Research a topic with AI without trusting made-up facts

What do you need before you start?

Gather the items below before you open the chat. You already have most of them.

• One clear question (who, what, when, or should we…)
• A free chat that can search the web, or one where you will paste pages yourself
• 15 minutes to open the sources, not just read the summary

How do you frame the brief?

Tell the chat who the brief is for and what decision or learning goal you have. Vague questions get vague answers.

I need a research brief for [me / my team / a class discussion].
Question: [one clear question].
Audience: [beginner / coworker who already knows the basics].
Depth: short. About 400 words plus a source list.

Give me:
1. A 5-bullet answer to the question
2. For each bullet, the claim and a source (title, publisher or site, and URL if you have one)
3. A "check these yourself" list: names, dates, numbers, and quotes I must verify
4. An "I am unsure" section: anything you cannot back with a source
5. Three follow-up questions I should ask next

Do not invent sources. If you cannot name a real source for a claim, put the claim only under "I am unsure."
Do not give medical, legal, or financial advice. Point to public pages instead.

How do you open every source?

Click each link. Skim for the claim. If the page does not support it, cut the claim from your notes. Prefer primary pages (official docs, papers, news reporting) over random blogs.

If the chat cannot browse, paste one page at a time:

Here is a page I opened: [paste title and URL, then key paragraphs].
Does this support claim: "[paste claim]"?
Answer yes, no, or partly, and quote the exact lines. Do not add outside facts.

How do you pressure-test the brief?

Copy the prompt below and fill in the brackets. Send only that, then stop.

Here is my edited brief after I checked sources:
[paste your notes]

Argue against it. What is weak, outdated, or one-sided?
List 3 things a skeptical reader would ask.
Do not invent new sources. Mark anything you are unsure about.

How do you write the version you will use?

Keep only claims you verified. Add the date you checked. Share that version, not the first chat reply.

Rewrite my verified notes as a 200-word brief plus a bullet source list.
Use only what I kept. Do not add new facts or links.

What should you check before you use it?

Answer each line below. If one is still a no, do not use the result yet.

Director check: look before you use it

• Did you open the sources, or only read the chat? If you did not open them, you do not have research yet.
• Would you bet your name on every number? If not, cut it or label it as unchecked.
• Is this medical, legal, or money advice? Stop. Use a professional and primary sources.
• Did the chat agree with you too easily? Run Step 3 again and ask what is wrong.

What is one extra tip?

For everyday research, turn on web search or a "deep research" mode if your free plan has it, then still open the pages yourself. For catching made-up facts in any reply, use Catch facts the AI made up. For a long PDF you already have, use Summarize a long report. More work guides are on Work.