Meeting Summary: How to Get Decisions, Open Questions, and Action Items from Text
The 'Meeting Summary' tool transforms meeting text into a structured summary: decisions, open questions, and action items. In this article we'll walk through how to prepare your source material, set your goal, verify and manually refine the result. You'll learn what limitations to keep in mind and how to avoid common mistakes. All examples are fictional and provided for clarity.
Open tool: Meeting summary ↗What the Tool Actually Does
'Meeting Summary' is a text-only tool. You paste your meeting transcript or notes into the 'Source material' field, set a goal in the 'Goal or desired result' field, and click 'Generate'. The tool returns a single result in a separate panel: decisions, open questions, and action items. The result can be copied or downloaded as a plain text TXT file.
The tool does not create audio, video, or images. It works exclusively with text. If you have a meeting recording, you first need a textual transcript obtained by other means. For example, you can use local speech recognition, but this is not a built-in feature of 'Meeting Summary'.
Important to understand: the tool does not make decisions for you. It only highlights what was already stated in the text. If the source text does not contain a clear decision, the tool should not invent one. Your task is to verify that every decision actually sounded during the meeting.
Source material: 'Anna will prepare a draft by Friday. Sam will check by Monday. The launch date has not been determined yet'. Result: Decisions: none. Open questions: launch date. Action items: Anna - draft by Friday; Sam - check by Monday.
How to Prepare the Source Text
The quality of the result depends directly on the quality of the source text. Before pasting, remove everything irrelevant: greetings, weather discussion, personal comments. Leave only substantive fragments: who proposed what, what arguments were brought, what conclusions were reached.
If you have a meeting transcript, check that it includes participant names and their roles. The tool works better when it is explicitly stated who is speaking. For example, 'Anna: I will prepare a draft' is clearer than 'A draft needs to be prepared'.
Break the text into paragraphs by topic. If the meeting was long, you can divide it into logical blocks: budget discussion, timeline discussion, task discussion. This makes it easier for the tool to extract decisions and open questions. Do not paste more than 8000 characters into the first field - this is a tool limitation.
Instead of: 'Well, we talked here, and it seems Anna might make a draft, but she's not sure, maybe by Friday, or maybe later'. Better: 'Anna: I will prepare a draft by Friday. If I can't finish, I will let you know in advance'.
How to Set the Goal
The 'Goal or desired result' field helps the tool understand what to focus on. If you want only action items, write: 'Extract action items with responsible persons and deadlines'. If you need all three categories, write: 'Decisions, open questions, and action items'.
The goal should be specific but not too narrow. For example, 'Prepare a summary for the team' is a good goal. 'Find all budget mentions' is too narrow, the tool might ignore other important points. It is better to indicate a general task and clarify details in the source text.
Do not use the goal as a place for formatting instructions. The tool does not support styles, bold text, or bulleted lists. It returns plain text. Your goal should be semantic, not ornamental.
Goal: 'Collect decisions, open questions, and action items. For each action item, specify a responsible person and deadline if they exist in the text'.
First Iteration: What You Will Get
After clicking 'Generate' you will see one result in a separate panel. It usually consists of three blocks: decisions, open questions, and action items. In each block there may be items or a note saying 'none'. This is a draft, not a final document.
The first result is often incomplete. The tool may miss an action item if it is formulated implicitly. For example, the phrase 'We need to discuss this' might not end up in open questions if the tool does not recognize it as a question. This is normal for the first iteration.
Check that all names are spelled correctly. If there was a typo in the source text, the tool may repeat it. Also check if the tool combined two different decisions into one. If so, go back to the source text and clarify the formulations.
First result: Decisions: launch postponed. Open questions: launch date. Action items: Anna - draft; Sam - check. Problem: no deadline specified for Sam.
How to Refine the Result into Finished Material
Refinement is your manual work. The tool does not have a comparison mode, change highlighting, or special settings panels. You simply edit the text manually. First, compare the result with the source text. Find all decisions that were stated but did not end up in the summary. Add them.
Then check the action items. Every action item should have a responsible person and a deadline, if they were stated. If no deadline was stated, write 'deadline not specified'. Do not invent deadlines yourself. If the responsible person was not named, write 'responsible person not assigned'.
Separately check the open questions. These are things that remained unresolved. Make sure you did not record in open questions what was already decided. And conversely: if a question remained unanswered, it should be in this block. After manual editing save the text as TXT.
After refinement: Decisions: launch postponed until the next meeting. Open questions: launch date not determined. Action items: Anna - draft by Friday; Sam - check by Monday, deadline for check - Monday.
How to Verify the Result
Verification consists of three steps. First: read the source text and the result side by side. Make sure every statement in the result is confirmed by the source text. If there is a phrase in the result that is not in the source, delete it.
Second step: check names and deadlines. Typos in names can lead to confusion. Deadlines should be stated as they sounded. If the deadline was relative ('until next Friday'), clarify the date manually.
Third step: check that all three categories are filled correctly. If there were no decisions in the source text, the decisions block should say 'none'. Do not leave a block empty - this may be perceived as an error. The tool can make mistakes, so final responsibility lies with you.
Verification: the source text says 'Sam will check on Monday'. The result: 'Sam - check on Monday'. All correct. If the result had been 'Sam - check on Tuesday', that would be an error.
Limitations and Common Errors
The tool does not support document upload, OCR, batch processing, automatic saving, integrations, or publication. You cannot send the result by email directly from the interface. There is no registration or premium plan. Everything you can do with the result is copy or download as TXT.
The tool does not work with audio or video. If you need speech recognition, use separate local solutions. Do not think that audio is sent to AI - it is not. Also the tool does not create photorealistic images and does not clone voices.
The model can make mistakes. Sometimes it misses action items or incorrectly identifies open questions. Queue and availability may change. Do not rely on the result as the sole source of truth. Always verify manually.
Typical error: the tool recorded 'discuss the budget' in decisions, although on the meeting they only proposed discussing it. Correct: this is an open question, not a decision.
'Meeting Summary' is a helper text tool. It helps quickly extract decisions, open questions, and action items from meeting text, but does not replace manual verification. Prepare clean source text, set a goal, get a first draft, then manually refine it: add what was missed, remove what is extra, check names and deadlines. Remember the limitations: text only, TXT only, no audio or video. Examples in the article are fictional. Use the tool as a draft, and create the final document yourself.
Open tool: Meeting summary ↗Interface translations are machine-assisted and may need refinement.