How to research a topic and save the sources to a file

Write a short summary and a list of source links into one Markdown file, with each claim in the summary pointing to the source it came from. With an AI agent, ask for exactly that: a summary, a sources list, and the file name to save it to.

A research note you can trust has two parts: what you found, and where you found it. Saving both in a plain text file means you own it, can search it later, and can check any claim against its source in one click.

Steps by hand

1. Write the question down first. A narrow question gives better research. "What changed in the EU's rules on X in 2026, and when does it apply?" beats "tell me about X".

1. Create the file. Markdown is plain text, so it opens anywhere and works in git.

mkdir -p ~/research && cd ~/research
touch topic-name.md

On Windows, mkdir %USERPROFILE%\research in Command Prompt, or New-Item -ItemType File topic-name.md in PowerShell.

1. Use this structure.

# <Your question>

Researched: 2026-10-25

## Summary
- Key finding one. [1]
- Key finding two. [2]
- Open question or disagreement between sources. [1][3]

## Sources
1. <Title>, <publisher>, <date published>. https://example.com/page
2. ...

The numbers tie each finding to its source.

1. Prefer primary sources. The official documentation, the law itself, the original study, the company's own announcement. Use news articles and blog posts to find primary sources, then cite the primary one.

1. Note the date of every source. Facts change. A date tells future you whether a source is still current.

1. Write the summary in your own words. Short quotes are fine when the exact wording matters; put them in quotes and cite them.

1. Check that every link works. Pull the links out of the file and test each one. On Mac and Linux:

grep -oE 'https?://[^ )>]+' topic-name.md | while read -r u; do
  curl -s -o /dev/null -L -w "%{http_code} $u\n" "$u"
done

A 200 means the page loaded. On Windows 10 and 11, curl.exe is built in; in PowerShell type curl.exe rather than curl, which can mean something else there.

1. Keep it with your work. Put the file in the project it belongs to, or in a notes folder you back up or keep in git.

Doing it with Mel

Mel's agent has a web search tool, so it can look things up on the live web in the middle of a conversation, with no search key to set up. Press ctrl+enter (cmd+enter on macOS) and ask for the result as a file:

Research: what are the current recommended settings for <topic>?
Use official documentation where possible. Write research/<topic>.md with
a short summary where each point cites a numbered source, then a Sources
list with title, publisher, date and URL. Say clearly where sources disagree.

What to expect:

• Searches show in the conversation as tool cards with the query and the results: each hit's title, URL and snippet. Searches only read, so they run without an approval card.
• Answers that use search end with a Sources list of the links the agent actually used.
• Saving the file is a write, so with Mel's approvals it shows on an approval card, and the result shows as a diff card you can read.
• Snippets, not full pages. The search gives the agent result titles, links and short summaries. That covers many "what is current" questions; for anything important, open the source and read it yourself.

Web searches have a daily cap that depends on your Mel plan. The agent can also turn the note into a styled PDF report with its PDF tool when you want something to send.

FAQ

What is the best way to save research sources? A plain Markdown file with a numbered sources list: title, publisher, date and URL for each. It is easy to search, share and keep in git.

How do I cite websites in my notes? Give the page title, the publisher, the date it was published or updated, and the full URL. Add the date you read it if the page changes often.

Can I trust AI research? Use it as a fast first pass. Check every important claim against its source, and prefer primary sources over summaries.

How do I check that my links still work? Run the curl loop in step 7. Anything other than 200 deserves a look.

Why Markdown and not a document app? Markdown is plain text: it opens in any editor, works with git, and will still be readable in twenty years. You can convert it to other formats when you need to.

• How to pull tables or totals out of many PDFs
• How to write a task for a coding agent
• How to search for text in files from the terminal
• What is an AI desktop agent?