PostHog AI FAQ

What access does PostHog AI have to my data?

While working with PostHog AI, you will certainly share some information with it. Data analysis is a core feature. When doing research, PostHog AI can read the following data in your project:

Events, persons, sessions, groups
Actions and cohorts
Event and property definitions
Data warehouse schema
Saved insights and dashboards

PostHog AI respects your project's access controls. You can only access insights and dashboards you have permission to view.

Based on that context, PostHog AI comes up with queries, runs them, and summarizes the results. Those results can be as high-level as a single number, or as detailed as a list of users with all their properties.

Is my data shared with third parties?

We use third-party providers to power PostHog AI, so messages and all data accessed by PostHog AI (see list above) are passed through them.

Our list of data subprocessors includes:

Anthropic, OpenAI, and Alphabet – responsible for the core of PostHog AI's functionality, have access to user messages and all data types.
Azure – responsible for semantic search across PostHog, has access to actions and cohorts (including their filters)

Do the AI providers train on my data? Does PostHog?

Third-party providers, no. DPA §2.1.3 prohibits subprocessors from using your data to fine-tune, train, or develop their models, and that covers chat content as well as event data. PostHog does train its own models on anonymized data. That's a separate setting, "Enable AI training on anonymized data", and you can disable it.

Where does AI inference happen? Does EU Cloud data leave the EU?

Our AI subprocessors' primary location can be found next to the applicable subprocessor on posthog.com/subprocessors. Those subprocessors have their own subprocessors, which may sit outside the USA. The Service Provider for AI Features entries on our Trust Center link to each of those lists.

More broadly, for EU Cloud: we have subprocessors outside the EU, and PostHog staff may access data from outside the EU, including for support. Any transfer mechanism in the DPA may apply, whether that's an adequacy decision, the Data Privacy Framework (DPF), or EU Standard Contractual Clauses (SCCs).

Is there zero data retention with the AI providers?

We can't yet promise zero data retention (ZDR) on our AI features, because the underlying model providers don't consistently offer it. Retention windows are those providers' defaults.

How can I disable PostHog AI?

With admin permissions for your organization, you can disable PostHog AI and all AI data analysis in your organization settings.

Go to Settings > Organization > General
Find the PostHog AI data analysis section
Toggle the switch to "off"

When disabled, PostHog AI won't be available and no data will be shared with AI providers. This can be reversed at any time.

Can I hide PostHog AI?

There currently isn't a toggle to hide AI features altogether. However, in any case, only org admins can approve AI features in the organization – so there's no risk of data being sent to any AI provider without explicit consent.

Is my chat history retained if I disable and re-enable PostHog AI?

Yes, PostHog AI retains your chat history.

Can I self-host PostHog AI and use my own models?

No, PostHog AI is only available in the cloud-hosted PostHog app, which uses third-party LLM providers. PostHog automatically selects the most efficient model for each AI feature.

What is PostHog's approach to AI sustainability?

For our enterprise friends: PostHog AI runs on third-party LLM providers (Anthropic, OpenAI, and Alphabet), so the environmental footprint of running PostHog AI largely tracks the sustainability commitments of those providers rather than PostHog's own infrastructure.