How Do You Use ChatGPT and AI in Zendesk? 4 Routes Compared (2026)

You use ChatGPT-style AI in Zendesk through one of four routes: Zendesk's own AI agents and Copilot (Copilot is $50 per agent a month on Professional plans and above), a Marketplace app running on your own OpenAI API key, a third-party AI agent layer such as Macha, or your own build on the OpenAI and Zendesk APIs. There's no single "ChatGPT button," and the right route depends on who you want running the model, who pays for it, and who owns the privacy posture.

We sell an AI agent ourselves (Macha, covered in route 3), so treat this as a practical map from people who watch this market closely, not a neutral encyclopedia. Each route gets real pros, cons, and the data-privacy notes that matter.

Does Zendesk already use OpenAI?

Zendesk already uses OpenAI under the hood. Zendesk runs a multi-LLM architecture: its Generative AI at Zendesk article lists OpenAI, Google Gemini, AWS Bedrock, Microsoft Azure, ElevenLabs, Groq and Fireworks AI as providers, depending on the feature. It also has a public partnership with OpenAI to power features like generative replies and AI agents (OpenAI). So when people say "I want ChatGPT in Zendesk," they're often already half-there: the native features are GPT-powered, just wrapped in Zendesk's own product and billing.

That matters for two reasons. First, you may not need to bolt anything on. Second, the privacy posture is better than people assume. Zendesk says it uses "zero data retention" endpoints for providers such as OpenAI, Google Gemini and ElevenLabs, so request and response bodies "exist only in memory," and that "no third-party will use your inputs to train their models" (Zendesk help: AI Data Use). If you wire up your own ChatGPT integration instead (routes 2 and 4), you own that privacy posture, so read the privacy note in each section.

How do you use Zendesk's own generative AI? (Route 1)

This is the path of least resistance. Zendesk ships GPT-powered features inside Agent Workspace and its messaging/ticketing, split into two buckets.

Copilot helps your human agents. It drafts and expands replies (turn bullet points into a polished response), rewrites tone with one click, summarizes long ticket threads so an agent can catch up fast, and suggests next steps. It also powers intelligent triage, auto-detecting intent, sentiment, and language so you can route tickets.

AI agents (Zendesk's rebrand of what used to be "Answer Bot") handle conversations end to end. They read a customer's question, search your connected knowledge (Help Center articles plus external sources via the web crawler) and reply to deflect or resolve the ticket without a human. The newer AI agent builder lets you define procedures in natural language; the agent then plans a course of action with adaptive reasoning and shows a preview of its proposed steps before you publish.

How to turn it on (high level):

1. Confirm your plan. Plans run from $19/agent/mo (Support Team) and $55 (Suite Team) to $115 (Suite Professional), billed annually (Zendesk pricing).
2. Add the AI capability. The Copilot add-on is $50/agent/mo (annual), and Zendesk says it "can be added to Professional and higher plan levels," so Team plans have to move up a tier first (Zendesk help: Buying Copilot).
3. Build your knowledge base. AI agents are only as good as the Help Center and sources you connect: thin docs produce a thin agent.
4. Configure agents, triage rules, and escalation, then test before going live.

The cost catch. Beyond seats and add-ons, Zendesk's pricing page says AI agents are "included in every Suite and Support plan" and billed per automated resolution, meaning a request resolved without escalation to a human. The same pricing page lists the rate: $1.50 per resolution on committed volume and $2.00 pay-as-you-go. Its plan comparison table includes 5 automated resolutions per agent a month on Support Team and Suite Team and 10 on Suite Professional (up to 10,000 a year), and Zendesk's help center describes that allowance as a monthly dollar amount per seat ($2, $5 or $10 depending on plan, capped at $5,000 a year, since 18 May 2026). The pricing page no longer lists a separate Advanced AI add-on; its featured add-ons are Copilot, the Workforce Engagement Bundle ($50/agent/mo) and Contact Center ($83/agent/mo). Resolutions beyond your allowance and commitment are what drive most of the AI spend, so check your own Zendesk billing page before you budget.

