Research with an AI assistant: how to get answers you can trust

An AI assistant can explain a new field in minutes, compare options you hadn’t thought of and turn a pile of notes into a clear summary. It can also be confidently wrong. The difference often comes down to method. Here is ours.

Understand what the model knows

A language model learns from a large body of text, up to a certain date. It’s excellent at explaining concepts, structuring ideas and reasoning through a problem. It’s weaker on very recent events, niche figures and exact quotations, which it may reconstruct imperfectly.

So use Nella where it’s strongest, and bring the facts yourself where precision matters.

Start broad, then narrow down

Begin with orientation questions:

• “Give me an overview of heat pumps for a cold climate: how they work, the main types and the key trade-offs.”
• “What are the main schools of thought on this question, and what does each one argue?”
• “Which terms should I search for to go deeper?”

You come away with a map of the subject and the vocabulary to explore it.

Work from your sources

For anything precise, give Nella the material instead of asking it to remember. Paste the article, the report excerpt, the regulation or your notes, then ask:

• “Summarize this text in five points, and quote the sentence that supports each one.”
• “What does this document say about cancellation fees? Answer only from the text.”
• “Compare these two contracts clause by clause, in a table.”

Asking for quotations keeps the answer anchored to the text, and makes it quick to verify.

When web search is available on your account, Nella can also look things up online and show the pages it relied on. Open them and read: a citation is a starting point, not a guarantee.

Separate facts from interpretation

Ask Nella to make its reasoning visible:

Answer in two parts: what the sources say, then your own analysis. Flag anything you’re not sure about.

That single instruction makes it much easier to see what needs checking.

Verify what matters

Not everything deserves the same scrutiny. A list of brainstorming ideas? Use it freely. A medication dose, a tax deadline, a legal obligation, a figure you’re about to publish? Check it against a primary source: the official text, the manufacturer, the government website, the original study.

Three follow-up questions that pay off:

• “What would make this answer wrong?”
• “What are the strongest counter-arguments?”
• “Which part of this answer is most likely to be out of date?”

Keep a research trail

For longer projects, keep one conversation per topic and give it a clear name. Ask Nella to maintain a running summary: “Update the summary with what we’ve learned, and list the open questions.” When you’re done, Export my data in Settings › Data gives you your conversations in Markdown, ready to file alongside your notes.

The right mindset

Treat Nella like a brilliant research assistant in their first week: fast, tireless, widely read, and in need of supervision on the details. You remain the editor-in-chief. With that division of labour, you’ll research faster and better. To phrase your questions well from the start, see how to write better prompts.