Research and Exploration

An AI Chatbot Is Not a Search Tool

There is one thing to understand before using AI for research, and it prevents most of the trouble people get into: an AI chatbot is not a search tool. It is not looking through a database and reporting what it found. It is generating text that resembles what an answer would look like. Those are completely different operations, and confusing them is how people end up citing papers that were never written.

When you search a library database, you get records of things that exist. You can open them, check them, and cite them.

When you ask a general AI chatbot for sources, you get text shaped like citations. Some may correspond to real work. Some will be real papers attached to claims they never made. Some will be entirely invented — with plausible authors, a real journal, a volume and page range, and occasionally a DOI that resolves to nothing or to something unrelated.

The invented ones are not marked. They look exactly like the real ones.

This is not a bug that will be fixed in the next version; it follows from how the technology works. Some newer tools do retrieve real documents and link to them, which helps — but you still have to open the link, because a tool that retrieves can still summarize inaccurately.

The practical rule: never let an AI tool be the reason you believe a source exists.

A Workflow That Actually Works

Invert the usual order. Find sources the reliable way, then use AI on material you already have.

When you are new to a topic, the hard part is vocabulary. You cannot search well for something you cannot name.

Ask an AI tool: what are the main debates in this area? What terminology would researchers use? What are the major schools of thought? What related fields touch this?

Treat every specific claim as unverified — you are collecting search terms and a mental map, not facts. This alone can save an hour of flailing in a database.

Take the vocabulary from step 1 into the Maricopa Libraries databases, Google Scholar, or a subject-specific index.

This is also the moment to ask a librarian. Maricopa librarians will help you build a search strategy, and they are noticeably good at the "I have a vague topic and need to narrow it" problem. This is their actual job, not an imposition.

Now that you have real documents, AI becomes genuinely valuable:

  • Summarize a dense paper before deciding whether to read it fully
  • Explain a methods section in a field you do not work in
  • Compare several papers you have gathered and identify where they disagree
  • Pull out the definitions being used, which often differ between authors
  • Generate questions to ask of a text

Tools built for this — NotebookLM, for instance — answer only from sources you supply and tell you which one each claim came from. That is a fundamentally safer design than asking a general chatbot what it remembers, and it is provided by the district.

Even so: AI summary is not a substitute for reading what you cite. If it is going in your bibliography, you have read it.

AI is useful for finding weaknesses in your own argument — ask what a critic would say, what you are assuming, what evidence would undermine your position.

The caveat: these tools tend toward agreement. Ask "is my argument good?" and you will be told it is. Ask "what is the strongest case against this?" and you get something useful. Phrase it as a challenge or you will get flattery. See Critical Evaluation Skills.

Verifying What You Get

  1. Search the title in a library database or Google Scholar — not the whole citation, which may be malformed.
  2. If nothing, search the author name and check their publication list.
  3. If the paper exists, open it and confirm it says what was claimed. Real citations attached to invented claims are common and harder to catch than pure fabrications.
  4. If you cannot find it, it does not exist. Do not cite it, and do not assume the library simply lacks access — check with a librarian, who can tell you the difference.

A caught fabrication is worth mentioning to your instructor or colleagues. It is the most convincing demonstration available of why this matters.

  • A citation that fits your argument a little too neatly
  • Statistics with no attributable source
  • A confident claim about something very recent
  • An author name you cannot find anywhere
  • Anything about Maricopa specifically — our programs, policies, or data
  • A summary that is smoother and more conclusive than the messy reality of most research

That last one deserves attention. Real scholarship disagrees with itself. A tidy consensus is often a sign that nuance has been smoothed away.

For Faculty

Watching a student verify an AI-generated citation and discover it does not exist is one of the more effective teaching moments available right now. It makes the abstract warning concrete in a way that saying "be careful" does not.

Consider building it in deliberately: generate a short annotated bibliography with AI, hand it to students, ask them to verify every entry and report what they find. It teaches source evaluation, database skills, and appropriate skepticism in a single exercise.