Describe what happened. Find the sessions that show it.

You cannot filter for "gave up"

Support sends you a message: "I tried to pay three times and gave up." It has no order number and no time. You filter to the checkout page, get 600 sessions and start to watch them at double speed.

Filters find sessions by facts that you can name: an event, a page, a country. But some behaviour has no single event. Someone typed a postcode that the form did not accept, and left. No event says that.

Ask is for that behaviour. Describe what you would see on the screen. Write "Clicked Save on the billing page again and again", not "Frustrated users". Ask gives you the sessions that look like it.

One sentence in. Ranked sessions out.

1. 01

   Describe it.

   Write one line in the words of your product: button text, page names, error text. Your date range and filters make the set of sessions smaller first.

2. 02

   Sequential judges recent sessions.

   For each session that it checks, it reads a short summary of the events, not the recording. It judges how likely it is that the session shows what you described.

3. 03

   Read the ranked list.

   Only likely matches come back, strongest first. Each one has a confidence. In our tour: "matches · 91%", "probably matches · 77%", "possibly matches · 64%".

4. 04

   Go to the moment.

   When Sequential can pick the moment, the row shows it, for example "at 0:54 — support_opened". "Watch from here" opens the replay just before that moment.

An empty answer tells you why

"No results" usually tells you nothing. Did Ask find nothing, or did it check nothing?

Every Ask answer says how many sessions it checked, out of how many. When some sessions came close, or had too few events to judge, the answer says that too. So you know if you must search more or narrow the range.

Do you need the number first? Start from a finding. Sequential counts a finding across sessions.

From 38 people to the one session that shows it

Example: Northlight Outfitters, our fictional demo shop.

Sequential has raised a finding on the shipping page: a request that fails. It hit 38 people. The head of product wants to see it happen before she approves the work.

She types: "typed a postcode, got an error, and left at the shipping step".

The top result says "matches · 88%", at 2:14 — "Enter a valid postcode". She clicks "Watch from here". A customer types a postcode with a space, the form rejects it two times, and the customer leaves.

She puts the search link in the team thread. When a teammate opens it, the same search runs with the date range and filters of that teammate.

Your agent finds the sessions. You decide.

Ask the agent that you already use to find the sessions: Claude Code, Cursor, Codex, Claude Desktop or any MCP client. It gets each match back with a replay link and, when Sequential picked one, the moment. For a click or a form, the agent does not get the label text.

At Northlight, her agent takes the strongest matches and opens WEB-231 in the team’s tracker, with the replay links. Sequential gave the evidence. The agent filed the ticket. A person decides what ships.

Good to know

• Ask finds sessions. It does not count them. For "how many people did this", use Funnels, or start from a finding.
• Each run checks up to 200 sessions, from the most recently active in your range and filters. The "Search more" button checks the next batch of up to 200.
• The percentage is a confidence, not measured accuracy. Weaker results say so: "possibly matches". A moment shows only when Sequential can pick one.
• Ask reads events, not recordings. So it also finds sessions that have no recording. A session needs at least 3 events. You need a recording only to watch the replay.
• Ask has its own daily allowance for each project. Background judging does not use it. When the allowance is used up, Ask tells you when it resets. If you ask the same question about the same sessions within one hour, Sequential gives the saved answer. It does not use the allowance.
• Agents have their own allowance, with a lower cap for each token. An agent searches the last 7 days by default. The hosted preview keeps 7 days of data by default. An agent judges up to 50 sessions in each call and cannot apply filters. If its client approves read-only tools automatically, the agent can search and not ask you first. A "Read everything, no Ask" token stops that.
• What leaves for the answer. Your question and an event summary of each checked session go to the third-party service that judges sessions. Sequential does not send the user id. It shortens long text but does not scan it for personal data, so an email in a button label or an error message goes as it is. Sequential keeps questions for 180 days, or for 400 days if you mark a result right or wrong.

Questions