Building an Automated Meeting Intelligence Pipeline with Workspace Studio

If you spend a lot of time in Google Meet, you know that the real work happens after the call. Processing raw transcripts, structuring unstructured conversations, and routing action items to the right tools is a massive context switch.

Traditionally, automating this required wiring up webhooks, deploying a Cloud Function to handle the LLM prompts, and integrating with multiple Workspace APIs.

Today, I’m going to show you how to build this exact pipeline natively using Google Workspace Studio. We are going to take unstructured meeting data, pass it through Gemini for summarization and entity extraction, and route the structured outputs directly into Google Docs, Google Chat, and Google Tasks—all without writing a single line of code.

Prefer to watch instead of read? Check out the video version of the guide here.

Let’s dive into the architecture and build the flow.

The Architecture

Here is a high-level look at the pipeline we are building:

1. Ingestion: Triggered asynchronously when Google Meet finishes processing a transcript.
2. Processing (LLM): Gemini processes the text in two parallel logical steps: summarizing the broader context, and extracting specific action item entities.
3. Routing: The processed data is appended to a Google Doc for historical archiving, a payload is pushed to Google Chat for immediate alerting, and the action items are iterated over to generate individual Google Tasks.

Step 1: Set the Trigger

First, we need to set up the event listener that kicks off our pipeline.

1. Create a new flow in Workspace Studio.
2. Under Choose a starter, select When meeting outputs are ready.

This starter listens for the completion event from Google Meet, ensuring our flow only runs when the transcript and notes payload is fully generated and ready for processing.

Step 2: Contextual Summarization (Prompt Engineering)

Next, we’ll use the Summarize AI action. We are intentionally separating our summarization from our task extraction to ensure the LLM returns predictable, deterministic outputs for each step.

1. Click Add step -> AI actions -> Summarize.
2. For What to summarize, select Meeting notes.
3. Map Select meeting notes link to the variable from Step 1.
4. In the Additional instructions for Gemini field, we need to apply some strict prompting constraints. Paste the following:

Goal:
Provide a clear, high-level overview of the meeting based on the provided notes/transcript.
Requirements:

1. Logic: If the document is empty or contains only metadata, state "No substantive content found."
2. Content: Focus on the big picture, skipping minor small talk. Do NOT include action items here, as they will be extracted separately. Output Format:
3. Meeting Topic/Purpose - 1 sentence summary.
4. Key Decisions - Bulleted list of final decisions made.
5. Major Discussion Points - Bulleted list of topics covered.

Step 3: Entity Extraction (Action Items)

Now for the extraction phase. We need a clean array of tasks that we can iterate over later. Instead of writing a complex Regex or a highly-structured JSON prompt,Workspace Studio includes predefined extraction models, so no custom prompting is required here.

1. Click Add step -> AI actions -> Extract.
2. Map Content to analyze to the summary from Step 2.
3. Scroll down to What to extract. Click the Predefined dropdown and select Action items.

By using the predefined model, Gemini reliably returns an iterable list of tasks, avoiding the parsing errors you might get from a custom prompt.

Step 4: Data Routing (Docs & Chat)

Now that our data is processed and structured, we need to route it to our tools.

Archiving to Google Docs:

1. Click Add step -> Docs -> Create a doc (or Add to a doc if you have an existing doc).
2. Build your document schema by mapping the variables:

• Summary: [Map Output from Step 2]
• Action Items: [Map Output from Step 3]

Alerting via Google Chat:

1. Click Add step -> Chat -> Notify me in Chat.
2. Construct your payload. I like to keep this brief:

Meeting Processed! 📝
Summary: [Map Output from Step 2]
Tasks extracted: [Map Output from Step 3]

Step 5: Iterating over the Array (Google Tasks)

Finally, we need to take the array of extracted tasks from Step 3 and generate individual Google Tasks. To do this, we use a loop.

1. Click Add step -> Tools -> Repeat for each.
2. Map List to repeat to the Action Items variable from Step 3.
3. Inside the loop block, click Add substep -> Tasks -> Create a task.
4. Map the Task Title to the Current Item variable generated by the loop.

This is a powerful pattern: iterating over LLM-extracted entities to trigger downstream write operations.

Step 6: Testing the Automation

Before setting the automation live, Workspace Studio allows you to run a test using data from a past meeting.

1. Click the Test run button at the bottom of the screen.
2. Under "Meetings," select a recent meeting that has a transcript available (e.g., a "Project Sync" meeting).
3. Once done, click Start.

Workspace Studio will process the historical data as if the meeting just ended. Here is what the automated outputs look like:

1. The Chat Notification You will receive an instant direct message from Workspace Studio confirming the successful run, complete with a link to your newly generated notes.

2.The Formatted Google Doc Opening the linked Google Doc will show a clean, formatted summary detailing the Meeting Topic, Key Decisions, and Major Discussion Points, followed by the extracted action items.

3.Populated Google Tasks Checking your Google Tasks panel will reveal that every action item discussed on the call has been individually extracted and added as a distinct task, complete with the "Auto-generated from recent meeting" note we configured.

Seeing the test run execute successfully highlights exactly why this workflow is so powerful. Instead of manually reviewing a 45-minute transcript, copying out tasks, and pasting them into separate project management tools, the data flows seamlessly across your Workspace environment.

Notice how the extraction model (in Step 3) accurately identified only the actionable items, completely ignoring the small talk and general discussion points. This level of AI-driven precision prevents your task list from becoming cluttered with irrelevant information. By routing the data immediately to Google Chat, you get a real-time notification that the process is complete, so you never have to guess if the automation fired correctly.

Conclusion

Once you have verified your test run looks good, click Turn on at the bottom of the Workspace Studio editor.

Moving forward, this automation will run silently in the background every time a meeting output is generated. You’ve successfully transformed a tedious administrative chore into an automated pipeline, ensuring your tasks are always centralized and your meeting records are impeccably archived without a single manual keystroke.

Google Workspace Studio opens up incredible possibilities for streamlining your day-to-day operations. Start building, experiment with different AI prompts, and take your time back!