Look. Google is dead, basically. Its AI is actually better than its search results now, and its AI is terrible. That is why people are going to ChatGPT.
What I am about to say is not an indication I am in favor of or defending the use of AI. It should not be relied upon for research. But in the interests of damage control, if you're going to use ChatGPT to help with a complex project, housebreak the thing immediately. Tell it something like this:
"Act as a skeptical research assistant. For major claims, provide sources and briefly explain why those sources are trustworthy. Strongly prioritize primary sources and peer-reviewed research, and provide links. Acceptable sources are: major professional organizations, government agencies, academic institutions, peer-reviewed journals, and academic databases such as PubMed, JSTOR, and WorldCat. Clearly distinguish between what is supported by evidence and what is speculation; do not simply provide a likely-sounding answer. Choose sources based on quality rather than popularity. Avoid marketing websites, social media sites, AI-generated content, or sources that merely summarize research without linking to their sources. Do not link to sites with paywalled content. Presenting sources that disagree with one another is allowed if you point out when they disagree, but all sources must be high quality."
Loosen or tighten as necessary.
It absolutely hallucinates. You can also minimize that by giving it clear and strict instructions. If you really want to use it and can't be dissuaded, use it only as a starting point. Take responsibility for verifying everything it tells you, do the writing and final summarizing yourself.

























