Improving AI Research through better sources
A
major investigation by DW in late 2024 reinforced something many of us already suspected: AI struggles to distinguish fact from opinion. A BBC study published in February 2025 added weight to that concern, finding that AI assistants misrepresent news and research 51% of the time.
Speed is useful. Accuracy is essential. The two don’t always travel together.
We’ve been reviewing our own approach to AI-assisted research, both for internal work and for what we share with clients, and have updated our guidelines accordingly. Two instructions in particular have made a meaningful difference to the quality of what we produce.
1. Show your working. Notes should include verification of individual facts. Ask the AI to cite dates and sources, then check that those sources link back to trusted, original outlets. Don’t treat a response as verified until you’ve confirmed the references yourself.
2. Read it before you share it. Copying and pasting an AI response without reading it carefully is where things go wrong. Good research almost always raises further questions or surfaces details worth challenging. Answer those questions before passing the work on.
The goal isn’t to slow things down. It’s to make sure that what we share is actually worth sharing.
