Use AI to answer questions in your notes

TL;DR

Clone a pre-built Reflect AI prompt and edit its instructions so it returns an answer instead of repeating the question.

Briefing Cornell Notes

Briefing

A practical way to turn an AI assistant inside Reflect Notes into a “question-and-answer” and “command-execution” tool is to build custom prompts that control both the output format and how much text the assistant is allowed to generate. Instead of selecting a pre-built prompt that merely echoes a question back, the workflow centers on cloning an existing prompt and rewriting its instructions so the assistant returns a clean answer directly into the note.

The setup starts with a hypothetical article and a question the user wants answered from within their notes. After selecting the AI assistant, the default behavior can be unhelpful—returning the question rather than answering it. The fix is to choose a prompt template that’s closer to the intended behavior (for example, a “prompted cell” style), then clone it so it can be customized without starting from scratch. The key changes are in the prompt text: the instructions are rewritten to “answer the question” using plain, simple language, while keeping the response concise and constrained to the content the user needs.

To make the output reliable, the prompt is configured with strict formatting rules. The assistant is told to use the input area (shown as triple quotes) as the question or parameters, and to return only the answer—no extra commentary, no surrounding quotation marks, and no fluff. Once saved, running the new prompt produces an actual answer inserted into the note. The user can then iterate by adding parameters such as length control—for instance, requiring “three to five” sentences—then re-running the prompt to see whether the response matches the desired depth. This “tune and rerun” loop is presented as the fastest path to a prompt that fits a specific note-taking style.

A second custom prompt shifts the assistant from answering questions to executing commands. Instead of asking for capabilities in a question form (which would yield a less useful response), the workflow uses a dedicated “execute command” prompt that generates a clean, structured list tailored for insertion into an article or notes. The prompt’s instructions again emphasize output discipline: avoid including the prompt text itself (including the triple-quote wrapper) and return only the requested result. The result is a formatted list that can be dropped directly into writing, supporting tasks like article outlines, research summaries, and reusable note templates.

Overall, the approach treats Reflect’s AI assistant like a ChatGPT-style interface—typing prompts and getting actionable outputs—but with tighter control over what gets returned. With a small set of reusable, cloned prompts, users can repeatedly generate concise answers or ready-to-paste lists on demand, turning notes into a workspace where AI can both respond and produce structured content.

Cornell Notes

Custom prompts in Reflect Notes can turn an AI assistant into a reliable note-taking tool that either answers questions or executes commands. The workflow starts by cloning a pre-built prompt and rewriting its instructions so the assistant returns only the answer, in plain language, without extra quotes or commentary. Output quality improves by iterating on parameters like response length (e.g., three to five sentences). A separate “execute command” prompt produces cleaner, insert-ready lists—useful for article outlines and structured research—by instructing the assistant to return only the requested result and omit prompt text. This makes AI outputs consistent enough to reuse regularly inside notes.

Why does the assistant sometimes just repeat a question instead of answering it, and how is that fixed?

Selecting an AI assistant with a basic “answer this question” setup can lead to a prompt that echoes the question back. The fix is to clone a prompt template that’s closer to the desired behavior (a “prompted cell” style), then customize the cloned prompt so it explicitly instructs the assistant to “answer the question” and to output only the answer. Once saved, running the prompt generates the actual response inserted into the note.

What formatting constraints make the AI output usable inside notes?

The custom prompt includes strict output rules: it tells the assistant to use the triple-quoted input area as the question/parameters, respond in plain and simple language, and “do not return anything other than the answer.” It also specifies not to wrap responses in quotes. These constraints reduce fluff and prevent the assistant from adding meta-text that would clutter the note.

How can response length be controlled without rewriting the whole prompt?

After cloning and saving the initial “answer the question” prompt, additional parameters can be added—such as requiring “three to five” sentences. Running the prompt again applies the new length constraint, producing a more detailed answer while still keeping the output concise and note-friendly.

Why use an “execute command” prompt instead of asking a question like “what can large language model tools do”?

Asking for capabilities in question form can yield a less useful response for direct insertion. The “execute command” prompt is designed to generate a clean, structured list of items tailored for note-taking and writing. It’s built to return only the requested list content, making it easier to paste into an article or outline.

What does the “execute command” prompt do differently from the Q&A prompt?

The Q&A prompt is instructed to answer a question concisely. The “execute command” prompt is instructed to execute the requested task and output the result (like a comprehensive list) in a clean format. Both prompts rely on the same general structure—triple quotes for inputs and strict “return only the result” rules—but the intent changes from answering to producing an insert-ready output.

Review Questions

When cloning a prompt, what specific instruction changes ensure the assistant returns an answer rather than echoing the question?
How would you modify the prompt to require a longer response (e.g., 5–7 sentences) while still preventing extra commentary?
What output rule would you add to keep the assistant from including prompt text (including triple-quote wrappers) in the final inserted result?

Key Points

1
Clone a pre-built Reflect AI prompt and edit its instructions so it returns an answer instead of repeating the question.
2
Use strict output constraints like “return only the answer” and “do not wrap responses in quotes” to keep notes clean.
3
Control response depth by adding parameters such as “three to five” sentences, then rerun the prompt to compare results.
4
Create a separate “execute command” prompt for tasks that need structured, insert-ready outputs like lists.
5
In command-style prompts, explicitly prevent the assistant from including prompt text (including triple-quote wrapper content) in the final output.
6
Treat prompt iteration as a workflow: adjust wording and parameters, rerun, and keep the version that matches the desired note-taking style.
7
Maintain reusable prompts in notes so AI can be called on demand for Q&A and structured writing tasks.

Highlights

Cloning and rewriting a prompt can change AI behavior from echoing a question to generating a direct answer inserted into the note.

Telling the assistant to return only the answer—and not include quotes or extra commentary—makes outputs practical for real writing.

Adding a sentence-count parameter (like three to five) lets users dial in depth without rebuilding the prompt from scratch.

A dedicated “execute command” prompt produces clean lists that can be pasted directly into articles and outlines.

Explicitly banning prompt-text echoes (including triple-quote content) prevents the assistant from adding clutter to the final result.

Topics

Reflect Notes AI prompts
Question Answering
Command Execution
Note-Taking Workflow
Prompt Engineering