Pros: zero integration work; native to Agent Workspace; strong privacy posture; one vendor, one invoice. Cons: per-resolution pricing gets unpredictable at scale; answers are largely confined to what's in Zendesk; you don't choose the model. Privacy: handled by Zendesk's zero-retention endpoints, the safest of the four routes.

Can you add ChatGPT to Zendesk with a Marketplace app? (Route 2)

If you want raw ChatGPT behavior (your own prompts, your own OpenAI key) without engineering, the Marketplace has dozens of apps that drop a GPT panel into the agent sidebar. Most are bring-your-own-API-key: you paste an OpenAI key, and the app calls the ChatGPT API on your behalf.

Common examples:

• Triggers+ChatGPT: run ChatGPT workflows that categorize, translate, or parse tickets (Marketplace).
• ChatGPT Assistant by Canadesk: summarizes messages, analyzes sentiment, and recommends replies/macros from your own prompt (Marketplace).
• AI Agent Assist: uses the ChatGPT API with custom and predefined prompts for consistent, on-brand replies (Marketplace).

How to set up (typical flow):

1. Get an OpenAI API key from your OpenAI account (this is a separate, usage-based bill).
2. Install the app from the Zendesk Marketplace into your account.
3. Paste your API key and configure prompts/macros.
4. Use the panel from the ticket sidebar to summarize, draft, or translate.

Pros: cheap to start (you pay OpenAI usage, often a few dollars to dozens per month); full control over prompts and which GPT model you use; fast to install. Cons: quality and support vary widely between third-party developers; most are copilots for agents, not autonomous resolvers; you're stitching together two vendors. Privacy, important: when you bring your own OpenAI key, Zendesk's zero-retention setup does not cover that traffic. Your data goes to OpenAI under your account terms. Confirm the app's data handling and your own OpenAI data-retention settings before sending real customer data (eesel AI: Zendesk ChatGPT).

What do third-party AI agent layers add on top of Zendesk? (Route 3)

This is the middle ground between "use whatever Zendesk gives you" and "build it yourself." A class of tools connects to your Zendesk, ingests your knowledge from more places than the native agent reaches, lets you pick the underlying model, and either drafts replies or resolves tickets autonomously, while leaving Zendesk as your system of record.

Macha is one honest option here, and it's worth being clear about what that means. Macha is an AI agent layer on top of Zendesk, not a Zendesk replacement, and it runs the same way on Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom. Tickets, customers, and reporting stay in Zendesk; Macha adds the AI resolution layer over it. Teams reach for this route when they want answers grounded in sources beyond the Help Center (docs, Notion, Slack, past tickets), more control over model choice and behavior, and pricing structured around tickets rather than per-resolution overages. Other tools play in this space too (eesel, myaskai, and others), so evaluate a couple against your own ticket mix. The right fit depends on your channels, your knowledge sources, and how much autonomy you want to grant. You can see how the layer works on the Macha for Zendesk page, and our broader rundown of AI agents for Zendesk compares the category.

How to set up (typical flow):

1. Connect the tool to Zendesk via OAuth.
2. Point it at your knowledge sources (Help Center, websites, docs, internal wikis).
3. Configure agents/behaviors and choose a model.
4. Run in copilot/draft mode first, then enable autonomous resolution once you trust it.

Pros: broader knowledge grounding than native; model flexibility; keeps Zendesk as the source of truth; often more transparent pricing. Cons: another vendor in the stack; quality still depends on your documentation; you should pilot before trusting full autonomy. Privacy: depends on the vendor. Check for SOC 2, a DPA, data residency, and whether they (or their model providers) train on your data. Pricing note: Macha bills per ticket (one conversation, charged once however many messages it takes, whichever model you pick), not per resolution, so the bill never hangs on an outcome definition you don't control. It starts at $299 a month for up to 750 tickets, about $0.40 each, and the setup is done for you: the Macha team analyzes your past tickets, builds the knowledge base and the agent, and runs it in safe mode until the drafts are right, with a dedicated success manager who handles your changes afterwards, all included. You can start on the free trial ($50 of usage, no credit card required) to see real behavior on your own tickets before committing.

