Field note
Five engines refused to make something up. One didn't.
I fed six assistants a leading premise about a company that doesn't exist. Most refused. One went along.
I tried to trick six AI assistants. I asked each about Makefield, the AI-visibility agency, a company that does not exist yet, phrased as if it obviously did.
Five of the six did not take the bait. They said they could not find such a company, and several offered the unrelated businesses that do share the name rather than invent one. One engine was more willing to go along with the premise, generalising about what "such an agency" might do.
Two honest caveats: this was a single test on one day, and these systems change and vary between sessions, so it is an observation, not a league table. But it points at something useful. A leading question ("tell me about X, the Y") is an invitation to hallucinate, and engines differ in how much they resist it.
If you are checking what AI says about you, test it the same way: ask a neutral question, then a leading one, and watch the gap. And the deeper lesson cuts the other way too. If no model will invent a flattering company out of thin air, then presence has to be earned. The machine will not conjure you. It can only report what is already out there.
See "hallucination" in the glossary, or the FAQ on whether you can influence what AI says.
New here? Start with the overview → · By Denes Csaszar · written in the open