Meeting Summary: How to Formulate an Accurate Prompt for Summarization
The Meeting Summary tool helps quickly extract three key blocks from meeting text: decisions, open questions, and action items. However, the quality of the result directly depends on how accurately you formulate the task. In this article, we will explore how to prepare source material and objectives to get a useful summary instead of generic phrases. We will examine the structure of the fields, provide examples of successful and unsuccessful phrasing, and show how to verify the result.
Open tool: Meeting summary ↗Why Precise Task Formulation Matters
The tool analyzes only the text you provide it. If your input lacks clear formulations of decisions or tasks, the model cannot extract them. For example, the phrase 'we discussed the budget' does not contain a decision, while 'we decided to increase the budget by 10%' does. Therefore, your task is to provide maximally specific source material.
The goal of the task helps the tool understand exactly what you want to get at the output. If the goal is vague ('make it beautiful'), the result will be less predictable. If the goal is specific ('extract decisions, open questions, and tasks with responsible persons'), the model will focus more accurately on the desired elements.
Moreover, a precise formulation saves time on verification and editing. The better the task, the fewer edits will be required after generation.
Poor goal: 'Make a summary'. Good goal: 'Extract decisions, open questions, and tasks with assignees and deadlines'.
Source Material Field: What to Include
In the 'Source material' field, paste the full meeting text: a transcript, notes, or a protocol. It is important that the text contains explicit markers of decisions ('decided', 'approved', 'agreed'), questions ('unclear', 'requires clarification'), and tasks ('prepare', 'check', 'contact').
If you are pasting a dialogue, make sure the lines are separated and it is clear who is speaking. This will help the tool correctly attribute tasks. Avoid overly long monologues without structure — they are harder to analyze.
The maximum volume of the first field is 8,000 characters. If the text is longer, split it into parts or shorten it, keeping only significant fragments. Do not insert unrelated discussions.
Good fragment: 'Anna: I will prepare a draft by Friday. Sam: Agreed, I will check on Monday. The launch date decision is not yet made'. Poor fragment: 'Well, we talked about something, like we decided something, but I'm not sure'.
Goal or Desired Result Field: How to Specify
In this field, describe exactly what you want to get. For example: 'A list of decisions with who approved them', 'A list of open questions with context', 'Tasks with assignees and deadlines'. The more detailed the goal, the better.
You can specify the desired format: 'Three separate blocks: Decisions, Open questions, Tasks'. The tool has no preset styles, so the format is defined by you in text.
If you need a specific output language, specify it in the goal. For example: 'Output the result in Russian'. By default, the language may be determined by the source text, but it is better to clarify.
Goal: 'Extract three blocks: 1) Decisions made, 2) Open questions, 3) Tasks with assignees and deadlines. Output in Russian'.
Example of Poor vs. Good Task
A poor task often contains minimal context and a vague goal. For example, source text: 'We discussed the project. Anna will do something. Sam will check. Launch is uncertain'. Goal: 'Make a summary'. The result will be inaccurate because there are no clear decisions and tasks.
A good task includes structured text and a clear goal. Source text: 'Anna: I will prepare a draft by Friday. Sam: I will check it on Monday. The decision on the launch date has not been made, we need to clarify with the client'. Goal: 'Extract decisions, open questions, and tasks with assignees and deadlines'.
In the good example, the tool will be able to accurately determine: Anna's task is to prepare a draft by Friday; Sam's task is to check on Monday; the open question is the launch date; and the decision is currently absent. This is the desired result.
Poor input: 'The meeting happened, we discussed things'. Good input: 'Anna will prepare a draft by Friday. Sam will check on Monday. The launch date is not determined'.
How to Verify the Result
After generation, carefully read the received text. Make sure all decisions from the source material are reflected, open questions are not lost, and tasks have assignees and deadlines (if they were specified).
Compare the result with the source text: did any extra entities appear that were not in the input? The tool can make mistakes, so it is important to verify facts.
If the result is incomplete, try to clarify the goal or add context to the source material. Sometimes it helps to split the task into several requests: first extract decisions, then extract tasks.
Verification: the source text says 'Anna will prepare a draft by Friday'. The result should be: 'Anna — prepare a draft by Friday'. If this is missing, the task requires clarification.
Tool Limitations
The tool works only with text. It does not analyze audio or video directly, so you need to get a transcript first. The maximum text volume is 8,000 characters, which roughly corresponds to 5–7 minutes of speech.
The tool does not save history, has no integrations, and does not support batch processing. The result is output as plain text, which can be copied or downloaded as TXT. DOCX or PDF export is unavailable.
The model may make errors, especially if the source text contains ambiguities. Do not rely on it for making critical decisions without human verification.
If the meeting text says 'we might launch in June', the tool should not output this as a decision. It may mark it as an open question instead.
An accurate task for the Meeting Summary tool is a combination of structured source text and a specific goal. Specify exactly which elements you want to extract, in what format, and in what language. Check the result against the source data. Remember the limitations: text only, up to 8,000 characters, no integrations. By following these recommendations, you will quickly get useful meeting summaries and save time on routine processing.
Open tool: Meeting summary ↗Interface translations are machine-assisted and may need refinement.