Blog • Workflow

How I Research a Topic With AI Without Trusting It Blindly

A five-step research routine that uses AI to map a topic and find questions, then relies on real sources for every fact.

By Akesh Chandrasiri20 October 20263 min read

AI can make research faster, and it can make it worse. It is quick at explaining a topic and just as quick at inventing a detail that sounds right. The trick is to give it the jobs it does well and keep the facts for sources you can open.

Here is the routine I use for the topics I write about, including the Blockbuster and Netflix story.

Step 1: Ask for a map, not answers

Start by asking the AI to lay out the topic, not to state facts:

I'm going to write about [TOPIC]. List the main questions a curious reader would ask, the common claims people make about it, and the claims that are often disputed. Do not state any numbers or dates.

The "no numbers or dates" rule keeps it on questions and stops it handing you figures you'd have to unpick later. For Blockbuster and Netflix, the useful output is a list of disputed claims, such as whether Blockbuster really passed on a chance to buy Netflix.

Step 2: Turn the map into a reading list

Take each question and find a real source. For company stories, I look for:

  • Company filings and earnings releases.
  • Reporting from established outlets, preferably with named sources.
  • Interviews or books by the people involved, remembering that they have their own point of view.

Open each source yourself. A search snippet is not a source.

Step 3: Write your own notes from what you read

Write short notes in your own words, with the source next to each one. This is the material you will actually use. For example, from my Blockbuster research:

  • Netflix offered to sell to Blockbuster for $50 million in 2000, per CNBC's reporting.
  • Blockbuster collected about $800 million in late fees in 2000, per the same report.

Notes like these are checkable. If someone challenges a line, you know where it came from.

Step 4: Use AI on your notes, not on its memory

Now the AI is useful again. Give it your notes and ask it to work only from them:

Using only the notes below, draft an outline of the article. If a section needs a fact that isn't in the notes, put [NEED SOURCE: what] there. Do not add facts.

NOTES: [paste]

You get the speed of AI without the invention, because it is working from material you have already checked.

Step 5: Ask what would prove you wrong

Before publishing, I ask:

Here is my draft. What are the strongest counterarguments or the most likely corrections a knowledgeable reader would raise? Don't tell me who is right; tell me what I should check.

This often points to a claim I stated too strongly. In the Blockbuster story, it is the temptation to say the $50 million offer was the cause. The reporting I read points to a more tangled story, so I changed my wording.

Signs the AI is guessing

  • Round, confident numbers with no source.
  • Names of studies or reports you can't find.
  • Quotes with no date or place.
  • A very smooth explanation of something that is genuinely disputed.

When you see these, don't argue with the AI. Go to the source.

The takeaway

Use AI to find the questions and organise your notes. Use real documents for the answers. The more a sentence matters, the more it needs a source you opened yourself.