Who to read on conversational AI that actually converts — the builders, the operators, the researchers, and the practitioners documenting the stack in production. X handles rendered as links only where verified; unverified ones are flagged, not faked.
Four layers: consumer platforms, messaging channels, enterprise resolution engines, and build-on-API toolkits. Cost per layer where published. Filter by category. Real logos via logo.dev → DuckDuckGo favicon → monogram fallback.
The working vocabulary of conversational AI — from disclosure-first and masquerade through to vertical RAG and database reactivation. Every pattern is a design decision with a documented consequence. The two poles — AI-drafted, human-fronted and machine-fronted, disclosure-first — are the spine of this whole field.
The primary sources beneath the practice — FCA regulation, ICO guidance, the CASA framework, and the academic literature on persuasive design and vulnerable users. Every link goes to the source itself, not a write-up of it. Expands to Finding · Context · Why it matters.
The live arguments as of mid-2026. Should a bot hide being a bot? Voice or text? Where is the regulatory line in lead-gen AI? No settled answers — that's the point.
The teardowns, regulation documents, and platform references that inform every editorial decision on this site. Cards for the library canon; the full table below searches across all sections.
| Kind | Name | Detail |
|---|