Use AI Without Outsourcing Your Judgment
Research Like a Professor · Guide 9 of 9
AI chat tools can be good research partners. They can also say false things in a confident voice, and make up sources that look real. Used well, they speed up the parts of research that don't need your judgment, so you have more time for the parts that do.
What AI is good at, and what it gets wrong
An AI chat tool predicts likely words. That makes it fluent, fast and tireless, and it means it can produce a sentence that sounds right without any way of knowing whether it is.
It's good at work with language:
- suggesting search words you hadn't thought of
- explaining a hard passage you paste in
- summarizing a text you give it
- turning your notes into a cleaner outline
- formatting a citation you've already checked
It gets these wrong, often confidently:
- Facts without sources. Dates, names and numbers can be invented or blended.
- Citations. It can produce articles that don't exist, with real-sounding authors and journals.
- Recent events. Its knowledge stops at a training date unless it's searching the web.
- Local history. Small places are thin in its training. It may fill a gap with a plausible guess about your county rather than admit it doesn't know.
The pattern is simple. Trust it more with what you give it, and less with what it tells you.
Five rules of thumb
- Bring your question first. Decide what you're asking (guide 1) before you open the chat. Your question should decide what gets read, not the tool's suggestions.
- A summary of a title is not a reading. If an AI describes an article it hasn't been given, it's guessing from the title. Paste in the text, or read it yourself, before you rely on what it says.
- Open every source it names. Ask for sources, then find each one through guide 2 or guide 3. If you can't find it, assume it doesn't exist. Cite the source you opened, never the AI.
- Check claims the way you'd check a stranger's. The moves in guide 5 apply: trace the claim to the original, and find better coverage.
- Say when you used it. A short note, such as "I used an AI tool to suggest search terms and check my summary," is honest, and it's increasingly expected.
The principle underneath all five: let the machine assist, but keep a person in charge of verifying, and write down what you did. Your judgment is the part of research that can't be outsourced.
These rules come from my own practice. How I Create shows them at work, including the errors the checking caught.
AI across the series
Each earlier guide has a step where an AI tool can help, and a check that keeps you in charge.
| Guide | A good use of AI | The check that stays yours |
|---|---|---|
| 1. Know What You're Asking | Suggest narrower versions of your question, and alternate search words | You choose the question that's actually yours |
| 2. Find It | Explain a search result you don't understand | You find the sources in Scholar yourself |
| 3. Get It | Draft a polite email asking an author for a copy | You confirm the article exists before you send it |
| 4. Read It | Explain a hard paragraph you paste in | You write the claim, the evidence, and the meaning in your own words |
| 5. Check It | List questions to ask about a source | You do the lateral reading |
| 6. Read the Past | Explain an old term or abbreviation in a record | You confirm it in another source from the period |
| 7. Find the Numbers | Explain what a statistic measures | You get the number from the agency that produced it |
| 8. Keep Track | Format a citation from details you supply | You check it against the style guide |
This is the end of the series. Start again at guide 1 whenever a new question finds you.
Research Like a Professor is a nine-part series. See all nine guides.