The chatbot trap + News/analysis

The Spectrum of Misinformation returns for 2026 with more news, analysis & practical advice for communicators.

In News digest: nostalgia porn, bad baby advice, AI overview fails, missing scientist myths, fake influencers, London under attack, Grok still X-rated, Aussie disinformation campaigns, antivaxx u-turns. Plus: find out who has created a "crisis of public trust", and how AI could be both the misinformation problem and the solution. Ready for your check-up? The 'chatdoc' will see you now, we ask the panel about your Trust issues before swotting up on Case histories.

The Guardian covers how AI 'boomerslop' memes and conspiracy theories are radicalising older people in the UK, Sara Wilford says 'nostalgia porn' on social media is often the first step towards engaging with extremist content.

BBC investigates self-described experts giving dangerous baby sleep advice, Wes Streeting comments: "dangerous misinformation dressed up as expert advice is putting babies' lives at risk - and it must stop".

New York Times tested Google's AI overviews showing that, despite being 90% accurate, it delivers tens of millions of incorrect overviews every hour. Many answers did not link to evidence fully supporting their results and overviews could be manipulated by fake news stories.

BBC documents the 'missing scientists' conspiracy theory linking the deaths of 10 people in the US with ties to sensitive research despite there being explanations in each case. Loved ones called the baseless speculation 'disgusting' and 'terrible'.

A New York Times investigation found over 300 TikTok accounts sharing AI-generated pro-Trump fake influencers posing as ordinary Americans.

London's Mayor has warned the capital faces a "dark blizzard of disinformation" with false, often AI-generated, content portraying it as a violent 'fallen city' [BBC].

NBC News finds GrokAI is still making sexual deepfakes despite X promising to prevent their creation of such images. Researcher Genevieve Oh said Grok was "unmistakably the largest nonconsensual synthetic nudity generator”.

France 24 reports on a network of Facebook pages run by users in Vietnam spreading disinformation about Australian politics.

The Independent covers the European Council imposing sanctions on two Russian organisations linked to disinformation campaigns.

Straits Times tells the stories of US antivaxxers who have changed their minds about vaccines as measles cases surge, exploring their motivations and how health professionals engaged with their concerns.

The Guardian covers a report arguing misinformation from RFK Jr and top US health officials has created a "crisis of public trust" over autism and vaccines with actions “detrimental to the autistic community”.

City of Sanctuary UK has been cleared of all wrongdoing by the Charities Commission after false claims that it encouraged children to send Valentine's Day cards to adult asylum seekers.

Science hosts a long read on how AI is super-charging misinformation but can also be used to fight it. Will its use lead to an 'AI arms race' between misinformers and those trying to stop them? Or is AI the only way to defend democracy?

The 'chatdoc' will see you now

A message pings in your health app. Your test results are in and, from what you can tell, it doesn't look good.

A pop-up asks: would you like to discuss your diagnosis?

But instead of a face-to-face consultation (or call) with a doctor your first conversation is with an LLM-powered chatbot...

Just a few years ago this would've been science fiction. But with new health technologies, digital patient records, the 'app-ification' of the NHS, and the rush to embed LLMs/GenAI in everything, it seems all but inevitable that, as with utilities and shopping sites, chatbots will become the mediators, interpreters and assistants on our healthcare journey.

What does this mean for misinformation?

That's what I was pondering recently during a meeting to discuss the challenges of bringing new digital technologies into the NHS.

AI tools 'under the hood' of healthcare can deliver impressive results: for example analysing CT scans to spot blockages in blood vessels in the head and neck that emergency medics may have missed.

But as we discussed 'empowering people' by giving them control over their health data, it made me think about the difference between raw data and information and the vital role of health professionals as interpreters and communicators.

With health services increasingly accessed through our smartphones, and human medics in short supply, it's only a matter of time before chatbots will be integrated into apps to take on some of this communicator role.

So why should that worry us?

Firstly there's accuracy. We know LLMs such as ChatGPT and Google Gemini often make mistakes, are overconfident in their answers, and can give bad advice. Who will check that chatbots in health apps are reliable? Who is accountable if they give false or misleading answers?

