AI Hallucinations in Design Research - How to Spot and Avoid Them

Content Authenticity Standards
1

AI hallucinations have been causing real problems in design research workflows and I don’t see enough discussion about it in this community.

Two specific examples from my own recent work:

Asked an AI assistant for statistics on accessibility adoption rates in digital product design. Got confident numbers with plausible-sounding source citations. Spent 30 minutes tracking down those sources. Two of them didn’t exist. One existed but the statistic cited was not in it.

Asked for examples of brands that had successfully used a specific design approach. Got a mix of real and fabricated case studies. The fabricated ones were indistinguishable from the real ones in how they were presented.

The problem is the confidence. Wrong information delivered tentatively is catchable. Wrong information delivered with the same authority as correct information is dangerous - especially when you’re presenting research to clients or in academic contexts.

My current rule: never use AI-generated statistics or case studies without primary source verification. AI is good for synthesis of known information. It’s unreliable for specific claims.

How are others handling this in their research workflows?

2

The confidence problem is the key issue. I’ve started treating AI research output the way I’d treat an anonymous tip - useful for direction, not for citation. Follow the lead, find the real source.

3

Primary source verification rule is correct and should be standard. Any statistic or case study that will be presented externally needs to trace back to a verifiable original source. AI is a research assistant, not a reference.

4

@Snaxx_TechGrid non-existent citations are the worst version of this. It’s not even paraphrase error - it’s fabricated sources. That’s a credibility disaster if it gets into a client presentation or academic submission.

5

I build a verification step into my research workflow explicitly now. AI draft → flag all statistics and named sources → verify each before using. Adds time but the alternative is presenting false information as researched fact.