\n\n
Recent independent studies reveal a stark divide in how reliably AI chatbots handle news content. The BBC and European Broadcasting Union found that Gemini’s sourcing‑inaccuracy rate sits at 72%, while Perplexity and Copilot achieve just 15%, with ChatGPT at 24% 【"On the specific measure of sourcing inaccuracy…"】. This 72% vs 15% gap underscores the growing risk of misinformation when AI summarises news.
In response, the research community is tightening controls. The preprint server arXiv has begun limiting submissions to curb the flood of AI‑aided papers 【"started limiting submissions to mitigate the tsunami of A.I.-aided results"】, and citizen‑science platforms such as iNaturalist are adding safeguards against fabricated observations.
Search providers are also stepping up. Google announced upgrades to its AI‑driven search, moving source links higher in responses to make fact‑checking easier 【"AI responses will display links more prominently, so users can fact-check"】. These changes come as news publishers report a 40 % traffic decline from Google over the past year 【"news publishers saw a 40 per cent drop in traffic from Google"】, highlighting the commercial pressure to preserve attribution.
The BBC’s own methodology—testing 100 BBC stories across four major chatbots—provides the backbone for these findings 【"BBC journalists put 100 BBC news stories through…"】 and signals that more rigorous evaluation will be essential as AI tools proliferate.
What to watch next:
- Further rollout of arXiv submission limits and similar policies at other preprint servers.
- Adoption of Google’s enhanced source‑link UI across other AI platforms.
- Ongoing independent benchmarking studies (e.g., by the BBC, EBU, and the Columbia Journalism Review) to track accuracy trends.
- Potential regulatory moves aimed at mandating transparency and attribution standards for AI‑generated news. [[1, 2]]