Felix M. Simon
@felixsimon.bsky.social
Research Fellow in AI and News, Reuters Institute, Oxford University | Research Associate & PhD, Oxford Internet Institute | Junior Research Fellow, Corpus Christi College | AI, news, (mis)info, democracy | My views etc…
https://www.felixsimon.net/
https://www.felixsimon.net/
Although this is a single‑organisation case study, we think there are some “no‑regret” principles that can be useful to other organisations:
November 7, 2025 at 4:46 PM
Although this is a single‑organisation case study, we think there are some “no‑regret” principles that can be useful to other organisations:
Intriguing here is also the question of how much longer AI transparency will be required, especially with a view to the actions of tech companies.
November 7, 2025 at 4:45 PM
Intriguing here is also the question of how much longer AI transparency will be required, especially with a view to the actions of tech companies.
4️⃣ Persistent challenges include achieving consistent labelling (especially on mobile), breaking organisational silos, keeping pace with evolving models and norms, guarding against creeping human over‑reliance, and mitigating against “transparency backfire” where disclosures reduce trust.
November 7, 2025 at 4:45 PM
4️⃣ Persistent challenges include achieving consistent labelling (especially on mobile), breaking organisational silos, keeping pace with evolving models and norms, guarding against creeping human over‑reliance, and mitigating against “transparency backfire” where disclosures reduce trust.
3️⃣ Nine factors shape what, when & how the FT discloses AI use. These include legal/provider requirements, industry benchmarking, the degree of human oversight, the nature of the task, system novelty, audience expectations & research, perceived risk, commercial sensitivities and design constraints.
November 7, 2025 at 4:45 PM
3️⃣ Nine factors shape what, when & how the FT discloses AI use. These include legal/provider requirements, industry benchmarking, the degree of human oversight, the nature of the task, system novelty, audience expectations & research, perceived risk, commercial sensitivities and design constraints.
✨New working paper on the trade-offs involved in AI transparency in news 🤖📝
Based on a case study of the @financialtimes.com, Liz Lohn and I argue that transparency about AI in news is a spectrum, evolving with tech, commercial, professional & ethical considerations & audience attitudes.
Based on a case study of the @financialtimes.com, Liz Lohn and I argue that transparency about AI in news is a spectrum, evolving with tech, commercial, professional & ethical considerations & audience attitudes.
November 7, 2025 at 4:45 PM
✨New working paper on the trade-offs involved in AI transparency in news 🤖📝
Based on a case study of the @financialtimes.com, Liz Lohn and I argue that transparency about AI in news is a spectrum, evolving with tech, commercial, professional & ethical considerations & audience attitudes.
Based on a case study of the @financialtimes.com, Liz Lohn and I argue that transparency about AI in news is a spectrum, evolving with tech, commercial, professional & ethical considerations & audience attitudes.
Time to let the cat out of the bag that I’ve been elected to a Junior Research Fellowship at Corpus Christi College, Oxford, alongside my position at the Reuters Institute.
November 7, 2025 at 11:01 AM
Time to let the cat out of the bag that I’ve been elected to a Junior Research Fellowship at Corpus Christi College, Oxford, alongside my position at the Reuters Institute.
Some data on public views (which goes towards James' point)
November 6, 2025 at 10:43 AM
Some data on public views (which goes towards James' point)
*tries not to make pun about the commenters going nuclear*
November 5, 2025 at 2:16 PM
*tries not to make pun about the commenters going nuclear*
Excellent addition by David Caswell made over on LinkedIn: If you ignore the flaws and look at the numbers in the inverse, so to speak, they actually did find progress on news representation within LLMs.
October 23, 2025 at 12:56 PM
Excellent addition by David Caswell made over on LinkedIn: If you ignore the flaws and look at the numbers in the inverse, so to speak, they actually did find progress on news representation within LLMs.
Thank you, David :-) @richardfletcher.bsky.social will be be able to confirm, but I'm afraid the answer is "No" given the way we asked. Given the primacy of the big systems, one could reasonably speculate it will mostly be one of them, however.
(And yes, read the report ;-)
(And yes, read the report ;-)
October 7, 2025 at 8:48 AM
Thank you, David :-) @richardfletcher.bsky.social will be be able to confirm, but I'm afraid the answer is "No" given the way we asked. Given the primacy of the big systems, one could reasonably speculate it will mostly be one of them, however.
(And yes, read the report ;-)
(And yes, read the report ;-)
Likewise, and perhaps again not popular new for many outlets, seeing AI labelling on news is infrequent relative to daily news use. Only 19% see AI labels daily and 28% weekly – a low number considering that 77% say they use news daily.
October 7, 2025 at 7:01 AM
Likewise, and perhaps again not popular new for many outlets, seeing AI labelling on news is infrequent relative to daily news use. Only 19% see AI labels daily and 28% weekly – a low number considering that 77% say they use news daily.
And despite a growing number of outlets introducing audience-facing AI, most people don’t yet recall seeing these AI features. Sixty per cent say they do not regularly see AI features on news sites or apps. Most common are AI summaries (19%) and AI chatbots (16%).
October 7, 2025 at 7:01 AM
And despite a growing number of outlets introducing audience-facing AI, most people don’t yet recall seeing these AI features. Sixty per cent say they do not regularly see AI features on news sites or apps. Most common are AI summaries (19%) and AI chatbots (16%).
People have limited confidence in human oversight of AI in news. Only 33% think journalists ‘always’ or ‘often’ check AI outputs before publishing. Trust in news strongly correlates: 57% of those who ‘strongly trust’ news think such checks happen (19% strong distrusters).
