zoho
Does Zoho CRM Send Your Data to OpenAI? 2026 Privacy Guide
Does Zoho CRM data go to OpenAI? See which AI models Zia uses, how bring-your-own-key works, data center storage and GDPR checks before switching AI on.
5 Nov 2026 · 10 min read · Abhijeet Singh

On this page 8 sections
If you are about to switch on AI in Zoho CRM, the first question from your data protection lead is usually the same: does our Zoho CRM data go to OpenAI? As of November 2026 the answer depends on which model your admin turns on. Zoho's own model keeps processing inside Zoho's infrastructure. ChatGPT, Gemini, Claude and the other third-party models only receive data when an administrator deliberately connects them, and then that provider's terms apply.
This guide walks through which models Zoho CRM uses, how bring-your-own-key works, what actually gets sent, where your data sits by data center, and the GDPR and UK GDPR checks worth doing before anyone presses "Deploy".
Does Zoho CRM data go to OpenAI? The short answer
Not by default, and not without an admin action. Here is the picture based on Zoho's own documentation:
- Zia, Zoho's in-house model, is the native option. Zoho's help page for Zia Models describes it as operating "entirely within Zoho's secure infrastructure", so prompts and records processed by Zia are not sent to an outside AI company.
- Third-party models are opt-in. OpenAI (ChatGPT), Google Gemini, Anthropic and Cohere can be connected, but only an administrator can enable them, and each one has to be configured separately.
- Once connected, the provider's terms govern the data. Zoho's Zia privacy principles say that for these integrations the provider's terms cover data handling and training, and your organization must review and accept them.
- Zoho says it does not train its models on customer data. Its Zia privacy principles state that "Zoho does not use customer data to train or improve its AI models."
So the real question is not "does Zoho send our data to OpenAI?" but "has anyone in our org connected OpenAI, and do we have the paperwork to cover it?"
Which AI models Zoho CRM actually uses in 2026
Zia LLM, Zoho's own model
Zoho announced Zia LLM in July 2025. According to coverage of the launch in IT Brief UK, it was built in-house on NVIDIA's platform in three sizes (1.3, 2.6 and 7 billion parameters), and Zoho said its models are not trained on consumer data and that customer information is not retained. In the CRM, it appears under Setup, then Zia, then Models as the Zoho Hosted LLM, which Zoho's help page lists as available in all data centers.
Zoho's community announcement on switching between models says Record Assistant opens with the Zoho Hosted LLM by default, and users can switch to another active model from a dropdown if their admin has enabled one. For which of these AI features are worth enabling at all, see our practical review of Zia features.
Third-party models you can connect
Zoho's Zia Models help page lists these providers for Zoho CRM:
- ChatGPT (OpenAI)
- Gemini (Google)
- Anthropic (Claude)
- Cohere
- DeepSeek and SiliconFlow, which Zoho lists for its China data center only
Several can be active at the same time. Zoho's release information lists the multi-model feature on the Professional, Enterprise, Ultimate, CRM Plus and Zoho One editions.
Zia Agents follow the same pattern
If you build agents in Zia Agents (Agent Studio), the model choice works similarly. A June 2026 overview by Zoho partner Clixlogix describes two paths: Zoho-hosted models that need no API key, and bring-your-own-key access to OpenAI, Anthropic or Google Gemini. As always with agents, check which model each agent is configured to use, because one agent on an external model is enough to send data outside Zoho. If you are still deciding where agents should live, our comparison of Zia Agents versus n8n covers the trade-offs.
How bring-your-own-key (BYOK) works
BYOK means your company signs up with the AI provider directly, creates an API key in its own account, and pastes that key into Zoho. Zoho's overview of its OpenAI integration describes exactly this: users with an OpenAI account generate a secret key on the OpenAI platform and enter it in their Zoho account.
That has three practical consequences:
- You are the provider's customer, not Zoho. Billing, usage limits and the data processing agreement sit between your company and OpenAI (or Google, Anthropic, Cohere).
