How to analyze PostHog session recordings for repeatable issues

Teams that already analyze PostHog session recordings usually have plenty of captured data and too little time to act on it. The recordings pile up, someone watches a handful, and most of the evidence never turns into a decision.

The good news is that PostHog ships more replay tooling than most teams use. Before you consider migrating or bolting on another vendor, it is worth knowing exactly what the built-in filters, collections, and Max AI features can do - and where a dedicated analysis layer starts to earn its place.

This guide is based on PostHog's official session replay documentation, reviewed on July 24, 2026.

What PostHog replay already gives you

PostHog Session Replay is a playback product first: it records the DOM, network activity, and console logs so you can diagnose UI issues, support customers, and get context on nuanced behavior. Inside the player you can inspect the DOM, open a network waterfall, view a page performance score, jump into a heatmap built from the recording's own HTML, and capture screenshots or short GIF/MP4 clips to attach to a bug in Jira or Linear.

The part most teams underuse is everything that happens around a single recording:

• Saved filters. Build a dynamic list by date, active duration, activity counts, events, properties, console logs, or feature flags, then save it. New replays that match are added automatically, and the filter is shared with everyone on the project through the Shared filters tab.
• Activity-count filters. Filter recordings by clicks, key presses, or mouse activity with numeric operators. A click_count > 0 filter strips out bot and scanner traffic; keypress_count > 10 finds sessions where users typed a lot.
• Collections. Where saved filters are dynamic, collections are permanent lists you curate and share for a specific investigation.
• Built-in collections. PostHog auto-surfaces recordings into ready-made lists, including Watch history, Recordings with comments, Expiring soon, and a Frustration signals collection of sessions with multiple rage clicks or exceptions in the last 7 days.

If your team is watching recordings one at a time from the default list, these features alone will save hours.

What Max AI adds

PostHog's AI assistant, Max, layers natural-language search and summarization on top of replay. There are two distinct capabilities worth separating.

Finding replays. Instead of building filters by hand, you describe the behavior - "users who rage clicked on the pricing page" or "enterprise users who encountered an error on the dashboard" - and Max translates that into replay filters and returns matching sessions. It understands replay-specific signals: rage clicks, dead clicks, error events, navigation patterns, and any person property you have captured.

Summarizing sessions. Max reads the structured event stream of a recording - every page view, click, input, scroll, error, and custom event - and produces a human-readable summary of key actions, behavioral patterns, errors, and the session outcome. It analyzes the event data rather than watching the video, so summaries stay fast even on long sessions, and you can summarize a batch at once to spot what is common. In the player, AI summary segment markers appear on the timeline as green (success) or red (failure) overlays you can click to jump to the moment.

Two constraints matter. Session summaries are PostHog Cloud only - they are not available on self-hosted deployments, though Max itself can run self-hosted with a configured ANTHROPIC_API_KEY. And summarization is something you ask for on a chosen set of sessions; it is not a background job that watches all your traffic and tells you what to fix next.

Where the built-in workflow runs out

PostHog's replay AI is genuinely useful for triage: point Max at a question, get a shortlist, read a few summaries, move on. The friction shows up when you want analysis rather than answers to a query.

Job to be done	PostHog replay + Max	A dedicated AI analysis layer
Find sessions matching a described behavior	Strong - natural-language filters	Also supported
Summarize a chosen batch of sessions	Strong - fast, event-stream based	Also supported
Continuously scan all sessions in the background	Not the model - you ask per query	Core function
Group the same failure across many sessions into one issue	Manual - you infer it from summaries	Automatic grouping
Estimate how many users an issue affected	You assemble it from filters	Reported per issue
Rank issues by impact for a weekly queue	Not built in	Core function

The dividing line is cross-session grouping. A summary tells you what happened in one recording, or what a batch had in common when you asked. It does not, on its own, maintain a ranked, deduplicated list of the repeated problems across everything you record. When your question shifts from "show me sessions like X" to "what are the top issues this week and who did they hit," you are describing a different product.

When to add an AI analysis layer on top

You do not have to leave PostHog to get that. A complementary analysis layer makes sense when:

• You record enough traffic that manual triage never catches up.
• You need repeated rage clicks, dead clicks, errors, and drop-offs grouped into named issues, not re-derived each time.
• You want each finding tied to a funnel step, goal, or revenue figure so product and engineering can prioritize.
• You need a short weekly queue the team actually trusts, with the original replay evidence attached.

Flowsery is built for exactly that layer. It combines privacy-first web analytics, recordings, funnels, journeys, goals, sources, and revenue, and its AI session layer groups repeated friction into ranked issues you can evaluate against business context - complementing the raw capture and query workflow you already have.

Frequently asked questions

Does PostHog use AI to analyze session recordings?

Yes. PostHog's assistant, Max, finds replays from natural-language descriptions and summarizes sessions by reading their event stream. Summarization runs on PostHog Cloud only, while Max itself can also run on self-hosted instances with an ANTHROPIC_API_KEY.

Do PostHog AI summaries watch the video?

No. Max analyzes the structured event data - page views, clicks, inputs, scrolls, errors, and custom events - rather than the rendered video. That is why summaries stay fast even for long sessions and can be produced in bulk.

Can I find specific sessions without building filters manually?

Yes. Describe the behavior you want - such as sessions with rage clicks on the pricing page or users who dropped off during checkout - and Max converts it into replay filters. You can also save filters and reuse built-in collections like Frustration signals.

Do I need to migrate off PostHog to get AI issue analysis?

No. You can get more from PostHog's own filters, collections, and Max features first, then add a dedicated analysis layer such as Flowsery when you need repeated issues grouped, ranked by impact, and tied to funnels or revenue.

What is the difference between a saved filter and a collection in PostHog?

A saved filter is a dynamic, shared list that automatically adds new replays matching its criteria. A collection is a permanent, curated list you build and share for a specific investigation.

Turn PostHog-style replay evidence into ranked issues with Flowsery - start free and review the sessions that actually matter.

Sources: PostHog Session Replay docs, Watch recordings, Find replays with PostHog AI, and Summarize sessions with PostHog AI. Checked July 24, 2026.