How do you connect Zendesk to the OpenAI API yourself? (Route 4)

If you have engineering resources, or just want a no/low-code automation, you can wire Zendesk's API to the OpenAI API directly. The two common variants:

• Automation platforms (no/low-code): Zapier, Make, n8n, or Pipedream all have Zendesk and OpenAI connectors. A typical recipe: new Zendesk ticket → send body to OpenAI for summary/classification → post the result back as an internal note or set a tag (Make, Pipedream).
• Fully custom: your developers call the Zendesk REST API (triggers/webhooks fire on ticket events) and the OpenAI API, with your own prompt and business logic in between.

Pros: total control over prompts, models, and routing; can encode bespoke business rules; no per-seat AI markup. Cons: you own the build and the maintenance: prompt drift, model upgrades, error handling, and rate limits all become your problem; no one to call when it breaks. Privacy: entirely on you. You're sending customer data to OpenAI under your own account, so set zero-data-retention where eligible, redact PII where you can, and document the data flow for GDPR (eesel AI: GDPR).

How do the four routes compare?

A reasonable default: start with Route 1 to see how far native AI gets you, add a Route 2 app if you just need agent-assist on the cheap, and graduate to Route 3 when you want resolutions grounded in knowledge beyond the Help Center or pricing that scales with actions rather than outcomes. Reserve Route 4 for genuinely custom workflows where you have engineering to maintain it.

Frequently asked questions

Does Zendesk use ChatGPT/OpenAI? Yes. Zendesk's native generative AI runs on a multi-LLM architecture that includes OpenAI alongside Google Gemini, AWS Bedrock, Microsoft Azure and others, under an official OpenAI partnership.

Is my customer data used to train OpenAI's models? Not for Zendesk's native AI. Zendesk says it uses zero-data-retention endpoints and that its LLM sub-processors cannot use your data to train their models. If you bring your own OpenAI key (Marketplace apps or the API), that traffic runs under your OpenAI terms, so set retention and PII handling yourself.

Can ChatGPT actually reply to and resolve Zendesk tickets automatically? Yes, via Zendesk's own AI agents (Route 1) or a third-party AI agent layer (Route 3). Most Marketplace apps (Route 2) assist human agents rather than resolve autonomously.

How much does AI in Zendesk cost? Native AI adds the Copilot add-on ($50/agent/mo, annual, Professional plans and above) plus per-resolution charges ($1.50 committed to $2.00 pay-as-you-go on Zendesk's pricing page), offset by a per-seat monthly allowance since 18 May 2026. Marketplace and API routes mostly cost your OpenAI usage. Verify current numbers on Zendesk's pricing page.

Do I need to be a developer? No for Routes 1–3. Route 4's automation-platform variant is low-code; the fully custom variant needs engineering.

Which route is right for your Zendesk?

There is no single "ChatGPT button" in Zendesk, but there are four solid ways to get LLM help into your helpdesk. The native features are the fastest and most privacy-clean start; Marketplace apps are the cheap agent-assist option; a third-party AI agent layer (Macha among them) gives you broader knowledge grounding and model choice while keeping Zendesk as your system of record; and the API route trades control for maintenance burden. Match the route to how much control, autonomy, and predictability you actually need, and pilot before you let any of them answer customers unsupervised.

If the AI-agent-layer route fits, you can start a free trial ($50 of usage, no credit card required). You connect Zendesk and the tools your answers come from, and the Macha team builds the first agent and runs it in safe mode, with drafts as internal notes, so nothing reaches a customer until you say so.

Sources: Zendesk pricing, Zendesk AI Data Use, Generative AI at Zendesk, Buying the Copilot add-on, Moving to automated resolutions, OpenAI × Zendesk, and Zendesk Marketplace listings linked above. AI features and pricing change quickly; figures checked 24 September 2026, so re-check before budgeting.