- The provider's API terms apply. OpenAI's policy page states that it does not train on inputs or outputs from its products for business users, and that organizations are opted out of data sharing by default. Anthropic's API documentation says retained data is never used for model training without your express permission. Zoho's own Models help page adds a note that Anthropic may retain interactions for model improvement unless opted out, so read the current terms for whichever provider you connect rather than relying on a summary.
- Retention is not zero by default. Providers keep some data for abuse monitoring or specific features. Anthropic's documentation, for example, describes zero data retention as something you request from its sales team rather than the standard setting.
There is also a second route. Zoho's announcement of Zoho Key Service (ZKS) describes a setup where Zoho provides access to models without your own API key and your org receives complimentary credits: 2,000 on Professional, 4,000 on Enterprise and 6,000 on Ultimate. Admins must enable ZKS and accept the applicable terms before use. The same announcement says ZKS gives access to OpenAI models, with Anthropic being rolled out in phases, so a Zoho-provided key still means a third-party model is processing the request. Confirm which model each feature calls.
What data is actually sent when a third-party model is on
This is the part most teams never check, and Zoho's documentation is reassuringly specific for the Models feature in the email composer and record tools:
- Models only processes the content in the email composer window, not the full email thread.
- No recipient information (to, cc, bcc) or attachments are sent to the language model.
- Suggestions, once copied into the composer, are not stored.
- Models processes fields individually rather than analyzing entire records or notes.
- Sensitive and protected field types are excluded, and Zoho states that none of the models process PHI-encrypted fields.
In plain terms: when a rep asks ChatGPT inside Zoho to rewrite an email, the draft text goes to OpenAI. The customer's whole history does not. But the draft itself can still contain personal data (a name, a deal value, a complaint), so it counts as a transfer of personal data and needs to be covered.
Where your Zoho CRM data is stored, by data center
Your CRM data lives in the Zoho data center your account was created in. Zoho's accounts help page lists separate data centers for the United States, Europe, Australia, Canada, Japan and Saudi Arabia, among others, and states that it only allows each account to store its data in one data center. Moving later means a migration request through support, and Zoho notes that a few services may not support data center migration.
For UK businesses there is an extra detail. Zoho announced a UK data center in September 2025, and in April 2026 told the trade press it was in the process of launching it in the second quarter of 2026. When we checked Zoho's public data center list on 4 November 2026, it still showed no UK location. If UK residency matters to you, confirm with Zoho whether the UK region is live for Zoho CRM and whether your org can move to it, rather than assuming.
The key point: the data center controls where your stored CRM records sit. It does not control where a third-party AI provider processes a prompt. Connect OpenAI from an EU-hosted Zoho account and the prompt is processed under your OpenAI account's settings. OpenAI announced data residency for business customers in several regions, including Europe, the UK, the US, Canada and Australia, on 25 November 2025, but it requires an Enterprise customer approved for advanced data controls and a project set up for that region.
The GDPR and UK GDPR angle
This is general information, not legal advice, so loop in whoever owns privacy in your company. The questions that usually come up:
Who is the processor?
Zoho acts as a processor for your CRM data under its data processing agreement. When you connect an AI provider with your own key, that provider becomes a processor (or sub-processor, depending on how your contracts are set up) for the prompts it receives. You need its data processing agreement in place, not just Zoho's.
Is there an international transfer?
If your Zoho account is in the EU data center but your AI provider processes data in the US, that is a transfer outside the EEA. Under GDPR and UK GDPR it needs a legal basis, typically the provider's certification under the EU-US Data Privacy Framework (and the UK Extension for UK data) or standard contractual clauses. Check the provider's current transfer mechanism and record it.
Do you need a DPIA?
If AI will process customer personal data at scale, or touch anything sensitive, a data protection impact assessment is sensible and in some cases required. It also forces the useful conversation about which features are actually needed.
