Research and Source Checking

Introduction

Research is where AI-assisted content most often goes wrong. A model can sound certain about a statistic, a quote or a study that does not exist. Used carefully, AI is a good research assistant: it helps you plan, find angles and organise. It is not a source of truth.

What AI is good for in research

• Mapping a topic: "List the main sub-questions a beginner would ask about composting."
• Suggesting search terms and angles you had not thought of
• Explaining a concept in simpler words so you can understand it
• Summarising documents you supply, so you can find key points faster
• Drafting interview questions and outlines
• Spotting gaps in your draft

What it must not be trusted for

• Statistics, dates, names and quotes from memory
• Citations and links: models can invent plausible ones
• Recent events, unless connected to current sources
• Medical, legal and financial facts where errors can harm
• Claims about specific people or companies

A verification workflow

1. List every factual claim in your draft: numbers, dates, names, quotes, definitions and "studies show" statements.
2. For each, find a primary or reliable source you can open yourself: official statistics, original research, standards bodies, reputable institutions and the organisation's own documentation.
3. Read the relevant part. Check that the source actually says what you claim, in context, and for the same population and period.
4. Check the date. Is the information still current?
5. Cross-check important claims in at least two independent sources.
6. Prefer primary sources over articles that merely repeat them.
7. If you cannot verify a claim, remove it or state it clearly as an opinion or uncertainty.
8. Record the source in your log.

Judging sources

Ask:

• Who wrote it, and what expertise do they have?
• Why was it written, to inform, to sell or to persuade?
• Is it current and does it show its methods and evidence?
• Do other reliable sources agree?
• Is there a conflict of interest?

Be careful with anonymous pages, content farms and posts that cite nothing.

The source log

Keep a simple table for each article:

It makes editing faster, lets colleagues check your work, and helps you update the piece later. When you quote, use the exact words and attribute them. When you paraphrase, still credit the idea if it is not common knowledge.

Using AI with documents you supply

Paste in text you have the right to use, and ask for a summary, a list of claims or questions:

Using ONLY the text below, list the five main findings.
For each, quote the sentence it is based on.
If something is not stated in the text, write "not stated".

Then check the quotes against the original. This reduces invented details, but does not eliminate mistakes. Do not paste confidential or personal data into tools that are not approved for it.

Primary versus secondary

If a blog says "a study found 80 percent of...", find the study. Often the figure is misquoted, out of date or about something different. Linking to the real source helps readers, and signals trustworthiness.

When not to publish

If the stakes are high and you cannot verify, do not publish. Ask an expert, or reframe the piece around what you do know.

Try it yourself

Take an AI-drafted paragraph with at least five factual claims. Build the source log, verify each claim from an independent source, and note how many you had to change or remove.

Summary

• Use AI to plan, explain and summarise supplied text, not as a source of facts or citations.
• Verify every claim against sources you open yourself, prefer primary sources and check dates.
• Keep a source log, quote accurately, and remove anything you cannot verify.