Guide
Find patterns across your Granola customer calls.
Put your customer calls in one Granola folder and ask the folder what keeps coming up. Save the question as a recipe so you ask it the same way every week. When you need the same themes over months, counted by company, keep a simple log by hand or use a tool that keeps it for you.
Updated . Plans and prices checked on Granola's site that day.
One Granola folder, counted
Customers redo our reports before they trust them
Granola is good at the single call. You come out of a customer conversation with clean notes and a transcript. The harder question comes a month later: across all those calls, what do customers keep running into?
This guide covers the routes Granola gives you for that question, what each one needs, and where each one stops.
Can Granola find patterns on its own?
Yes, for a while. Make a folder for customer calls, for example "Customer calls", and move each customer conversation into it. Leave internal meetings out, or they will crowd the answers.
Then open the folder and ask Granola's chat a question about all of it. Granola answers from every note in the folder and links each point to the meeting it came from, so you can open the passage and check it.
Folders and chat across meetings work on every Granola plan, including the free one. The free plan keeps 30 days of meeting history, so on it the folder only ever shows the last month.
A recipe is a chat question you save and run again. Once you have a question that gives a useful answer, save it as a recipe and run it every week. Asking the same way each time is what makes this week's answer comparable to last week's. Granola Recipes for customer feedback has three recipes to copy.
What should you ask the folder?
Ask for the customer's exact words, the company, and whether they raised it without being asked. Those three details are what let you trust a pattern later. A paraphrase hides how strongly something was said, and a count of mentions hides that one customer may have said it on four calls.
A question that works as a recipe:
Go through every call in this folder. List each problem or request a customer raised. For each one give: - the customer's exact sentence, word for word - the company - whether they raised it themselves or answered our question Then group them by the problem underneath, and for each group say how many different companies raised it.
Read the answer against the sources before you share it. Chat answers are written fresh each time, so a theme can be named differently from one week to the next.
How do you read your calls in Claude or ChatGPT?
Granola has an MCP server, a connector that lets an assistant such as Claude or ChatGPT search and read your Granola notes. It helps when you want to work through your calls in a longer conversation, or combine them with documents you already have in that assistant.
- In Claude, add Granola under Settings, then Connectors.
- In ChatGPT, search for Granola in Settings and connect it.
- In Claude Code, run
claude mcp add granola --transport http https://mcp.granola.ai/mcpand sign in with/mcp.
On the free plan, the MCP reaches personal and public notes from the last 30 days, and the transcript tools are not available. On Business it reaches your full history and the transcripts. The same recipe question above works well as a first message. The guide Use the Granola MCP to read your customer calls has more prompts and the setup in detail.
Which Granola plan does each route need?
| Route | Basic, free | Business, $14 a person a month |
|---|---|---|
| Folders and chat across meetings | Yes, 30 days of history | Yes, unlimited history |
| Recipes | Yes | Yes |
| MCP in Claude or ChatGPT | Last 30 days, no transcripts | Full history and transcripts |
| API keys, for other tools | No | Yes |
| Zapier, Notion, HubSpot | No | Yes |
Granola also has an Enterprise plan at $35 a person a month, where admins decide what the MCP and API keys may read. Prices from granola.ai/pricing on 17 November 2026.
Where do these routes stop?
Chat answers the question you ask today. It does not remember last month's answer, so it cannot tell you that a theme has grown from three companies to nine since the summer.
It also counts what it finds in the notes: meetings and mentions. If one customer brings up the same report problem on four calls, that easily reads as four separate requests. For deciding what to build, that customer is one company with one problem, however often they mention it.
And the answer is only as good as the folder. A call that was never moved into it does not exist for the question.
How do you keep the count by hand?
A spreadsheet is enough. After you run the weekly recipe, copy each new item into one row with five columns: date, company, the exact sentence, the theme, and whether they raised it without being asked. Reuse the same theme names week after week.
Then count companies per theme, not rows. Here is an example from a made-up team:
We export the report every Monday and fix it in Excel before anyone sees it.
| Theme | Rows | Companies | Without being asked |
|---|---|---|---|
| Reports need fixing by hand before they are shared | 11 | 6 | 4 |
Eleven mentions sounds bigger than six companies, but six is the number that tells you how many customers would notice a fix. The four who raised it without being asked tell you it hurts. The guide How to hear your customers has the full method behind those columns.
When is a separate tool worth it?
The log works, and it costs you the copying, every week, for as long as you want the numbers to stay true. Most teams stop once they get busy, which is when the numbers matter most.
That copying is the job Ordagrann does. Ordagrann reads the customer calls you record in Granola, groups every problem and request into themes, counts how many companies raised each one, and turns a theme into a one-page brief your team can build from. It connects with a Granola API key, so it needs Granola Business. Every count opens to the sentences behind it, and you decide which calls it may read.
Works with Granola Business. Any other transcript can be pasted in.