Skip to content
NEXOTECH
Artificial intelligence

How to research a topic with AI without losing rigour

A workflow for using AI in research and assignments: map the topic, find primary sources yourself, track citations and write claims you can back up.

By NEXOTECH Editorial TeamPublished 4 min readLeer en español
In this article

AI can save you hours at the start of a research project: it helps you learn the vocabulary, see which sub-topics exist and organise what you find. But use it as your final source and your work rests on something you can neither cite nor verify. The trick is to give it the right job at each stage.

What role AI should play

Think of AI as a well-read colleague with a poor memory for where they read things. It's useful for getting oriented and for working with text you give it; it's unreliable as the origin of specific facts. That leads to one simple rule: everything that ends up in your work should come from a source you've opened yourself.

Stage 1: Map the topic

Start by asking for an overview, not facts. For example:

I'm researching the impact of remote work on productivity for a 3,000-word university essay. Give me a map of the topic: main sub-topics, open debates, technical terms I should know and the types of sources usually used. Don't include any statistics.

The map tells you what to look for and which words to search with. It's also worth asking for synonyms and the terms used in different fields — "remote work", "telework" and "work from home" pull up different literature.

Stage 2: Find primary sources yourself

Using the terms from your map, search in places suited to the type of work:

  • Google Scholar for journal articles, theses and books. Its help pages explain how to filter by date, search for an author or find free versions.
  • Library catalogues at your university or local library.
  • Official bodies for statistics and regulations.
  • Reports from recognised organisations in the field.

If your AI tool can search the web — ChatGPT search, for example — it can suggest links. Treat them as leads: open each one, check it exists, that it's the original source and that it says what it's claimed to say.

Stage 3: Read with AI, not instead of reading

Once you have real sources, AI becomes very useful again, because now it's working on text you supply:

  • Ask for a summary of a long paper to decide whether it's worth reading in full. See how to summarise documents with AI.
  • Ask it to explain a technical paragraph in plain words.
  • Ask what limitations the authors themselves acknowledge.
  • Ask it to compare two sources: where they agree and where they don't.

Then check anything important against the original. Summaries can drop key nuances, like a study's sample size or setting.

Stage 4: Track citations from day one

Every time you note an idea, record where it came from: author, year, title, page or URL, and date accessed. A simple document works, or a reference manager such as Zotero. This habit prevents the classic end-of-project problem: ideas in your draft with no idea which source they came from.

A minimal template:

Idea or factSourcePage or sectionChecked
Remote work cuts commuting time…Author (year), titlep. 12Yes

Stage 5: Write with your own judgement

The structure and conclusions should be yours. AI can help you:

  • Check whether a paragraph is clear.
  • Spot gaps in your reasoning.
  • Suggest a different order for your sections.
  • Tidy up wording without changing the content.

What it shouldn't do is add new claims without a source. If it suggests a fact, treat it as a lead and go back to stage 2.

Final checks

Before you submit, run through this list:

  1. Does every specific fact have a source I've opened and read?
  2. Are direct quotes verbatim, with page numbers?
  3. Are my sources recent enough for the topic?
  4. Have I checked contested claims against more than one independent source? Lateral reading — looking at what other sources say about a claim or a website — is a simple way to do it.
  5. Am I following my institution's rules on AI, and have I disclosed its use if required?

For more on checking facts, see our six-step method to fact-check AI answers.

Common mistakes

  • Starting by asking for conclusions. It colours everything you read afterwards. Start with the map.
  • Citing the AI as the source of a fact. A chatbot reply isn't a verifiable source for a specific fact; cite the original document.
  • Quoting from a summary. Nuance gets lost. At least read the section you're quoting.
  • Not recording access dates for web pages that may change.

The bottom line

Researching with AI works when each tool does its own job: AI orients, explains and organises; primary sources supply the facts; and you decide what matters and what you conclude. Split the work that way and you gain speed without giving up rigour. If you're a student, pair this workflow with our techniques for using ChatGPT to study.

Sources and documentation

  1. Google Scholar Help — Google
  2. ChatGPT search — OpenAI Help Center
  3. Lateral reading (Civic Online Reasoning) — Digital Inquiry Group

Information reviewed on

  • research
  • studying
  • fact-checking
More in Artificial intelligence