PostHog will train AI models with your data (opted-in by default)

TEXT ANALYSIS PROTOCOL

TEXT START: "I really think we're on the verge of some of our best work through the next six months."

1. THE DISSECTION

PostHog is announcing it will train proprietary AI models on customer behavioral data—session replays, feature usage patterns, user flows—and wrapping this in a transparency performance. The "honest disclosure" framing is strategic, not ethical: it's a reputational inoculation. The company's core pitch is that they're building a product editor (not just a code tool), which requires converting user behavioral data into a proprietary training corpus. They're transparent because the alternative (burying it in a ToS update) carries more regulatory and competitive risk in an environment where technical users are increasingly data-sovereignty-conscious.

The "opt-out" versus "opt-in" framing is a direct admission: they know this is extraction, and they're asking users to bear the cost of refusal.

2. THE CORE FALLACY

PostHog assumes the data moat is durable. It is not. The thesis:

Session replay data, user behavioral flows, conversion patterns = a proprietary training corpus that produces differentiated AI features.

Under DT logic: This is a transitional arbitrage. As AI systems become more capable, the marginal value of task-specific proprietary training data declines relative to the cost of maintaining the infrastructure and trust to collect it. Synthetic behavioral data generation already exists. Multi-modal models trained on broader corpora will increasingly replicate what PostHog is spending user trust to build.

The moat isn't the data. The moat is the workflow lock-in built on top of that data. PostHog knows this—which is why they're framing this as "product editor" territory, not "better autocomplete." They're betting that embedding AI into the product lifecycle creates sufficient switching cost to survive commoditization.

That bet may work for 3-5 years. It is not durable under structural pressure.

3. HIDDEN ASSUMPTIONS

• Behavioral data remains irreplaceable. Not guaranteed. Synthetic data quality is rising, and foundation model generalization is compressing the value of domain-specific corpora.
• Users won't leave over this. Conditional. The opt-out mechanism requires admin access, which means most users in organizations will be trained-on without meaningful agency. This is a ticking trust bomb.
• EU regulatory distinction is stable. The EU cloud opt-out is framed as a legal accommodation. What happens when US competitive pressure and regulatory arbitrage make that distinction untenable?
• Competitors won't replicate the features with synthetic data. Unlikely. The features PostHog is building (synthetic user testing, replay analysis, conversion prediction) are not defensible at the model level—only at the workflow integration level.

4. SOCIAL FUNCTION

Classification: Transition Management / Self-Exoneration Theater

This is a company signaling: "We have chosen to become a data-extracting AI entity, and we want credit for telling you about it." The transparency is real. The framing is self-serving. The function is:

• Manage the reputational risk of becoming a data harvester
• Position the "opt-out" as a user choice rather than a corporate extraction
• Signal to investors that PostHog is building proprietary AI assets
• Preempt regulatory scrutiny by demonstrating "informed consent"

The audience is technically sophisticated users who can appreciate the candor and mistake it for ethical rigor. It is a sophisticated marketing operation dressed as a transparency manifesto.

5. THE VERDICT

PostHog is positioning itself as a data-owning Sovereign in a transitional economy. The transparency is real. The strategy is fragile.

The core value proposition—that behavioral data + proprietary training = defensible AI features—will face mounting compression from falling training costs, synthetic data quality improvements, and foundation model capabilities. PostHog has approximately 3-5 years of genuine moat before the features they're building are replicable without the extraction.

The "opt-in by default for US, opt-out for EU" architecture is not an ethics statement. It is regulatory arbitrage. PostHog is treating user data as a capital asset and constructing the governance framework accordingly.

For users: The opt-out exists, but the features you won't receive are precisely the ones PostHog is betting on as its competitive core. Opting out means using a degraded product. This is the prototype for how mass employment-equivalent data extraction will function under the new regime: not slavery, but a subscription model where you pay in behavioral data and receive a progressively differentiated product in return.

The verdict: PostHog is not evil. They are a clear-eyed transitional entity executing the Sovereign playbook at the mid-market level. The transparency is genuine. The long-term defensibility is not.