Improving AI Research through better sources
In late 2024, a
major piece of research by DW was released that reinforced a suspicion that many of us already had – AI struggles to distinguish fact from opinion. This, combined with the findings of a BBC study from February 2025 that revealed that AI assistants misrepresent news and research 51% of the time, has reinforced the importance of including sources in your AI responses.
As you’d expect from high level design and technology partners, we’re constantly reviewing outputs from AI and how we can improve the information we’re sharing internally and with clients, and constantly adjusting guidelines to our teams about what we expect from them before they submit research. We’ve included the two core instructions below that have greatly increased the value of the information we’re getting, and not just the speed:
– Always show your working: notes should include the verification of individual facts, from asking the AI to include dates for sources and to check for multiple links from trusted outlets to the original source. manually verified and with multiple sources provided.
– Don’t just copy and paste the answer: read and challenge it before sharing. Whenever you read research, you almost always have additional questions or red flags that seem to challenge the findings. Find the answers to those questions before passing on your work.
