How can I tell if a Generative AI system (chatbot) got its information from a specific website or other source?

AIs are often unreliable in identifying the source of the information they provide. Frequently, the chatbot’s response is a lightly edited version of a single source, even if the chatbot claims otherwise. While these chatbots often will say they compile info from multiple sources, there are times when this is inaccurate and the answer is in fact derived from a single source. 

If the information comes from a single source, it's best to be able to view and cite that source, not the chatbot.

Ways to guess or investigate whether information may have come  from a particular single source:

• Two or more chatbots produce nearly identical paragraphs
• A more efficient way of discovering if this has happened (and of consulting the original source!) is to paste language from the answer into a search engine like Google. For instance, you may find the web page where the information was taken from. 

Some AIs will always cite their sources; others will do it sometimes; some AIs have "scholarly" versions that are meant to cite their sources. Each AI chatbot has different levels of reliability, and different prompts will get different results. They all work different ways. (Harvard faculty, staff and students may request a consultation with a librarian to review the chatbot you’re interested in.)

It is important to verify these sources. In addition to sometimes being set to speculate, AIs are only as correct as their training data, and any part of that data can be contradictory, incorrect, or out of date. While AIs are amazing, and can revolutionize our work, it's important to be aware of both their features and limitations.