Zendesk is betting your customers will ask ChatGPT before they ask you — and building for that
At its Relate conference in 2026, Zendesk announced it is extending its AI agents beyond its own interface and into third-party platforms including ChatGPT and Gemini, alongside voice and messaging channels. The company frames this as a shift toward "meeting customers where they are" rather than pulling them into a branded support app or website. Shashi Upadhyay, Zendesk's President of Product, Engineering and AI, argued the logic behind the move by pointing to how people already search: "You're just going to go to Perplexity or Gemini… you're not going to go to a website and search for it." Zendesk says the expansion sits alongside a new framework meant to keep context consistent across channels, and is separate from the company's other announcements at Relate around MCP (Model Context Protocol) integrations. Voice AI agents were also updated to handle language switching mid-conversation.
What is actually new here, beyond the branding?
Zendesk is positioning its AI agents to answer queries surfaced inside ChatGPT and Gemini, not just its own widget, with a shared context layer across channels.
That's a meaningful architectural claim if it holds up: today, most vendors' "omnichannel" pitch means your web chat, email and phone queues share a queue and a dashboard. Zendesk is describing something further out — the support agent itself showing up inside a general-purpose AI assistant a customer already has open, rather than the customer coming to you. TechRadar's coverage, drawn largely from Zendesk's own framing at its conference, doesn't detail how the underlying integration works technically, what data crosses into OpenAI's or Google's infrastructure, or under what terms.
Why should a buyer care where the conversation happens?
Because if your support answers surface inside someone else's chatbot, you lose visibility into how, when and to whom they're delivered.
Handing resolution to a third-party AI interface changes who controls the first impression of your brand and, potentially, what data leaves your own systems. Support leaders who've spent years consolidating channels into one platform for auditability should ask whether "meeting customers in ChatGPT" reintroduces the fragmentation they were trying to eliminate — just one layer further from their own logs.
How much should you trust the numbers Zendesk is citing?
Cautiously — the reporting here relays Zendesk's own framing from its own conference, with no independent measurement disclosed.
The source article, from TechRadar, is essentially a writeup of vendor claims made at Relate: extended language switching for voice, a "consistent context" framework, reach into ChatGPT and Gemini. None of it is benchmarked against a third party, and no methodology for what "consistent context" means in practice — resolution rate, handoff accuracy, latency — is given. That's normal for a launch story, but it means the burden of proof sits with your own pilot, not the press release.
What should you ask your vendor before adopting this?
Ask exactly what data is shared with ChatGPT or Gemini, who owns the transcript, and how "consistent context" is measured.
Push specifically on data residency and retention when a query routes through a third-party AI platform rather than staying inside your existing support stack, and ask for a definition — not a demo — of what "consistent" context means across a handoff from voice to messaging to a third-party chatbot.
Frequently asked questions
Is this the same as Zendesk's MCP announcement?
No. TechRadar's report is explicit that this cross-platform AI agent expansion was announced separately from Zendesk's MCP efforts at the same Relate conference in 2026.
Does this replace Zendesk's own chat widget?
The reporting doesn't say Zendesk is retiring its own interfaces — the framing is additive, extending agents into ChatGPT, Gemini, voice and messaging alongside existing channels.
Source: TechRadar, "Zendesk expands AI agents across ChatGPT, Gemini, voice and messaging," by Craig Hale, published 21 May 2026.