Summarizing and synthesizing are two of the most useful everyday AI tasks. They are also two of the easiest to misuse, because a smooth answer can hide what was omitted, softened, or invented.
Use AI here as a source-processing assistant, not as an authority. The job is to help you see the material more clearly while keeping the source in charge.
Summary And Synthesis Are Different
A summary compresses one source. It should preserve the source’s meaning, structure, and important details while removing noise.
Examples:
- turn a meeting transcript into decisions, action items, open questions, and risks
- turn a long article into the three points a teammate needs to understand
- turn raw notes into a status update draft
A synthesis combines multiple sources. It should compare them, show where they agree, show where they differ, and identify gaps or questions.
Examples:
- compare three stakeholder notes to find shared needs and unresolved disagreements
- combine research notes into themes, implications, and next questions
- compare product feedback from several channels without pretending everyone said the same thing
The mistake is asking for synthesis when you only need a summary, or asking for a summary when the real work is comparing sources.
Start With A Source Packet
Do not ask the model to summarize material it does not have. Give it a source packet: the approved text, notes, excerpts, transcript, requirements, or bullets it is allowed to use.
For a small task, the source packet might be a pasted note. For a larger task, it may be a labeled set of excerpts:
Source A: Team meeting notes
[approved notes]
Source B: Customer feedback themes
[approved excerpts]
Source C: Product constraints
[approved constraints]Labeling sources helps the model keep track of where ideas came from. It also makes your review easier because you can ask, “Which source supports that claim?”
Use A Grounded Summary Prompt
For one source, use a prompt that tells the model how to compress the material and what not to add.
Example:
Task:
Summarize the source below for an internal teammate who needs the practical takeaways.
Context:
Use only the source text provided. The source may be messy and incomplete.
Constraints:
- Do not add facts, dates, owners, decisions, or recommendations that are not in the source.
- Mark unclear items as "Needs verification."
- Preserve important nuance, caveats, and disagreements.
Output shape:
Use sections for Key Points, Decisions, Action Items, Open Questions, and Risks.
For action items, include Owner and Due Date only when stated.The important parts are “use only the source” and “mark unclear items.” Without those constraints, the model may fill gaps because a complete answer looks more helpful.
Use A Synthesis Prompt For Multiple Sources
For several sources, do not ask for “a summary of everything.” That usually produces a blended answer that hides differences. Ask for comparison.
Example:
Task:
Synthesize the three labeled sources below for a project lead deciding what to investigate next.
Context:
Source A is team notes. Source B is stakeholder feedback. Source C is implementation constraints.
Constraints:
- Use only the labeled sources.
- Do not smooth over disagreement between sources.
- Identify which source supports each major point.
- Mark unsupported interpretations as "Possible interpretation, not stated."
Output shape:
Use sections for Shared Themes, Differences, Gaps, Implications, and Next Questions.This prompt asks the model to preserve tension. That matters. When synthesis turns disagreement into false consensus, the output may feel tidy but mislead the team.
What To Check In A Summary
Review a summary against the source before handing it off.
Check for:
- Coverage: Did it include the main points, decisions, risks, and open questions?
- Faithfulness: Did it preserve meaning without exaggerating or softening it?
- Unsupported detail: Did it add names, dates, owners, numbers, quotes, or causes not in the source?
- Omission: Did it leave out a caveat, disagreement, blocker, or uncertainty that matters?
- Action safety: Are action items limited to what the source actually states?
If the source says “timeline is uncertain,” the summary should not say “the project is delayed two weeks” unless that delay is actually stated.
What To Check In A Synthesis
Review a synthesis for both accuracy and structure.
Check for:
- Source separation: Can you tell which idea came from which source?
- Agreements: Are shared themes supported by more than one source?
- Differences: Are conflicts or tensions named instead of erased?
- Gaps: Does the synthesis say what the sources do not answer?
- Interpretations: Are conclusions clearly separated from source facts?
- Next questions: Does it help the reader decide what to verify next?
A good synthesis is not just shorter. It improves judgment by making patterns visible without pretending the source material is cleaner than it is.
Worked Example: Three Notes Into One View
Suppose you have three short source notes:
- Source A: The support team says customers are confused by setup steps.
- Source B: Sales says buyers ask whether setup requires technical help.
- Source C: Product notes say setup has three required configuration decisions.
Weak synthesis:
Customers need easier setup.
Better synthesis:
Shared theme: setup clarity is a barrier. Source A reports customer confusion, Source B reports buyer questions about technical help, and Source C shows setup has three required configuration decisions. Gap: none of the sources shows which step causes the most confusion. Next question: identify the highest-friction setup step before proposing a fix.
The better synthesis does not jump straight to a solution. It preserves evidence and names the gap.
Handoff Language
When sharing a summary or synthesis, include a short review note:
Review state: Source-checked against the provided notes.
Unverified: Owners and due dates were not stated in the source.
Main risk: The synthesis identifies likely themes but does not prove priority.
Next step: Confirm the highest-impact theme with the source owner.That note tells the next person how much trust to place in the output.
Reference Links
These public references are useful starting points for deeper study. They are linked for attribution and further reading; the lesson above is synthesized as original LIW training guidance.
- OpenAI: Prompt engineering - useful for general prompting patterns.
- Anthropic: Prompt engineering overview - useful for source-grounded prompt habits.
- People + AI Guidebook - useful for trust and user experience patterns around AI-assisted work.
Compact Exercise
Take two short source notes about the same topic. First, write a summary of each one. Then write a synthesis that lists shared themes, differences, gaps, and next questions. Check whether every major claim can be traced back to a source.