Investigating Replay

Investigating a session recording

When a user asks “what happened in this session?” or provides a recording/session ID to investigate, gather all relevant context in parallel rather than making them ask for each piece.

Available tools

Workflow

Step 1 — Get recording metadata and person profile

Start with the recording to get metadata and the person’s distinct ID:

posthog:session-recording-get
{
  "id": "<session_id>"
}

The recording id and the event $session_id are the same value. It selects the recording here and the same-session events in Step 2. The response includes distinct_id, person, start_time, end_time, duration, interaction counts, console error counts, and viewing status. Use the distinct_id to fetch the full person profile:

posthog:persons-retrieve
{
  "id": "<person_uuid_from_recording>"
}

Step 2 — Query same-session events

Use the recording id from Step 1 as the $session_id value. Get the timeline of what the user did during the session:

posthog:execute-sql
SELECT
    timestamp,
    event,
    properties.$current_url AS url,
    properties.$browser AS browser,
    properties.$os AS os,
    properties.$device_type AS device_type,
    properties.$screen_width AS screen_width
FROM events
WHERE $session_id = '<session_id>'
ORDER BY timestamp ASC
LIMIT 200

For sessions with many events, focus on the most informative ones:

posthog:execute-sql
SELECT
    timestamp,
    event,
    properties.$current_url AS url,
    if(event = '$exception', properties.$exception_values[1], null) AS exception_message,
    if(event = '$exception', properties.$exception_types[1], null) AS exception_type
FROM events
WHERE $session_id = '<session_id>'
    AND event IN ('$pageview', '$pageleave', '$autocapture', '$exception', '$rageclick')
ORDER BY timestamp ASC
LIMIT 100

No rows? Recover the event session ID

The recording id is the session ID. No rows means the session’s events were ingested without it. Find candidates from the person’s events in the recording window, padded by 100 seconds like the replay events query. person_id covers all of the person’s distinct IDs:

posthog:execute-sql
SELECT
    properties.$session_id AS session_id,
    count() AS event_count,
    min(timestamp) AS first_seen,
    max(timestamp) AS last_seen
FROM events
WHERE person_id = '<person_uuid>'
    AND timestamp >= toDateTime('<start_time>') - INTERVAL 100 SECOND
    AND timestamp <= toDateTime('<end_time>') + INTERVAL 100 SECOND
    AND properties.$session_id IS NOT NULL
GROUP BY session_id
ORDER BY event_count DESC
LIMIT 10

Continue only when one session ID clearly matches. Use it for the Step 2 and Step 3 queries only. The replay URL and all Replay Vision calls take the recording id.

Step 3 — Check for linked error tracking issues

If the recording has console errors or exceptions, find related error tracking issues:

posthog:execute-sql
SELECT DISTINCT
    properties.$exception_fingerprint AS fingerprint,
    properties.$exception_types[1] AS type,
    properties.$exception_values[1] AS message,
    count() AS occurrences
FROM events
WHERE $session_id = '<session_id>'
    AND event = '$exception'
GROUP BY fingerprint, type, message
ORDER BY occurrences DESC
LIMIT 10

If fingerprints are found, search for the corresponding error tracking issues to provide links and status:

posthog:query-error-tracking-issues-list
{
  "searchQuery": "<exception_type or message>"
}

Step 4 — Synthesize the investigation

Present the findings as a coherent narrative:

1. Who — person properties (name, email, country, plan, etc.)
2. What — sequence of pages visited and key actions taken
3. Problems — exceptions, console errors, rage clicks, and their frequency
4. Related issues — linked error tracking issues with their status (active/resolved)
5. Context — session duration, device/browser, activity score

Optional: AI summary via Replay Vision

If the user wants a deeper analysis without reading through events manually, offer a Replay Vision summary. Follow “check-then-scan” — don’t scan blindly, a scanner can only observe a given session once.

1. Check for an existing summary. A scheduled scanner may already have one:

   jsonCopy

   posthog:vision-observations-list
   {
     "session_id": "<session_id>"
   }

   The rows come back narrowed to id, session_id, status, summary_line and scanner_id. Look for one whose status is succeeded, then read it in full with vision-observations-retrieve for that id: its scanner_snapshot.scanner_type tells you whether it is a summarizer, and scanner_result.model_output carries title, summary, intent, outcome, friction_points and keywords. If you find one, you are done — no new scan needed.

2. Generate one with an inline scan. Pass this exact config: inline scans are keyed by a fingerprint of the whole config, so the config below reuses the same scanner row the player’s Summarize button uses on its built-in prompt, while a different prompt or length mints a separate scanner and a separate summary.

   jsonCopy

   posthog:vision-scanners-inline-scan-create
   {
     "session_ids": ["<session_id>"],
     "scanner_type": "summarizer",
     "prompt": "Summarize what the user did in this session: which pages they visited, what they tried to accomplish, and any notable moments like errors, confusion, or successful completions. Be concrete and don't speculate.",
     "scanner_config": { "length": "medium" }
   }

   Leave model out so the server default applies. Warn the user this is async and takes several minutes (rasterize + LLM). Nothing is scheduled and there is nothing to clean up: the scanner an inline scan mints never sweeps on its own. A 400 here usually means the organization has not approved AI data processing yet.

3. Read results[0].scan_outcome before polling. started means poll vision-observations-list (step 1) until the new observation reaches succeeded. already_scanned means a terminal observation already exists — read it via step 1, and if its status is failed or ineligible say so rather than polling. A null scan_id or skipped_quota means nothing ran: report the quota, do not poll.

The project already has a summarizer scanner

Use an inline scan for a one-off summary even then. Only reach for a saved scanner when the user wants that scanner’s own prompt rather than the built-in one:

posthog:vision-scanners-list
{
  "scanner_type": "summarizer"
}

Show the user the scanners (name + prompt), ask which to use, then run it against the session and poll vision-observations-list until the observation reaches succeeded:

posthog:vision-scanners-scan-session
{
  "id": "<scanner_id>",
  "session_id": "<session_id>"
}

Never create a scanner to answer a single question — vision-scanners-create leaves a scheduled sweep behind that the inline scan does not.

Tips

• Run steps 1-3 in parallel when possible — they’re independent queries.
• If the recording has very few events, the session was likely very short. Note this rather than suggesting something is broken.
• Console error count from the recording metadata is a good signal for whether to dig into exceptions. If it’s 0, skip step 3.
• The start_url from the recording tells you where the user’s journey began — use this to frame the narrative.
• If person is null on the recording, the user was anonymous. Person properties won’t be available, but events still are.

Related skills

• finding-sessions-to-watch — choose which sessions are worth investigating in the first place
• finding-replay-for-issue — start from an error tracking issue and find its linked recordings
• diagnosing-missing-recordings — when a recording that should exist doesn’t
• creating-replay-vision-scanners — automate this kind of watching as a scheduled scanner