There's also the complexity of interactions. A recent Oxford study (testing GPT-4o, Llama 3 and Command R+) found that, despite in-depth medical knowledge, the LLMs suffered a 'two-way communication breakdown' when interacting with real human patients, with people not knowing what information to provide to get accurate advice and LLMs combining good and poor recommendations.

And then there's the problem of perception...

When we consult with a doctor we're not just looking for communication, we're looking for connection. We want to discuss our problems with someone who understands us, who we trust to filter information and tell us what we need to hear or what they'd do in our position. LLMs may pretend to do this but however realistic their replies they cannot think or empathise, tend to be 'flattery engines' that tell us what we want to hear.

Embedding a chatbot within an NHS app will make it appear more trustworthy and reliable (because we trust the NHS/doctors) even if it's just as prone to hallucination and misinformation as other chatbots.

Without better education about what LLMs are and aren't, and tackling misinformation about their capabilities, we risk the most vulnerable in society - the lonely, the elderly, the desperate - developing dangerous parasocial relationships with their chatbot 'nurse' or 'doctor' and trusting them over real humans who have their best interests at heart.

Chatbots have their uses in healthcare. But I worry that, without adequate human supervision, their inability to provide accurate, consistent, reliable answers risks undermining both information integrity and public trust.

Most agree we need to keep the human element in health. But with health budgets and systems under such pressure, how much are we willing to pay to 'keep it human'? Or, unable to get the information we need elsewhere, will consulting an unreliable 'chatdoc' be our only option.

Trust issues

How AI is upending the information ecosystem was one of the topics covered at a recent panel discussion organised by Media Trust.

On 22 April I joined the panel alongside Steve Nowottny, Editor at FullFact & Sara Allison, Head of Content at Macmillan Cancer Support.

You can watch a recording of the full session online, but I thought I'd pull out some key takeaways and my reflections on a fascinating chat...

We agreed that AI is disrupting not just information but how we search for it, that it leads people to less reliable sources and at the same time can prevent carefully-curated accurate info from being seen.

The rise of fake experts was another issue: with deepfakes of real doctors and entirely made-up experts (including bogus LSHTM experts!) used to endorse products or lend credibility to AI slop news sites.

Not knowing who to trust was a recurrent theme, Sara highlighted a Patient Information Forum survey (2024) which found half of adults in the UK struggle to access trusted health information and that 1 in 10 adults in the UK had been affected by misinformation. She also talked about the influence of influencers, people choosing what to believe based on 'gut instinct', and that 'trust comes with proximity' with many people turning to friends or family for advice.

On countering misinformation, Steve made several points that had me nodding vigorously, including that you should be careful not to amplify false claims when responding and that you shouldn't be dismissive or make people feel stupid for believing them: think instead about the majority of people who are genuinely uncertain or don't know.

Meanwhile I talked about LSHTM's Health Information Integrity Network, and my take on information integrity, the need to identify and fill information gaps and the approach of curate, create and defend. I also touched on existential threats to the integrity of the information ecosystem, such as political attacks on bodies that safeguard information and fake paper mills undermining the research on which accurate high-quality information depends.

I gave my tips for counter-misinformation on a budget: how even if you can't do a big campaign you can still do things like investigating people's genuine questions and concerns (using free tools like AlsoAsked) and examples of common misinformation on a topic (by prompting ChatGPT/CoPilot etc), and using this intel to create responses and explainers. This sort of 'audience first' approach can be embedded in everyday communications work.

Reflecting on the session overall, the challenges of overcoming distrust of authority and of responding rapidly enough while ensuring information is accurate were ones we were all grappling with. Some of the answers were the same too, making sure content was respectful and relatable, working closely with experts, and engaging with local organisations that are closer to the vulnerable audiences we are trying to reach.

Thanks to Media Trust for the invite, sign up to their newsletter to find out about future events.

Case histories

Last time I gave tips for navigating the choppy waters of information integrity and what it means for counter-misinformation work, as well as treating readers to an update on UK misinformation policies.