The Golden Retriever Problem: When AI Wants to Please You Too Much
We were researching the Canadian mining market for a grant application. It’s 2026, so naturally we used AI to help gather some initial figures.
The tool confidently produced stats: “200+ active mines, $50B+ annual GDP contribution, 40,000+ contaminated sites under management. Environmental consulting market in Canada: $3B+ annually.”
It even cited a source: Mining Association of Canada.
We asked for the reference link:
At least it was honest when pressed.
The Eager-to-Please Problem
Consumer AI tools are like golden retrievers. They desperately want to make you happy. Ask for market statistics, and they’ll fetch you market statistics—whether or not those statistics exist.
This isn’t malice. These systems are trained to be helpful, and “I don’t know” feels unhelpful. So they improvise. They pattern-match from training data, generate plausible-sounding figures, and dress them up with authoritative-sounding citations.
The citation is a costume.
How Citations Should Work
At Statvis, citations are structural.
When our system surfaces a claim, it points to a specific page in a specific document. Click the citation and you see the exact passage—highlighted, in context, on the original page.
There’s no opportunity for fabrication because the system doesn’t generate claims from patterns. It retrieves text from documents you uploaded, then shows you exactly where that text lives.
If the information isn’t in your documents, the system says so. No improvising. No costumes.
The Test
Next time an AI gives you a statistic, ask for the source link. If it can’t show you the exact document and page, you’re talking to a golden retriever—friendly, eager, and making things up to please you.
Yes, we used AI to help write this blog post. No, we didn’t let it make up the citations. The screenshot is real—we have the receipts.