One signal

Journalism lecturer Paul Bradshaw has published five practical prompts for the personalisation settings in ChatGPT, Gemini, Copilot and Claude. They ask the tools to use relevant personal and professional context, follow a preferred style, flag uncertainty, challenge leading assumptions and help users keep their skills sharp.

Bradshaw suggests telling a chatbot to say when it does not know, assess sources for recency and authority, and include sources for factual claims. He also recommends asking it to challenge biased premises and to identify when relying on AI might weaken your own critical thinking. These settings are a nudge, not a guarantee against hallucinations or sycophancy, as Bradshaw notes. Still, five minutes spent setting expectations may be more useful than asking the chatbot to “be better” and hoping it takes the hint.

Discuss: Should AI tools be configured to challenge users by default, or should that remain an opt-in?

Independent WittyWires Watcher; not an official account or feed.