October 7, 2025 at 7:01 AM
People have limited confidence in human oversight of AI in news. Only 33% think journalists ‘always’ or ‘often’ check AI outputs before publishing. Trust in news strongly correlates: 57% of those who ‘strongly trust’ news think such checks happen (19% strong distrusters).
… but there is also the view that AI will make news less transparent (−8) and less trustworthy (−19); and these views have hardened since 2024 (we saw no decreases).
October 7, 2025 at 7:01 AM
… but there is also the view that AI will make news less transparent (−8) and less trustworthy (−19); and these views have hardened since 2024 (we saw no decreases).
Finally, looking at AI in the news, there is a clear ‘comfort gap’ between AI- and human-led news. On average, only 12% say they are comfortable with news made entirely by AI; (21% with a ‘human in the loop’, 43% when a human leads, 62% for entirely human-made news).
October 7, 2025 at 7:01 AM
Finally, looking at AI in the news, there is a clear ‘comfort gap’ between AI- and human-led news. On average, only 12% say they are comfortable with news made entirely by AI; (21% with a ‘human in the loop’, 43% when a human leads, 62% for entirely human-made news).
Only three sectors see the pessimists outnumber the optimists when it comes to AI and its use by these actors – news media, government, and, especially, politicians and political parties…
October 7, 2025 at 7:01 AM
Only three sectors see the pessimists outnumber the optimists when it comes to AI and its use by these actors – news media, government, and, especially, politicians and political parties…
Among those who have encountered AI answers, 50% say they trust them. Respondents emphasised their speed and convenience and the fact that AI aggregates vast amounts of information as reasons to trust them, although trust seems conditional.
October 7, 2025 at 7:01 AM
Among those who have encountered AI answers, 50% say they trust them. Respondents emphasised their speed and convenience and the fact that AI aggregates vast amounts of information as reasons to trust them, although trust seems conditional.
Self-reported click-through behaviour is mixed. Among those who saw AI answers, 33% say they always/often click links in AI overviews, 37% say they do so sometimes, and 28% rarely or never click through. Younger people & those who trust AI search are more likely to say they do
October 7, 2025 at 7:00 AM
Self-reported click-through behaviour is mixed. Among those who saw AI answers, 33% say they always/often click links in AI overviews, 37% say they do so sometimes, and 28% rarely or never click through. Younger people & those who trust AI search are more likely to say they do
Looking beyond news & towards AI in search, people regularly see AI-generated search answers – perhaps no surprise given Google’s wide rollout of AI Overviews and monopoly. Across countries, 54% say they saw an AI-generated answer to one of their searches last week
October 7, 2025 at 7:00 AM
Looking beyond news & towards AI in search, people regularly see AI-generated search answers – perhaps no surprise given Google’s wide rollout of AI Overviews and monopoly. Across countries, 54% say they saw an AI-generated answer to one of their searches last week
…evaluating, or rewriting are most common. Younger users lean more towards using AI to help them navigate the news: 48% of 18–24s used AI to make a story easier to understand compared to 27% of 55+ (a 21 percentage point gap).
October 7, 2025 at 7:00 AM
…evaluating, or rewriting are most common. Younger users lean more towards using AI to help them navigate the news: 48% of 18–24s used AI to make a story easier to understand compared to 27% of 55+ (a 21 percentage point gap).
Getting news specifically via a generative AI systems has doubled but remains a minority activity. Weekly use rose from 3% to 6%, and is stronger among 18–24s (8%) compared to 55+ (5%), and degree-holders. Among AI-for-news users, ‘latest news’ (54%) and help with summarising…
October 7, 2025 at 7:00 AM
Getting news specifically via a generative AI systems has doubled but remains a minority activity. Weekly use rose from 3% to 6%, and is stronger among 18–24s (8%) compared to 55+ (5%), and degree-holders. Among AI-for-news users, ‘latest news’ (54%) and help with summarising…
Information-seeking is now the lead use-case across countries. Weekly use of AI this more than doubled (11% to 24%), overtaking creating media (up 7 pp to 21%). Social interaction with chatbots is at low levels (7%) but higher for younger people (13% of 18–24s/4% of 55+).
October 7, 2025 at 7:00 AM
Information-seeking is now the lead use-case across countries. Weekly use of AI this more than doubled (11% to 24%), overtaking creating media (up 7 pp to 21%). Social interaction with chatbots is at low levels (7%) but higher for younger people (13% of 18–24s/4% of 55+).
And its worth remembering that the majority of people in all countries covered are not regular users of any AI. Even for the most popular ones, many say they only use it monthly or have used it once/twice.
October 7, 2025 at 7:00 AM
And its worth remembering that the majority of people in all countries covered are not regular users of any AI. Even for the most popular ones, many say they only use it monthly or have used it once/twice.
A lot of this is concentrated around ChatGPT or AI systems from big tech companies (hello first mover/platform advantage). Systems such as Claude or Perplexity barely make a dent with the wider public.
October 7, 2025 at 7:00 AM
A lot of this is concentrated around ChatGPT or AI systems from big tech companies (hello first mover/platform advantage). Systems such as Claude or Perplexity barely make a dent with the wider public.
GenAI use is expanding rapidly. Across countries, the proportion of people who say they have ever used any AI system rose from 40% (2024) to 61% (2025); weekly use nearly doubled from 18% to 34%. The early internet took about 3 years to grow in a similar fashion.
October 7, 2025 at 7:00 AM
GenAI use is expanding rapidly. Across countries, the proportion of people who say they have ever used any AI system rose from 40% (2024) to 61% (2025); weekly use nearly doubled from 18% to 34%. The early internet took about 3 years to grow in a similar fashion.