Ask for Proof: Why We Don’t Ask for Citations in LLMS, and Why We Should

Is the marketing of LLMs as “answer engines” determining user behavior?
Unless you’re already in the habit of asking for citations in LLMs you use, the answer is probably yes.

Large Language Models, or LLMs, have been marketed as the next evolution of search: tools that don’t just find answers, but deliver them. That framing matters. While these models are remarkable at pattern recognition and linguistic mimicry, they also reflect the expectations we set for them.

The best thing about these tools is that they mostly do what you ask. If we request sources, links, dates, or even excerpts, they’ll comply. But if we don’t, they’ll simply offer an answer that sounds right, confident, well-worded, and possibly wrong.

It’s worth remembering that today’s LLM experience already looks more like a traditional search engine than it initially did. There are now citations in LLMs as an option, snippets, and even inline references in newer interfaces. The line between chat and search is blurring.

If we, as users, collectively demand proof, transparency, sourcing, and traceable facts, that accountability can be designed back in. It’s not just about training better models; it’s about shaping better behavior on both sides of the prompt.

The truth is simple: these systems aren’t providing the answer unless it’s purely quantitative, like how many teaspoons are in a cup. They’re providing the answer they think you want to hear.

That’s exactly why curiosity still matters.
Ask for proof. Ask for sources. Keep curiosity alive.