Summary
AI helped me by connecting fragments of information and explaining how they fit together. But as its explanations became more comprehensive, I found myself searching inside the answer for the information I actually wanted. That led me to change how I asked AI to report, depending on the reader and the purpose.
When the background is shared, focus on results and what the reader needs to make a decision. When it is not shared, build the necessary understanding first. And when the purpose is exploration, preserve related findings that could lead to the next discovery alongside the overall synthesis.
The work that remained after AI connected the information
When I used Google for research, I had to choose search terms, collect fragments of information, and connect them myself. When I started using AI, it connected that information and explained it to me. That was helpful.
But the explanations I received tended to be comprehensive. They connected the background, mechanisms, and related examples. I then had to search within those answers to find what I actually needed. There was less work in connecting information, but finding the relevant part still took effort.
I tried asking for the main points first and for shorter answers. Even then, I could not make the next decision if I could not tell which assumptions supported a conclusion or how much had actually been checked.
So I started thinking about what AI should communicate, and in what order, as well as how much it should write.
When the background is shared focus on results and decisions
When we are continuing a discussion on the same topic, there is no need to repeat all the background. I want a brief reminder of the purpose, followed by:
What was learned
How much has been verified
What evidence supports it
What remains unknown and what needs to be decided next
Consider this illustrative example of reporting a software change.
Describing the work in sequence
I changed the settings, added processing logic, and then ran the tests.
Organizing the information around the next decision
The requested input-error detection has been added. Tests with normal inputs and blank fields passed, but high-volume input has not been tested. That condition needs to be checked before release.
The second version shows both the progress and what needs checking before we move on. The detailed work log can remain available for reference.
I use Reporting for this role: communicating results against a shared purpose and decision criteria.
When the background is not shared build the necessary understanding
The same report may not be enough for someone encountering the topic for the first time.
In the example above, a reader who does not know what inputs the software handles, or what problems input errors cause, may struggle to judge why the change matters.
In that case, explain the purpose, background, necessary terms, and relationships before presenting the results and decision material. I distinguish this role as Presentation. It is a function that prose can serve too, not just slides.
This does not require producing two documents every time. It means choosing where the explanation starts based on what the reader already knows and what they still need to understand.
When exploring preserve related findings too
The purpose of the investigation also affects what should remain in a report, independently of the reader's background.
For example, I was investigating whether the expansion of AI data centers could keep up with demand. In simplified terms, facility capacity and planned opening dates are not enough: the facilities also need electricity. If on-site generation is used to supplement grid supply, another question follows: can the equipment used to generate that electricity, such as gas turbines, be obtained in time?
The synthesized report was quicker for me to read, but it dropped the gas-turbine information. A summary focused only on facility capacity and dates can remove the prompt to ask, “Should we also investigate the supply of power-generation equipment?” This was not a finding that a particular facility would be delayed. The problem was losing sight of a question worth investigating next.
That led me to retain two parts in exploratory reports.
Overall synthesis
Summarize what has been learned about the original question, what cannot yet be determined, and what to check next.
Related findings
Include information uncovered during the investigation that could lead to further exploration. In this example: “If on-site generation is used to supplement electricity supply, we should also check gas-turbine procurement lead times.” Make the connection to the original question and the next inquiry visible.
Adding everything would increase the reading burden again. Keep related findings separate so readers can follow the ones that interest them. Conditions that could change the conclusion, and major uncertainties, must also remain in the main synthesis.
Check assumptions on both sides
When organizing a report, I also want to leave room to check assumptions.
If the connections in AI's explanation differ from the ones I had in mind, that can reveal a possible mismatch. We can examine the difference, return to the evidence, and revise the question if necessary.
I want to address both AI's omissions and my own misunderstandings. My being convinced is not, by itself, verification that something is correct.
I am organizing this exchange as a reporting protocol. Beyond filling in a set of headings, it gives AI and the human a way to check what they agree on and what is still unresolved.
