Skip to main content
You ran a strong discovery call on Tuesday. By Friday, you can recall three things from it. By the time someone needs the quote for next quarter’s roadmap argument, the call has dissolved into a vague impression. The conversation was valuable. The shape that would have kept it valuable wasn’t there. A customer research workflow is that shape. This guide is a specific instance of the workflows pattern. Same three pieces, applied to discovery, usability, and win/loss calls.

​
What the artifact looks like

Every call you run through this workflow ends with the same kind of structured output. A worked example:
Same shape, every call. That’s the property that makes 20 of these aggregate into something queryable when they land in a Notion table, an Airtable base, or a research repo.

​
The template that produces it

The artifact above comes from running this prompt as a pre-seeded task in Earmark:
Three things in this prompt are load-bearing. “Use the user’s words where possible” stops the model from smoothing distinctive language into corporate paraphrase. Verbatim is the asset. Paraphrase reads fine and aggregates into nothing. Severity tags with an explicit “Unknown” option stops the model from rating everything Critical. Without this instruction, every pain point looks like a five-alarm fire and the artifact is useless for triage. With it, the team gets something they can actually act on. Separation of request from proposed solution keeps you from chasing the feature the user named when the real problem is upstream. “I want a bulk export toggle” might be the right ask, or the right ask might be “I want closing the books not to take three days.” The template makes you capture both layers. The rest of the wording is up to you. The headings are the load-bearing part.

​
Save it as a workspace template

1

Run the closest built-in template on a real call

The Client Call template is the natural starting point. Add it as a task from the template library and run a real customer conversation through it.
2

Rework it in the Composer until the shape is right

Open the artifact in fullscreen. In the Composer, iterate the prompt toward the shape above — ask for severity tags, quoted verbatim, separated requests, open questions. Watch the preview reshape until the artifact lands.
3

Save with Workspace visibility

Open the Composer menu and choose Save as template. Set visibility to Workspace so the rest of the team produces the same shape. See Custom templates for who can edit and share.
A prompt proven on a real meeting transfers to the next real meeting better than one written from scratch. Tune in the Composer, then save.

​
Run it on a single call

1

Pre-seed the meeting

Open the call in Earmark before it starts and add your saved Customer Research template as a task. While you’re there, click the project chip next to the title and add the call to that customer’s project. See Before a meeting.
2

Add context

Use the Customize context dialog to note segment, stage, and what you want to learn. Sharper context produces sharper extraction. For a repeat conversation, ask the customer’s project “What open questions are left from our earlier calls?” and carry the answer into the context. That’s what the Open questions section is for.
3

Let it run live

Pain points, quotes, and open questions stream in as the conversation goes. You don’t have to babysit it.
4

Curate for five minutes after

Fix misattributed quotes. Re-tag any severity the model called wrong. Promote anything important out of Notes into a real category. This is the entire human-in-the-loop step.
5

Send it to the destination

Copy the structured artifact into the tool where your team’s customer record actually lives — a Notion table, an Airtable base, a research repo, or your CRM. See Integrations for direct copy targets if you also want Linear tickets from the same artifact.

​
When the artifact is wrong

The model gets things wrong. The workflow assumes a few minutes of curation, not none.
  • Everything is Critical. Re-tag in the artifact. After a few weeks of consistent curation, the model calibrates to the severity bar your team actually uses.
  • A quote sounds smoothed-out. The model paraphrased instead of quoted. Replace from the transcript before sending to the destination — the verbatim is the asset, the paraphrase isn’t.
  • Pain and feature request collapsed into one bullet. Promote each to its own line. If this happens on every call, sharpen the separation instruction in the template.
  • Severity drift across the team. Three people each tweak the prompt for their own calls. Within a month the outputs no longer aggregate. Lock the template at the workspace level and update it deliberately.

​
Across many calls

A single call is a data point. The research value is in the pattern across twenty. Two features carry the workflow past the single call. A project per customer. Keep every call with one customer in a project, even when the titles don’t match. Add calls before they start, while they’re live, or after. For segment-level research, add the same calls to a second project too, like “Healthcare, Q3”; a meeting can be in several projects. Chat for the synthesis. From the project page, ask Chat the questions that used to take an afternoon of rereading:
  • “What has Acme said about the export limit across our calls?”
  • “What are the most common pain points in this project, with quotes?”
  • “Which feature requests came up more than once?”
Answers cite each call with a source chip. Hover to see the excerpt; click to jump to that moment in the meetings you recorded. Treat the chips the way you treat the artifact: verbatim is the asset, so check a quote against its source before it goes into a roadmap argument. Name the period for older research (“since January”). Share it with the people who need the signal. Add product and design to the project as Collaborators. They can ask their own questions across the customer’s calls without you forwarding notes. They see each call’s title, who recorded it, and when, but can’t open your meetings; Chat shows them short excerpts as sources. The rest still applies:
  • Find a specific call by title with the command menu (Cmd+K / Ctrl+K).
  • Aggregate in the destination tool. Twenty artifacts in the same Notion table is one filter away from a quarterly view. This is the payoff for the stable template shape.
  • Run your own analysis on local transcript files if you want a rollup you control end-to-end.
Chat reads transcripts, so it can answer across calls whether or not you ran the template. The template is what keeps severity comparable across the team and gives your destination table something to filter on. That’s why it still comes first.

​
Variations

Same workflow, three close relatives: Each is a separate custom template. Same shape muscle. Different output.

​
Where to go next

  • Workflows — the general pattern this is an instance of
  • Custom templates — visibility, sharing, and edit permissions
  • Before a meeting — pre-seeding so the artifact runs live
  • Projects — one project per customer, shared with the team
  • Chat — asking questions across a customer’s calls
  • After a meeting — generating audience-specific outputs from the same transcript