Will AI replace search? Only if it learns to show its sources

In 2023, a New York lawyer needed precedents for a case against an airline. He asked ChatGPT. It gave him six, complete with case names, courts, and citations. He asked whether they were real. It assured him they were.

None of them existed.

The judge was not amused, and the lawyers and their firm were fined $5,000. The lawyer explained afterward that he had thought he was using a kind of super search engine.

That one misunderstanding is the whole debate about AI and search.

The case for AI

People are tired of sifting through ten blue links. An AI assistant skips the list and gives you a direct answer.

The shift is real. In 2024, Gartner predicted that traditional search volume would fall 25 percent by 2026 as people turned to AI assistants. Google itself now puts AI summaries on top of its results.

The Achilles' heel

The weakness has a name: hallucination. The AI generates something that sounds right and is wrong.

A search engine that returns a bad link at least shows you where it came from, so you can judge for yourself. An AI that invents an answer delivers it in the same confident voice as a correct one.

For an individual, that's embarrassing. For an organization, it's a liability. Think of a fabricated regulation in a policy memo, or an invented statistic in a grant application.

Citations are the fix

The remedy is the one your high school teacher demanded: show your work.

AI tools that retrieve real documents and link to them give you a way to check. They aren't foolproof. A tool can cite a genuine source that doesn't say what it claims, so "it has a link" is not the same as "it's right." But a reader who can verify is in a far better position than one who can only trust.

So, will AI kill search?

Not yet. Search engines still excel at indexing vast amounts of information, ranking credibility, and offering a range of viewpoints.

The future isn't a choice between search and AI. It's a merger. Whoever masters trustworthy AI retrieval will win the next phase of the internet.

Until then

• Treat an AI answer as a lead. It isn't a fact until someone checks it.
• Prefer tools that cite sources, and click through on anything that matters.
• Set a rule. Nothing AI-generated goes to a client, a court, or the public until a person has verified it. That belongs in your AI governance.

The lawyer's real mistake was trusting a confident answer without checking it. Don't be that lawyer.

A shorter version of this article first appeared on LinkedIn. Join the discussion there.