FullSession

View saved version →

---
name: analyze-funnel
description: How to analyze a conversion funnel in FullSession, find the biggest drop-off, see who dropped and why, compare audiences, and get Lift AI recommendations. Use when the user asks about conversion, drop-off, checkout or signup completion, or a specific funnel.
---

# Analyzing a funnel

## 1. Find the funnel

`list_funnels` returns the site's funnels with their step names. Match the user's wording ("checkout", "signup") to a funnel name; if several fit, ask. If no funnel exists for what they describe, say so: the user can create it in the FullSession app, and `generate_funnel_steps_from_prompt` can draft its steps.

## 2. Read the funnel

`get_funnel_analysis` returns, per step, how many sessions (or users) reached it, the drop-off from the previous step, and the median time between steps, plus the overall conversion rate.

- Lead with the **biggest drop** (the step pair losing the most people) and the overall conversion rate.
- Say the unit: funnels count sessions unless they count unique users.
- If the funnel says it is **processing**, it was just created or changed; ask the user to try again in a few minutes.

## 3. See who dropped and why

`get_funnel_sessions` with `outcome: "dropped"` and the step **before** the drop returns the sessions that reached that step but never the next, with replay links. Offer 2–3 replays.

To learn why, look for patterns: run `get_session` on a few dropped sessions, or check the drop step's page with `compare_pages_by_frustration` / `get_page_heatmap` (see `investigate-friction`).

## 4. Compare audiences

- By device, country, browser or OS: `get_funnel_analysis` with `breakdown`. A drop that exists only on mobile is a much clearer lead.
- By saved segment: `get_funnel_analysis` with `segments` (from `get_segments`), for example new vs returning visitors.

Breakdown and segment comparison can't be combined in one call.

## 5. Recommendations

If the site uses Lift AI, `list_ai_goals` finds goals built on funnels, and `get_ai_recommendations` returns ranked issues with the predicted lift from fixing each. `get_recommendation_sessions` gives the sessions behind a recommendation. Present the top 2–3 with their predicted impact.

## Notes

- Compare like with like: when comparing periods, use the same length of time (`get_trend` with `comparePreviousPeriod` works for overall traffic).
- Small numbers are noisy. If a step has only a handful of sessions, say the result isn't reliable yet.

View saved version →

---
name: get-started
description: How to answer questions about a website's user behavior with FullSession (session replays, funnels, heatmaps, frustration signals, Lift AI recommendations and dashboards). Use whenever the user asks about their site's visitors, conversions, drop-offs, rage or dead clicks, errors, or wants to find or watch specific sessions.
---

# Using FullSession

FullSession records how visitors use a website: every session (with a replay), clicks and frustration signals, funnels, heatmaps, feedback, and Lift AI recommendations. The tools read the connected account's own data.

## Always start by knowing the site

- If the user hasn't named a site, call `list_sites`. If there is exactly one site, use it without asking. If there are several, ask which one, or match the name or URL they mentioned.
- Pass `siteId` explicitly to every tool after that.

## Time ranges

- Default to the last 7 days unless the user says otherwise. Say which range you used.
- Pass dates the way the user means them ("this week", "since January"), as ISO dates. Most tools accept ranges up to 365 days.
- When an answer is empty, say so plainly and suggest a wider range before concluding there is no data.

## Pick the right tool

| The user wants… | Use |
|---|---|
| A quick health check of a site | `get_site_overview` |
| Pages where users struggle | `compare_pages_by_frustration`, then `get_page_heatmap` (clicks) on a page |
| What's wrong with one element | `get_element_interactions` |
| Where a funnel loses people | `list_funnels`, then `get_funnel_analysis` |
| Who dropped off, to watch them | `get_funnel_sessions` |
| Sessions matching criteria | `search_sessions` |
| What happened in one session | `get_session` |
| One visitor's history | `get_user_sessions` (by userId or email) |
| A number, trend or breakdown | `get_metric`, `get_trend`, `get_breakdown`, `get_page_ranking` |
| What to fix to convert more | `list_ai_goals`, then `get_ai_recommendations` (Lift AI) |
| Saved dashboards and charts | `list_dashboards`, `get_dashboard_data`, `list_cards` |

For deeper workflows, see the `investigate-friction` and `analyze-funnel` skills.

