You ask an AI to summarize a 25-page report. The answer is clear and much shorter than the original. Then you notice it missed the deadline on page 14, a financial risk, an unresolved decision, and an exception in a footnote.
A broad request such as “summarize this” leaves the AI to decide what matters. That may be fine for casual reading. For contracts, client work, project reports, or meeting records, use a process that protects the details readers need.
Define the reader and purpose
Before asking for a summary, decide who will read it and what they need to do with it.
Compare:
Summarize this report.
with:
Summarize this report for a project manager who needs to decide what must happen during the next 30 days. Preserve every deadline, decision, named owner, unresolved issue, financial commitment, and stated risk.
The second prompt defines “important” for this task. An executive may need decisions, risks, and financial implications. A project manager may need owners, dates, dependencies, and open questions. A technical team may need requirements, exceptions, definitions, and implementation constraints.
Extract key details before summarizing
For important documents, use two passes. First, extract the facts. Then write the summary.
Try:
Before writing a summary, extract every:
- date and deadline;
- monetary amount;
- percentage;
- decision;
- action item;
- named owner;
- stated risk;
- unresolved question;
- dependency;
- exception.
Preserve the original section or page reference where available. Do not combine or interpret items yet.
Review the resulting inventory, then ask for the summary. This makes it easier to spot a number or condition that might otherwise disappear from the narrative.
Protect conditions and exceptions
Suppose the report says:
The migration can begin in October, provided security testing is completed by September 18.
A loose summary might say, “The team plans to begin migration in October.” It leaves out the condition.
Tell the AI which details must remain:
Do not omit conditions, exceptions, deadlines, numerical values, or statements beginning with words such as “provided,” “unless,” “except,” “subject to,” or “however.”
This gives the AI a clear rule for deciding what can be shortened.
Summarize long documents section by section
Use the report’s own structure. For example:
- Executive Summary
- Current Situation
- Proposed Changes
- Financial Impact
- Risks
- Implementation
- Appendices
Ask for a structured summary of each section, then combine them. If the final version says nothing about financial impact, you can compare it with that section’s summary instead of searching all 70 pages again.
For important claims, keep the section or page reference. For example:
Estimated project cost: €420,000. Financial Impact, p. 18.
Check AI-generated references against the source before relying on them. They can be wrong.
Separate facts from interpretations
Ask the AI to sort details into four groups:
- Facts: directly stated in the document.
- Opinions: views attributed to a person or organization.
- Assumptions: conditions the document relies on.
- Open questions: issues not yet resolved.
For example:
| Type | Extracted item |
|---|---|
| Fact | Current contract expires December 31 |
| Opinion | Operations team believes migration will take six weeks |
| Assumption | New hardware arrives before November |
| Open question | Final migration vendor has not been selected |
This format can make an operational document easier to use than a polished paragraph.
Give the summary enough room
Compression can remove useful information. If a 60-page report has 14 deadlines, eight risks, 11 action items, and six major decisions, a 100-word summary cannot preserve them all.
Try:
Keep the executive overview under 250 words, but include complete separate tables for deadlines, risks, and action items.
You can keep the introduction brief while preserving operational details in tables.
Ask what was left out
After the summary, ask:
Add a section titled “Important details not included in the summary.” List information you excluded because it was too detailed, secondary, or difficult to classify.
The AI can still miss something, but this asks it to identify what it chose to omit.
You can also run a separate omission check:
Compare your summary against the source document. Find information present in the source but absent from the summary. Pay particular attention to names, numbers, dates, conditions, exceptions, commitments, risks, action items, and unresolved questions. Do not rewrite the summary yet. Give me the omissions first.
Review the list before deciding what belongs in the final version.
Example: a loose prompt and a controlled prompt
Loose prompt
Summarize this client project report in a few paragraphs.
Controlled prompt
Create a summary for the project owner.
First extract every deadline, numerical value, decision, risk, dependency, action item, and unresolved question.
Then write a 400–600-word summary organized as:
- Current status
- Decisions made
- Upcoming deadlines
- Risks and dependencies
- Actions and owners
- Open questions
Preserve section or page references for important claims.
Do not omit conditions or exceptions.
Finish with “Details excluded from this summary,” listing material you intentionally left out.
The second prompt does not guarantee a perfect result. It does make the output easier to review.
Use a table when readers need to act
A narrative can bury the relationships between tasks, owners, deadlines, and dependencies. For an operational report, a table may be clearer:
| Action | Owner | Deadline | Dependency | Status |
|---|---|---|---|---|
| Finalize specification | Project lead | Oct. 12 | Client approval | Open |
| Security review | IT team | Oct. 18 | Final specification | Not started |
Use prose to explain the situation and tables when readers need to take action.
Review high-risk details yourself
Before relying on an AI summary, compare the following with the source:
- names and action owners;
- dates and deadlines;
- monetary amounts, percentages, and units;
- contractual commitments;
- conditions and exceptions;
- negative terms such as “not,” “unless,” and “except”;
- decisions, proposals, risks, and unresolved questions.
A fluent paragraph can still contain a subtly wrong number.
Final review checklist
Before sending a summary to a client or colleague, ask:
- Did I define the reader and purpose?
- Did I extract important facts before summarizing?
- Are the critical dates and numbers correct?
- Are decisions distinguished from proposals?
- Are conditions and exceptions preserved?
- Are available action owners named?
- Are risks, dependencies, and unresolved questions included?
- Can readers trace important claims to the source?
- Did I ask what was excluded and run an omission check?
- Did a person compare critical details with the original?
A useful summary shortens a document while keeping the information its reader needs.