1/ For 10 weeks, our fact checkers and data scientists asked five leading AI chatbots the same 22 questions every day: Gemini 3.1, ChatGPT, Grok, DeepSeek and Claude.
None repeated the core misinformation they were tested on. But we found several points of failure 🧵
If you come across a picture or video you think could be fake, would you ask an AI chatbot about it? We’ve been looking at how various LLMs cope with queries like this, and often found the answer is not very well. fullfact.org/technology/full…
Who says size matters?
World Cup history is surprisingly short, and this tournament is unusually long.
So when you hear a World Cup stat that seems particularly striking or like some kind of record, think about the sample size.
fullfact.org/sport/world-cup…
Altitude vs Away Form...
We've crunched the numbers on @England vs Mexico in the @FIFAWorldCup
fullfact.org/sport/mexico-en…
What progress has Sir Keir Starmer made in delivering the Labour manifesto’s ‘first steps for change’?
In the week he announced his resignation, here’s our Government Tracker’s verdict.
fullfact.org/politics/keir-s…
I’m so pleased to see this piece and kudos to @stevenowottny for pulling it together. Please get in touch if you’re interested in licensing @FullFact’s AI tools: fullfact.ai/contact/
Full Fact's suite of AI tools on a typical day processes about a third of a million sentences — and has been used by over 40 fact-checking organizations across 30 countries. niemanlab.org/2026/06/full-f…
We first highlighted this misleading claim made by @KemiBadenoch in the @Conservatives ‘Alternative King’s Speech’ in mid May: fullfact.org/economy/income-…
It’s pleasing to know that the @UKStatsAuth have endorsed our analysis. We now need to hear from the Conservative Party.
Today, we have written to the Leader of the Opposition about welfare spending. uksa.statisticsauthority.gov…
We hear (via The Guardian’s live blog) that the Conservative Party intend to “correct a line in a document.”Hopefully, the claim will not be repeated. theguardian.com/politics/liv…
Mark Frankel retweeted
Across digital media, misinformation travels faster than the truth, and AI chatbots are its newest channel.
PJMF partner @FullFact is building the world's first AI Trust Benchmark for LLMs to measure how far their answers can be trusted. Learn more: tinyurl.com/yj2mc4j5
We're looking for an Associate Product Manager to help engage and grow Full Fact's audience.
You'll work across our website and email channels, using data-driven insights to improve acquisition, retention, and user journeys.
Apply here: fullfact.teamtailor.com/jobs…
Mark Frankel retweeted
🚨 Our Digital News Report 2026 is out!
Authored by Jim Egan, Craig T. Robertson, Amy Ross Arguedas, Nic Newman, Rasmus Nielsen, Mitali Mukherjee and Richard Fletcher, it is based on 100K interviews in 48 markets
Full report
reutersinstitute.politics.ox…
7 findings in thread
1/ Facts—and the institutions that verify and communicate them—are under intense pressure in the UK.
Our latest report finds the information environment is structurally weakening, putting democracy at risk.
fullfact.org/policy/reports/…
Mark Frankel retweeted
Fact-checking charity @FullFact secures new £400,000 grant from Patrick J McGovern Foundation to cover the first year of a new project to build and deploy an AI Trust Benchmark to measure the accuracy and reliability of LLMs like ChatGPT
Mark Frankel retweeted
As Black Rod knocked on the door, someone in the Commons shouted 'not now, Andy'
Can you trust those bar charts in the local election leaflets on your doormat?
We’ve found over a dozen examples of leaflets making claims about how people are likely to vote which are misleading, or not backed with reliable evidence.
Our investigation 👇 fullfact.org/politics/electi…
If politicians are serious about restoring trust in our democracy, we deserve more than dodgy election leaflets with unreliable and misleading data. @FullFact’s investigation for @guardian
#FactsMatter
theguardian.com/politics/202…
We’re using our own AI tools to identify how images with Google synth ID markers are being shared during this 2026 election campaign. A good example.
A real image posted by Reform UK MP @SarahForRuncorn on the campaign trail has been edited by social media users to change the supporters behind her.
The edited image has a SynthID watermark, indicating it was manipulated with Google AI tools.
fullfact.org/politics/edited…
Mark Frankel retweeted
7️⃣ Young people use chatbots more often and are more comfortable with AI-generated news
Around 15% are using AI to access news weekly compared with just 3% of those 55 and over. Up to 30% are comfortable with news being made mostly by AI
