## Presenting results

- Some tools show the user an interactive card (funnel chart, page ranking, session list, trend chart). When a card is shown, don't repeat its data as a table. Give the key insight in a few sentences and suggest the next step.
- Link to evidence. Session results include replay links; offer the most relevant 2–3 rather than listing everything.
- Be precise about units: funnels count sessions unless they're set to count unique users, and the tools say which.

## Visitor privacy

- Session lists show visitors' emails and names masked (e.g. `j***@acme.com`). This is intentional.
- When the user asks about a specific visitor or session, `get_session` and `get_user_sessions` return the full identity. Use them only when the question is about that person.
- Don't try to reconstruct masked identities from other fields.

## Things to know

- A funnel that was just created or had its steps changed is "processing" for a few minutes; its numbers are unavailable until it finishes.
- `generate_segment_from_prompt` and `generate_funnel_steps_from_prompt` only draft definitions with an AI model; they save nothing.
- These tools only read data. Segments, funnels, heatmaps, dashboards and charts are created and edited in the FullSession app; the `generate_*` tools can draft a definition for the user to paste there.

View saved version →

---
name: investigate-friction
description: Step-by-step investigation of where and why visitors struggle on a site, from frustrated pages down to the exact element and the session replays that prove it. Use when the user asks about rage clicks, dead clicks, error clicks, broken buttons, confusing pages, or "where are users frustrated?".
---

# Investigating friction

Work from broad to specific, and stop as soon as the user's question is answered. Each step narrows the evidence.

## 1. Find the pages

`compare_pages_by_frustration` ranks every page by one signal:
- `rageClicks`: repeated fast clicks, usually something that looks clickable but doesn't respond, or is slow.
- `deadClicks`: clicks that change nothing, often on non-interactive text or images that look like links.
- `errorClicks`: clicks followed by a JavaScript error, usually a real bug.

If the user didn't name a signal, start with rage clicks and mention the other two. Judge a page by **affected sessions relative to page views**, not raw clicks: 200 rage clicks from one session is one frustrated visitor, not a site-wide problem. Say when that's the case.

## 2. Find the element

For the page that matters, call `get_page_heatmap` with `type: "clicks"`. It lists elements ranked by frustration, with the visible text and a selector. Name elements by their visible text ("the 'Apply coupon' button"), not by selector.

To see how far people read or where the mouse goes, use `type: "scroll"` or `type: "movement"`.

## 3. Get the proof

`get_element_interactions` with the element's selector returns how many sessions interacted with it and the **sessions where it frustrated users**, with replay links. Offer 2–3 replays as evidence.

For a broader view, `search_sessions` with filters such as `hasRageClicks`, `hasDeadClicks`, `hasErrorClicks`, `hasJsErrors` or `hasNetworkErrors` finds sessions with those signals across the site.

## 4. Explain what happened

Use `get_session` on one representative session to read its timeline (clicks, errors, network failures, idle time) and explain in plain words what the visitor tried to do and what went wrong.

## 5. Suggest a fix

Tie the suggestion to the evidence:
- Dead clicks on text or images → make them clickable, or stop styling them like links.
- Rage clicks on a real button → check that it responds and shows loading state; look for slow network calls in the session.
- Error clicks or JS errors → name the error message and page so a developer can reproduce it.
- Form abandonment → `get_form_analytics` for completion rates.

Quantify the impact ("affects about 12% of sessions on /checkout") and suggest watching the replays before changing anything.

## Notes

- Device matters: if a problem looks mobile-only, check with `get_page_heatmap`'s `device` filter or `get_breakdown` by device.
- If nothing stands out, say so. Low frustration is a valid answer.