✳ IskraAI

Итоги встречи: как составить точное задание для суммаризации

The «Meeting Summary» tool helps quickly extract three key blocks from meeting text: decisions, open questions and follow-up actions. However, the quality of the result directly depends on how accurately you formulate the task. In this article we’ll review how to prepare the source material and goal to get a useful summary, not generic phrases. We’ll review the field structure, provide examples of successful and unsuccessful formulations and show how to check the result.

Abrir ferramenta: Resumo de reunião ↗

Why precise task formulation is needed

The tool only analyzes the text you pass to it. If the input lacks clear formulations of decisions or tasks, the model cannot extract them. For example, the phrase «discussed the budget» does not contain a decision, while «decided to increase the budget by 10%» does. So your task is to provide maximally specific source material.

The goal helps the tool understand exactly what you want to get out of it. 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 assignees and deadlines»), the model will focus more accurately on the desired elements.

Moreover, precise task formulation saves time on checking and editing. The better the task, the fewer revisions will be needed after generation.

Poor goal: «Make a summary». Good goal: «Extract decisions, open questions and tasks with assignees and deadlines».

Field «Source material»: 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 decision markers («decided», «approved», «agreed»), questions (« unclear», «requires clarification») and tasks («prepare», «check», «contact»).

If you paste a dialogue, make sure utterances are separated and it is clear who is speaking. This helps the tool correctly attribute tasks. Avoid overly long monologues without structure — they are harder to analyze.

Maximum volume of the first field is 8000 characters. If the text is longer, split it into parts or shorten it, leaving only meaningful fragments. Do not insert unrelated discussions.

Good fragment: «Anna: Let's prepare a draft by Friday. Sam: Agreed, I'll check on Monday. Decision on launch date not yet made». Poor fragment: «Well, we discussed something like, we decided something, but I'm not sure».

Field «Goal or desired result»: how to specify

In this field describe exactly what you want to get. For example: «List of decisions with who approved them», «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 you set the format in text.

If you need a specific output language, specify it in the goal. For example: «Output in Portuguese». 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 Portuguese».

Example of poor and good task

A poor task often contains minimum context and a vague goal. For example, source text: «Discussed the project. Anna will do something. Sam will check. Launch is up in the air». 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. Decision on launch date not made, need to clarify with client». Goal: «Extract decisions, open questions and tasks with assignees and deadlines».

In a good example the tool will be able to accurately determine: Anna's task — prepare a draft by Friday; Sam's task — check on Monday; open question — launch date; decision — none yet. This is the desired result.

Poor input: «Meeting happened, discussed things». Good input: «Anna will prepare a draft by Friday. Sam will check on Monday. Launch date not determined».

How to check the result

After generation carefully read the obtained 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 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 tasks.

Check: the source text says «Anna will prepare a draft by Friday». The result should say: «Anna — prepare a draft by Friday». If this is not the case, 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. Maximum text volume is 8000 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 output is plain text that can be copied or downloaded as TXT. DOCX or PDF export is not available.

The model may make mistakes, especially if the source text contains ambiguities. Do not rely on it for critical decisions without human verification.

If the text says «maybe we'll launch in June», the tool should not output this as a decision. It may mark it as an open question.

An exact task for the «Meeting Summary» tool is a combination of structured source text and a specific goal. Specify which exact elements you want to extract, in what format and in which language. Check the result against the source data. Remember the limitations: text only, up to 8000 characters, no integrations. Following these recommendations you will quickly get useful meeting summaries and save time on routine processing.

Abrir ferramenta: Resumo de reunião ↗

As traduções da interface são assistidas por máquina e podem precisar de refinamento.