---
title: "The Insights Lab: Create and Test Custom AI Insights"
description: Insights lab, custom Prompts, AI prompts
---

[Skip to content](https://help.cloverleaf.ai/knowledge-base/prompts-create-and-test-custom-ai-insights#main-content)

English

Show submenu for translations

![Cloverleaf AI Logo\_black\_full](https://help.cloverleaf.ai/hubfs/Cloverleaf%20AI%20Logos_2026/Black/Cloverleaf%20AI%20Logo_black_full.svg)

Open main navigation

Close main navigation

- English
  
  Show submenu for translations

 Welcome! How can we help you?

- There are no suggestions because the search field is empty.

1. [Cloverleaf AI Support](https://help.cloverleaf.ai/knowledge-base?hsLang=en)
2. [AI Tools](https://help.cloverleaf.ai/knowledge-base/ai-tools?hsLang=en)

# The Insights Lab: Create and Test Custom AI Insights

## The Insights Lab lets your organization write its own AI instructions for finding insights in meetings and documents. Each prompt runs in two steps, Analysis and then Scoring, so Cloverleaf AI surfaces the topics you care about and ranks them the way you would.

### Who Can Create Prompts

Only Operations Admin users can create prompts. Prompts live under Account \> Prompts, and once a version is made active it becomes available to use across your organization.

### How a Prompt Works

Every prompt has two steps that run one after the other.

- **Step 1: Analysis.** Reads the meeting transcript (or document) and pulls out what matters, based on your instructions. The output quotes the source and marks where each moment happened with a timestamp like \[ts:45:13\].
- **Step 2: Scoring.** Takes the output of Step 1 and scores it, for example from 1 to 10, with a short explanation.

You only write the instructions in the middle. Cloverleaf AI adds the rest for you: the meeting transcript is attached to Step 1, Step 1's output is attached to Step 2, and the timestamp rules are already built in. You do not need to mention the transcript or the previous step in your instructions.

Each step holds up to 2,000 tokens of instructions. A counter in the top right of each step shows how much you have used.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-06-58-8951-PM.png?width=847&height=698&name=undefined-Nov-05-2026-04-06-58-8951-PM.png)

### Create a Prompt

1. Go to Account \> Prompts.
2. Select New Prompt at the top right of the list. If you have no prompts yet, the button says **Create A Prompt**.
3. Enter a **Name**. Pick something your team will recognize, such as "Budget and Funding Signals."
4. Under **Start from**, choose **Blank meeting prompt** or **Blank document prompt**. A meeting prompt runs on meeting transcripts. A document prompt runs on documents and works the same way, with the same two steps.
5. Select **Create**. The editor opens with a new draft, version v1.0.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-05-49-2824-PM.png?width=1568&height=698&name=undefined-Nov-05-2026-04-05-49-2824-PM.png)

### Write Your Instructions

Type your instructions into the large box in each step. The gray text above and below the box is fixed. It is the wrapper Cloverleaf AI places around what you write, so you cannot edit it. A small toolbar gives you bold, italic, strikethrough, quotes, and bulleted or numbered lists.

1. In **Step 1: Analysis**, describe what to look for in the meeting and how to report it. Example: "Find any mention of budget, funding, or software replacement. Quote the relevant discussion."
2. In **Step 2: Scoring**, describe how to rate what Step 1 found. Example: "Score the finding from 1 to 10 based on how likely it is to lead to a software purchase. Explain the score in one sentence."

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-06-35-6060-PM.png?width=824&height=698&name=undefined-Nov-05-2026-04-06-35-6060-PM.png)

### Profile Variables

Profile variables let one prompt adapt to each person who runs it. Type the variable name in curly braces, such as {jobTitle}, and Cloverleaf AI fills it in from the profile of whoever runs the insight. Typed variables show in red in the editor. Select **Profile Variables** at the top right of the editor to see the full list.

| Variable | What it fills in | Good for |
| --- | --- | --- |
| {jobTitle} | The user's job title | Framing the insight for the reader |
| {industry} | The industry the user's company works in | Deciding what counts as relevant |
| {goals} | Focus Areas: what the user wants to get out of Cloverleaf AI, in their own words | Usually the best guide to what the insight should focus on |
| {competitors} | The user's competitors | Spotting competitor mentions |
| {productsServices} | The user's products and services | Matching discussions to what they offer |
| {painPoints} | Customer problems the user's products or services solve | Spotting discussions where those problems come up |
| {notRelevantSignals} | Signals the user has marked as not relevant | Steering away from similar results |

The same seven variables are available in both steps.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-07-29-5058-PM.png?width=1568&height=698&name=undefined-Nov-05-2026-04-07-29-5058-PM.png)

### Save a Version

Your work is not stored until you save it. When you change either step, an orange **Unsaved changes** tag appears next to the version number.

