PostHog 现已接入 Vercel 的 v0

We've released a brand new PostHog × Vercel integration. That integration made the setup intentionally boring. Flags and experiments are defined in PostHog, synced into Vercel’s native Flags system, credentials are handled automatically, and your app simply consumes them via the Flags SDK.

That integration focuses on how flags and experiments run in production. Now, we’re extending the same product context into Vercel's v0 via MCP, so it’s available while you’re building.

What you can do with PostHog in v0

v0 can build directly on real codebases, enabling teams to ship production-ready apps and agents rather than prototypes or toy projects.

With PostHog as a partner, v0 doesn’t just generate code against your repository – it can do so with an understanding of how your product actually behaves in production. It has access to experiment results, user behavior patterns, feature flag state, and error impact, which means the software you generate reflects real product data rather than assumptions about how the product should work.

There's so much you can do with PostHog in v0 that we asked our Growth team what types of prompts they'd start with.

Act based on experiment results

When iterating on onboarding or activation flows, existing experiments often already contain the answer to “which direction should we take.” With PostHog connected, v0 can reference experiment data and use it as context when generating UI, so new designs reflect what has already been tested in production.

Suggested prompt:

Based on the onboarding experiment testing long vs short signup, show me the metrics and then generate a signup flow aligned with the winning variant.

Insights from our Growth team: You can use this when redesigning signup or pricing pages. For example, if a shorter signup form improved completion rate but reduced activation, that tells you friction wasn’t the only issue. You might keep the shorter form but strengthen the first-run experience, add clearer value framing, or tailor the flow by acquisition channel instead of treating all users the same.

Turn ideas into multivariate flags

Product changes are rarely binary. v0 can help turn a concrete idea into a properly structured multivariate feature flag that matches PostHog’s experiment model, making it ready to roll out and measure without additional setup.

Suggested prompt:

Create a multivariate feature flag in PostHog for our new ‘Quick Actions’ button. Variant A is yellow, variant B is green.

Insights from our Growth team: Use this for things like CTA color, button placement, or dashboard layout variations. Instead of debating whether a brighter button “feels more clickable,” you can ship both versions behind a multivariate flag and measure downstream impact. Not just clicks, but whether users actually complete the next meaningful action.

Ask product questions and act on the answers

Understanding what users do often comes down to a few focused questions. v0 can run trends, funnels, and HogQL queries in PostHog and surface the results directly, so decisions about flows or copy are informed by actual usage rather than assumptions.

Suggested prompt:

Show me a funnel from page view to signup, broken down by referral source.

Insights from our Growth team: You can use this when evaluating campaign traffic. If paid users convert differently from organic users, you could generate different landing page variants tailored to intent. It’s also helpful before rewriting homepage copy – if most drop-offs happen before pricing is even viewed, the issue likely isn’t the pricing table.

Find drop-offs and close the loop

Identifying where users drop off is only useful if it leads to follow-up. With PostHog in v0, you can move directly from analysis to targeted feedback, without stitching together multiple tools.

Suggested prompt:

Find my worst-performing funnel, then create a survey targeting users who dropped off asking what went wrong.

Insights from our Growth team: You can use this for onboarding steps with unexplained friction. If users consistently abandon at, for example, “Connect your first integration,” you can automatically trigger a short survey for that segment and ask whether it’s confusion, missing documentation, or lack of perceived value. That feedback can immediately shape the next iteration.

Create and summarize surveys

Qualitative feedback is most useful when patterns are clear. v0 can help create surveys and then group responses by theme once data starts coming in, making it easier to move from raw feedback to concrete next steps.

Suggested prompt:

Create an open-text survey asking users: 'What's the one thing you'd change about our product?' and target it to users who've been active for at least 30 days.

Insights from our Growth team: Try to use this after launches or major UI changes. Instead of manually reading hundreds of open-text responses, look for recurring themes like “navigation confusion” or “missing export options.” That will make it easier to prioritize fixes based on frequency and impact rather than on the loudest individual comment. Once survey responses come in, use a follow-up prompt to pull survey stats and identify the most common themes.

Debug anticipating user impact

Errors are easier to prioritize when they’re tied to user impact. v0 can query error data from PostHog so investigation starts with which issues affect the most users, not just which ones appeared most recently.

Suggested prompt:

Show me the top errors affecting the most users this week, and any new errors that appeared after yesterday’s deploy.

Insights from our Growth team: This should be useful after releases. If a new deploy introduced an error affecting, let's say, 15% of active users in a critical flow, that takes priority over a rare edge-case bug. It will also help you connect technical issues to real user journeys – for example, whether the error blocks upgrade, onboarding, or core usage.

Keep LLM costs visible

When building AI features, cost trends matter as much as functionality. v0 can surface model-level spend from PostHog so usage decisions are based on actual data rather than surprises at the end of the month.

Suggested prompt:

Which LLM model is costing me the most this week, and how is the spend trending?

Insights from our Growth team: This comes handy when testing different models for AI-powered features. If GPT-4-level performance doesn’t meaningfully improve user outcomes compared to a cheaper model, that’s an immediate optimization opportunity.

Product data is most useful when it shows up at the moment decisions are made. With PostHog available in v0, you can build agents, features, and experiments with real context in mind instead of assumptions. Check it out: v0.app