Making AI Correct for Business and Trade

Why It Matters

Accurate government data in AI outputs protects citizens and businesses from costly mistakes and bolsters trust in public digital services.

Key Takeaways

• 150 stale GOV.UK pages identified, slated for redirects or retirement
• New semi‑automated review cycle mandates updates at least every six months
• “Last reviewed” tags help users see content freshness and next review date
• AI surface‑tracking toolkit monitors and flags misinformation in real‑time

Pulse Analysis

Generative AI models increasingly draw on public web content, and when they surface outdated government pages, the result can be costly misinformation. The UK’s Department for Business and Trade (DBT) recognized that legacy pages—once hidden in a "0‑view abyss"—are now resurfacing in AI‑driven answers, prompting a systematic audit. By pinpointing 150 pages that hadn’t been refreshed in five years, DBT is taking a data‑driven first step toward safeguarding the integrity of public information that powers tools like ChatGPT and Copilot.

DBT’s response blends manual curation with emerging automation. The agency is redirecting obsolete pages to current guidance, archiving where appropriate, and rolling out a semi‑automated maintenance schedule that forces a review at least every six months. A new "last reviewed" notification, already tested positively with users, signals when content was updated and when the next check is due, enhancing transparency. Complementary tools, such as an AI surface‑tracking toolkit, continuously monitor how government content appears in AI outputs, flagging discrepancies before they spread.

The initiative signals a broader shift in how public sector bodies manage digital knowledge in an AI‑first era. Reliable, up‑to‑date information reduces the risk of financial errors for businesses and preserves citizen confidence in online services. As other ministries observe DBT’s pilot, the model could become a template for nationwide content governance, ensuring that AI‑enhanced interactions with government data remain trustworthy and accurate.