1. Select **Save Version** at the top right.
2. Add an optional **Notes** entry that describes the change, such as "Added budget keywords."
3. Select **Save**. A "Version saved" message confirms it.

If you would rather start over, select **Discard Changes** to go back to the last saved text.

Saving a draft overwrites that draft. If you changed the steps, its earlier test runs are cleared, because those results came from the old text. Select **History** to see every version with its status, who saved it, when, and how many test runs it has.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-07-47-9551-PM.png?width=1568&height=698&name=undefined-Nov-05-2026-04-07-47-9551-PM.png)

 

### Test a Version

Test runs let you see what a saved version produces on real meetings before anyone relies on it. Save your changes first, then scroll to the **Test run** section at the bottom of the editor.

1. Open the **Campaign** dropdown and choose one of your campaigns. The list shows up to the top 20 meetings from that campaign.
2. Check the box next to each meeting you want to test. You can pick up to 5. The button counts your picks, such as **Run Test (2/5)**.
3. Select **Run Test**. Each meeting shows a **Running** tag while the two steps work.
4. When a meeting shows **Done**, select its row to open the results.

Each result shows Step 1: Analysis on the left and Step 2: Scoring on the right. Analysis lists the quotes it found, each with a timestamp such as \[ts:01:10:51\] that marks where it happened in the meeting. Scoring shows the score and a one-sentence explanation. Test runs use your own default profile and your not-relevant signals to fill any profile variables.

Test runs only show you the output. They do not attach the prompt to the campaign or change what the campaign shows. Select one meeting at a time while you are tuning, since each run may use AI credits. Selecting Run Test again starts another run, so wait for the button to respond before selecting it twice.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-08-14-7937-PM.png?width=1568&height=698&name=undefined-Nov-05-2026-04-08-14-7937-PM.png)

### Make a Version Active

Select **Make Active** when a version is ready for your organization to use. It sits next to History at the top of the editor, and also appears under each version in the History panel. Once a version is active, that prompt is available to use in the app for your organization.

Making a version active does not apply it to campaigns automatically. The campaign picker on the prompt page is only for testing. Make Active is greyed out while you have unsaved changes, so save first.

![](https://help.cloverleaf.ai/hs-fs/hubfs/undefined-Nov-05-2026-04-08-37-9218-PM.png?width=818&height=698&name=undefined-Nov-05-2026-04-08-37-9218-PM.png)

### Rename or Delete a Prompt

Select the three-dot menu next to the prompt name to see **Rename** and **Delete**. Select **All Prompts** at the top left to go back to the list.

### **Advanced Example: Early Modernization Signals**

This example finds early signs that a government body may buy new software before any RFP exists. It uses five profile variables, a fixed output format, and a scoring guide. 

### Step 1: Analysis

You are helping a {jobTitle} in {industry} find early signs that a government body may buy new software before any RFP exists. The user's goals are: {goals}. They sell: {productsServices}. Problems they solve: {painPoints}. Competitors they care about: {competitors}.

Read the meeting and find every moment where officials discuss any of these: replacing or upgrading an existing system, complaints about current software or manual processes, budget set aside for technology, vendor contracts that are expiring or up for renewal, staff asking for new tools, consultants or assessments about IT, or grant funding tied to technology.

For each moment, report the following on separate lines: Topic (a short label), Quote (the exact words, kept short), Speaker and role if stated, Stage (one of: Complaint, Discussion, Budget request, Approved funding, Vendor selection), and Named vendors or products if any. Include the timestamp for each. Skip anything that matches {notRelevantSignals}. If a topic is only mentioned in passing with no action or money attached, mark it as Passing mention. If you find nothing, say so plainly. Do not guess or fill gaps.

### Step 2: Scoring

Score each finding from 1 to 10 for how likely it is to become a software purchase in the next 12 months, using this guide.