Choose an output that helps you make the next decision
Making AI-generated prose sound natural is useful. I also think choosing a report structure to fit its reader and purpose can address the effort of searching for the information I need.
These are the conditions I would now add to a request:
If the background is shared, briefly retain the purpose and focus on results and decision material
If it is not shared, supply the assumptions and relationships needed for understanding
If the purpose is exploration, separate the overall synthesis from related findings that could lead further
In every case, make evidence, unknowns, and differences in assumptions visible
What I gained this time was the sense that I could read the report faster, along with a discovery about information lost through synthesis. This does not establish an improvement in accuracy or understanding.
I want less effort spent reading while preserving room to think and question. That is why I want to adapt AI's output to my purpose.
What makes an AI answer difficult for you: its length, finding the information you need, or not being able to see the assumptions and evidence behind a decision?
Three prompts you can add to an AI request
You can append these prompts to a question or task request to specify how the answer should be organized. Replace the text in square brackets with your purpose and audience.
Prompts 1 and 2 address the reader's prior knowledge. Prompt 3 addresses the purpose of the investigation. You can combine them as needed.
1 Reporting to someone who shares the background
Put results and information needed for the next decision first, without repeating the shared background.
Purpose: [What I want to verify or decide]
The reader shares the background and purpose of this conversation.
Organize the answer concisely in this order:
1. Conclusion: What was learned or changed. Briefly state what to focus on for the next decision.
2. What was checked: Evidence, scope of verification, and references.
3. What remains unresolved: Unknowns and outstanding decisions.
4. Next steps: Necessary checks or actions.
When reporting work progress, add the current status and its reason after the conclusion.
Do not repeat shared background, but briefly retain the purpose.
Do not omit changed assumptions or differences in understanding.
Distinguish facts, inferences, and unverified points. If there is no conclusion yet, say so.
Do not invent missing assumptions. Ask for clarification when they are essential to the decision.
Keep the conditions under which the conclusion holds and its connection to the evidence, even when shortening the answer.
2 Explaining to someone who does not share the background
Supply the necessary context and relationships so the reader can judge what the results mean.
Purpose: [What the reader should understand or be able to decide]
Reader: [Who they are and what they already know]
The reader does not fully share the background of this topic.
Organize the answer in this order so the explanation is easy to follow:
1. Purpose and main point: What the topic is and why it matters to the reader.
2. Necessary context: Background, terms, and relationships.
3. Findings: Results and supporting evidence.
4. Meaning and limits: What the results establish and what they do not yet establish.
5. What to consider next: Open questions, decisions, and necessary checks.
Explain only the background needed for this purpose, rather than covering it exhaustively.
Distinguish facts, inferences, and unverified points. Do not invent missing facts to make the explanation flow.
If the reader's prior knowledge is unclear, identify that uncertainty. Ask for clarification if it would change how the topic should be explained.
Even when shortening the answer, preserve the connections that explain how the results or judgments follow.
3 Reporting to support further exploration
Synthesize the answer to the original question while preserving information that could lead to another discovery.
Question to investigate: [What I want to learn]
Reader's prior knowledge: [What is already known and what needs explanation]
To support further exploration, organize the answer as follows.
1. Overall synthesis
- What has been verified about the original question, with evidence.
- The current interpretation and the assumptions behind it.
- What remains unknown.
2. Related findings
- Related information that could support further investigation, with sources.
- How it connects to the original question.
- New questions it raises.
3. Next investigation
- What to check and how to check it.
Do not present related findings as isolated keywords. Include enough explanation for a first-time reader to follow the connections.
Do not add loosely related material without limit. If there are no relevant related findings, do not invent any.
Distinguish facts, inferences, and unverified points. Do not fill in unknown assumptions.
It is fine not to have a conclusion yet. Keep conditions that could change the conclusion and major uncertainties in the overall synthesis too.
For an exploratory report aimed at an unfamiliar reader, combine the context-building instructions in prompt 2 with the synthesis and related findings in prompt 3. In every case, keep conditions that could change the conclusion and important uncertainties in the main explanation.
Top comments (0)