Skip to main content
Provision the standard PostHog views once PostHog is connected (2.1.4 Connect PostHog) — ask Pi with the posthog-setup skill, which reads /memory/posthog.md, discovers your lead event, creates the views, bundles them into a dashboard, and records the ids back to memory so re-runs stay idempotent (it won’t duplicate a view that already exists). Provision all five: session recording, tiered conversion actions, the Pages-with-AI-Traffic HogQL insight, a sitewide conversion funnel, and per-pillar funnels. Together they prove the two things that matter — AI-answer-engine traffic and pillar-level conversion — which is the whole attribution stance in Part 1 §1.10: GSC tells you what ranks, PostHog what those visitors did, and the join tells you which pillar to double down on. Then wire up attribution and validate it in the performance review.

​
What “standard” means: provision all five

  1. Session recording, a PostHog-side toggle. Turn it on in PostHog, then disable your ad-block to verify a session records. You set this one in PostHog directly, not through the skill.
  2. Conversion actions, in three tiers:
    • Soft: click demo, click sign up
    • Conversion: form submission, account setup
    • Hard: payment
  3. Pages with AI Traffic, the HogQL insight that isolates traffic referred by ChatGPT, Perplexity, Claude, and Gemini. Classic rank tracking can’t see this surface (see Part 1 §1.2).
  4. Sitewide Conversion Funnel, pageview to lead, to baseline conversion.
  5. Per-pillar funnels, the same funnel filtered to each content pillar (/blog, /uses, /features, and so on), so you can attribute conversion per pillar.

​
Step 1: Provision the insights, funnels, and dashboard

Set up my standard PostHog views. Discover my lead/conversion event first, then
create the "Pages with AI traffic" insight, a sitewide funnel to lead, and per-pillar funnels for
/blog, /uses, and /features. Bundle them into one dashboard and save the ids to /memory/posthog.md.
Don't duplicate anything that already exists.
Pi runs posthog-setup: it finds the lead event (confirming with you if it’s ambiguous, since a wrong event makes every funnel silently empty), creates each view, pins them to an "Acquisition & AI traffic" dashboard, and echoes the PostHog ids and URLs so you can eyeball them.

​
Step 2: Create the conversion actions

Create actions for the soft conversions (click demo, click sign up), the conversion events
(form submission), and the hard conversion (payment) — match each to my real buttons and
pages.
Pi creates the actions (matching on pageview URL or an autocapture click, e.g. a “Get a demo” button). Confirm the paths and labels against the live site before it writes.

​
The Pages-with-AI-Traffic insight (the SQL)

This is the insight that proves AI-answer-engine traffic. It’s provisioned as a HogQL data table. Reuse it exactly, and add a Number of Days variable for it to run: 
SELECT properties.$current_url AS page_url, count() AS event_count
FROM events
WHERE event = '$pageview'
  AND (
    properties.$referrer LIKE '%chatgpt.com%'
    OR properties.$referrer LIKE '%perplexity.ai%'
    OR properties.$referrer LIKE '%claude.ai%'
    OR properties.$referrer LIKE '%gemini.google.com%'
  )
  AND timestamp >= now() - INTERVAL {variables.number_of_days} DAY
GROUP BY page_url
ORDER BY event_count DESC
Good to know: when Pi provisions this via posthog-setup, it matches an expanded AI-source list, checking both properties.$referring_domain and properties.utm_source (either can be set), so it also catches chat.openai.com, copilot.microsoft.com, you.com, and so on. Same insight, wider net.
Good to know: on a PostHog account you already use, tell Pi to save to memory what you own and what not to touch, so a later session doesn’t clobber your existing PostHog setup.
🎥 Video: Conversion tracking on PostHog
Back to: 2.6 Analytics & review · Prerequisite: 2.1.4 Connect PostHog · Next: 2.6.2 Set up attribution · Theory: 1.10 Attribution