- 1 to 3: Passing mention, no money, no owner, or only about non-software items.
- 4 to 6: A real problem or plan is discussed, but no funding or timeline.
- 7 to 8: Funding is requested or a timeline is stated, and the need matches {painPoints}.
- 9 to 10: Funding is approved or a vendor selection is underway, and the need clearly matches {productsServices}.

Subtract 2 points if a {competitors} product is already named as the chosen solution. Add 1 point if a department head or elected official speaks in favor. Give one overall score for the meeting, using the strongest finding, then list each finding with its own score and a one-sentence reason. End with a line that says "Top lead:" followed by the single best finding.

### Why This Prompt Works

- **Finding and judging are split.** Step 1 only collects and quotes. Step 2 only scores. This makes results easier to tune.
- **The output has a fixed shape.** Labeled lines such as Stage give Step 2 something concrete to read.
- **The scoring guide is explicit.** Each score range says what must be true, so two meetings are scored the same way.
- **Variables do the personalizing.** One prompt works for every user, because {goals}, {productsServices} and {painPoints} come from each person's profile.

 Tips

- Start small. Write one clear question for Step 1 and one scoring rule for Step 2, test it, then add detail.
- Keep Step 1 about finding and quoting. Keep Step 2 about judging. Mixing them makes results harder to tune.
- Use profile variables such as {goals} and {painPoints} so one prompt works for sales and government affairs users alike.
- Add a note each time you save so History tells you what changed.
- Test on a mix of meetings, including some that should score low.

- [Getting Started](https://help.cloverleaf.ai/knowledge-base/getting-started?hsLang=en#main-content)

    - [Login + Settings](https://help.cloverleaf.ai/knowledge-base/getting-started?hsLang=en#login-settings)
    - [Default Email Client](https://help.cloverleaf.ai/knowledge-base/getting-started?hsLang=en#default-email-client)
- [Search](https://help.cloverleaf.ai/knowledge-base/search?hsLang=en)
- [AI Tools](https://help.cloverleaf.ai/knowledge-base/ai-tools?hsLang=en#main-content)

    - [Meeting Insights](https://help.cloverleaf.ai/knowledge-base/ai-tools?hsLang=en#meeting-insights)
    - [Channel Insights](https://help.cloverleaf.ai/knowledge-base/ai-tools?hsLang=en#channel-insights)
    - [Meta Analysis](https://help.cloverleaf.ai/knowledge-base/ai-tools?hsLang=en#meta-analysis)
- [Campaigns](https://help.cloverleaf.ai/knowledge-base/campaigns?hsLang=en#main-content)

    - [Campaigns](https://help.cloverleaf.ai/knowledge-base/campaigns?hsLang=en#campaigns)
    - [Signals Emails](https://help.cloverleaf.ai/knowledge-base/campaigns?hsLang=en#signals-emails)
- [Meetings and Documents](https://help.cloverleaf.ai/knowledge-base/meetings-and-documents?hsLang=en#main-content)

    - [Videos](https://help.cloverleaf.ai/knowledge-base/meetings-and-documents?hsLang=en#videos)
    - [Documents](https://help.cloverleaf.ai/knowledge-base/meetings-and-documents?hsLang=en#documents)
- [Channels](https://help.cloverleaf.ai/knowledge-base/channels?hsLang=en)
- [Best Practices](https://help.cloverleaf.ai/knowledge-base/best-practices?hsLang=en)
- [Admin Tools](https://help.cloverleaf.ai/knowledge-base/admin-tools?hsLang=en)
- [FAQs](https://help.cloverleaf.ai/knowledge-base/faqs?hsLang=en)

[![Chill listening crop-3](https://help.cloverleaf.ai/hs-fs/hubfs/Logo_Favicon.png?width=62&height=62&name=Logo_Favicon.png "Chill listening crop-3")](https://auth.cloverleaf.ai/u/login?state=hKFo2SB6aDdHZGk1ck03OFdGYjJiWmdpMU9Bd3h3cERxWjhKdqFur3VuaXZlcnNhbC1sb2dpbqN0aWTZIHB6M3Eyc1dMOE5QTkR6QVdOeXN3b0VDOHFoMnFKbnl0o2NpZNkgTjlJYk1iZnk1U2Z1R0Q1UFBLcmZYb2JEM1lDVUpvSVA)

Cloverleaf AI Knowledge Base

Copyright © 2026, Cloverleaf AI