Does your privacy notice cover it?
If customer data is processed by an AI provider, your privacy notice and records of processing should say so. Zoho's help page also notes that Zoho CRM does not record or process user information without consent and gives admins GDPR settings for processing and deletion requests, so align those settings with your notice.
The same checklist works for the CCPA in California and the Australian Privacy Act: know which vendor receives the data, have a contract that covers it, and tell people in your privacy notice.
What to check before you switch AI on in Zoho CRM
A short checklist we use as a starting point when reviewing AI settings for clients:
- Open Setup, then Zia, then Models. List every model that is enabled, and note whether each one uses ZKS or your own key.
- Decide on a default. If your privacy team is cautious, keep the Zoho Hosted LLM as the only active model and add external models later for specific use cases.
- Restrict by profile. Only admins can enable models, but you can limit which profiles use AI features. Start with a pilot group.
- Collect the paperwork. For each external provider: the DPA, the transfer mechanism, and its retention and training settings for API data.
- Review Zia Agents separately. Each agent has its own model setting, so audit them one by one.
- Keep sensitive data in protected fields. Zoho excludes sensitive and PHI-encrypted fields from model processing, so use encryption and field types properly rather than storing sensitive notes in plain text fields.
- Update your privacy notice and records of processing before go-live, not after.
For the wider picture of how Zia, Zoho MCP and external automation tools fit together, see the full guide to AI agents for Zoho CRM.
How this plays out in real implementations
In practice, the safest rollout is boring: start with Zoho's own model for summaries and drafting, prove the value with a small team, and only then connect an external model for a specific job where it clearly does better. When we set up Zoho CRM AI for clients, we usually begin with an audit of what is already switched on, because it is common to find an API key someone connected during a trial and forgot about. From there it is a short list of settings, contracts and a privacy notice update, not a months-long project.
If you are planning to switch on Zia, connect ChatGPT or Claude, or build Zia Agents and want the data protection side handled properly from day one, our Zoho CRM consulting and setup covers it. Book a call through the website and we can map out a setup that your privacy lead will sign off on.
Related resources
Keep building the automation map
Move from the guide into the services and proof pages connected to this topic.
Related services
Service
Explore Zoho CRM Consulting
Zoho CRM setup, custom modules, client scripts, Deluge functions, workflow rules, and cross-module automation.
Service
Explore AI Agent Development
AI agents for sales replies, lead memory, storyboard generation, data extraction, and structured automation outputs.
Service
Explore Custom CRM Implementation
Custom CRM architecture for sales pipelines, operations handoff, service workflows, invoicing, and reporting dashboards.
Further reading
Article
Read: Zoho CRM Zia AI Features in 2026: What's Worth Turning On
Every Zia AI feature in Zoho CRM sorted into enable now, pilot first, or skip — with links to the official docs and what real SMB rollouts actually show.
Article
Read: Do You Need Zoho CRM Enterprise for AI Features in 2026?
Do you need Zoho CRM Enterprise AI features? Here's which Zia and Zia Agents tools each edition gets in 2026, the USD cost by plan, and when to upgrade.
Article
Read: Connect Zoho CRM to Claude or ChatGPT with Zoho MCP (2026)
Connect Zoho CRM to Claude or ChatGPT with Zoho MCP: step-by-step setup, the right tools and auth mode, and how to fix the data center region error.
Proof pages
Case study
See case study: Custom Zoho CRM with Auto-Pipeline for a Freight Forwarder
A freight CRM that carries an enquiry through contacts, accounts, deals, quotes, jobs, cost sheets, and invoicing — with client scripts and Deluge automation removing the re-typing between every stage.
Case study
See case study: Complete Zoho Stack for a B2B Packaging Manufacturer
One Zoho operating system for a B2B packaging manufacturer — sales, service, inventory, accounting, ticketing, and analytics — replacing the legacy Excel and Salesforce files the business used to run on.