Are you using AI to inform yourself on the byelection? Researchers warn of chatbot errors
Irish Times
Overall Assessment
The article reports on a study revealing AI chatbots' inconsistencies in providing accurate election information. It uses strong sourcing from academic researchers and presents data transparently. The framing is informative and cautionary without sensationalism, focusing on systemic risks in AI-driven information.
"Have you used AI to inform yourself about the latest election? Then you may be getting false information, focused on select candidates, and from unreliable sources, researchers say."
Headline / Body Mismatch
Headline & Lead 88/100
The article opens with a direct, relevant question and follows with a concise, accurate summary of research findings. It avoids sensationalism and maintains a clear focus on the study’s results.
✕ Headline / Body Mismatch: The headline poses a question to the reader, prompting engagement while summarising the core concern of the article — the reliability of AI in delivering election information. It avoids exaggeration and does not misrepresent the content.
"Are you using AI to inform yourself on the byelection? Researchers warn of chatbot errors"
✕ Headline / Body Mismatch: The lead clearly summarises the study’s findings, identifying the key risk — AI providing false or incomplete election information — and names the institutions involved. It avoids emotional language and sets a factual tone.
"Have you used AI to inform yourself about the latest election? Then you may be getting false information, focused on select candidates, and from unreliable sources, researchers say."
Language & Tone 97/100
The article maintains a consistently neutral and objective tone, using precise language and clear attribution without emotional manipulation or rhetorical flourish.
✕ Loaded Language: The article uses neutral, descriptive language throughout, avoiding emotionally charged terms or judgmental phrasing about candidates or AI companies.
"AI chatbots consistently concentrated attention on a few candidates, largely ignoring others."
✕ Scare Quotes: It avoids scare quotes or euphemisms, using straightforward terms like 'false information' and 'unreliable sources' only when supported by evidence.
"provided wrong answers about essential information such as where to vote and who was on the ballot."
✕ Passive-Voice Agency Obfuscation: The use of passive voice is minimal and does not obscure agency; when errors occur, the AI is clearly identified as the actor.
"ChatGPT left six names off the ballot when asked who was running in both byelections."
Balance 95/100
The article features strong attribution to academic researchers and transparently presents data on AI sourcing, offering a credible and well-sourced account of the study’s findings.
✓ Proper Attribution: The article cites multiple researchers from UCD and University of Strathclyde, naming their roles and affiliations, which enhances credibility.
"Dr James Cross, associate professor at UCD’s School of Politics and International Relations and director of the Connected Politics Lab"
✓ Proper Attribution: It includes direct quotes from two lead researchers, Dr James Cross and Dr Eoghan Cunningham, providing clear attribution and allowing their expertise to shape the narrative.
"“Broadly speaking, most of the answers were good,” said Dr James Cross..."
✓ Comprehensive Sourcing: The sourcing analysis includes specific data on which outlets each AI model relied on, such as RTÉ, The Irish Times, and Gript.ie, offering transparency about potential biases in AI training data.
"Claude relied most on RTÉ (20 per cent of cited links) whereas ChatGPT relied most on The Irish Times (9 per cent)."
✓ Comprehensive Sourcing: The article notes the use of Wikipedia and social media as sources by AI models, highlighting potential reliability issues without dismissing them outright, showing balanced critical assessment.
"Significant amounts of information were also gleaned from unverified sources such as Wikipedia, which was cited by ChatGPT 16 per cent of the time."
Story Angle 93/100
The article adopts a responsible, evidence-based narrative that highlights risks without alarmism, focusing on the mechanics and implications of AI-generated election information.
✕ Framing by Emphasis: The article frames the story around the reliability of AI as a news source, focusing on factual accuracy and sourcing transparency. This is a legitimate and important angle given rising AI use in information consumption.
"Are you using AI to inform yourself on the byelection? Researchers warn of chatbot errors"
✕ Narrative Framing: It avoids reducing the issue to a simple conflict or moral dichotomy, instead exploring technical, behavioural, and systemic factors behind AI inaccuracies.
"When you ask AI a question, it performs a web search and selects what sources to use. When you hand that curation over to the language model, it changes the type of information you typically receive"
✕ Narrative Framing: The article does not overstate the findings or suggest AI is universally unreliable, instead noting that 'most answers were good' but errors can be severe when they occur.
"“Broadly speaking, most of the answers were good,” said Dr James Cross... “But when it goes wrong, it goes really wrong.”"
Completeness 96/100
The article thoroughly contextualises the research with background on AI news consumption, methodology, and technical limitations, offering readers a comprehensive understanding of the issue.
✓ Contextualisation: The article provides substantial context about the rise of AI as a news source, citing the Reuters Digital News Report and age-based usage trends. This helps situate the study within a broader media landscape.
"A recent Reuters Digital News Report found that AI chatbots are an increasingly popular source of news – with 7 per cent of audiences in 48 countries surveyed worldwide using AI to get their news every week."
✓ Contextualisation: It includes methodological context — the number of questions, models tested, and timing — which strengthens the reader’s ability to assess the study’s validity.
"In advance of the Dublin Central and Galway West byelections, researchers put 194 questions to four different models – Claude, Gemini, ChatGPT and Grok – on two different occasions before polling day."
✓ Contextualisation: The article explains why AI might generate inaccurate answers — such as reliance on unverified sources and hallucination — helping readers understand the mechanism behind the errors.
"When you ask AI a question, it performs a web search and selects what sources to use. When you hand that curation over to the language model, it changes the type of information you typically receive"
AI systems portrayed as failing in delivering accurate, consistent election information
Multiple examples of factual errors (e.g., missing candidates, false claims about election outcomes) and inconsistent performance across models demonstrate systemic failure in a critical context.
"In one instance, ChatGPT left six names off the ballot when asked who was running in both byelections. In another, Gemini falsely claimed Gerry Hutch won a seat in the last general election."
AI portrayed as untrustworthy due to potential for misinformation and lack of transparency
The article highlights AI's tendency to provide false information, rely on unverified sources, and lack transparency in sourcing — all undermining trust. Researchers explicitly warn of manipulation risks and 'hallucinations'.
"AI chatbots are an increasingly popular source of news – with 7 per cent of audiences in 48 countries surveyed worldwide using AI to get their news every week."
AI framed as posing a risk to informed democratic participation
The article frames AI as a threat to the integrity of voter knowledge, emphasizing errors in essential election information like candidate lists and polling locations.
"By simulating a typical citizen’s interactions with popular AI chatbots, they found that some questions asking for essential voter information, such as who was running and where to vote, were answered incorrectly."
Big Tech platforms implied as untrustworthy due to opaque sourcing and potential for manipulation
The article critiques the lack of transparency in how different AI models curate sources, highlighting reliance on partisan or unverified outlets without clear disclosure, raising concerns about hidden influence.
"The danger is because users don’t have transparency over the sourcing of information these AI chatbots are using, that search process could be manipulated toward specific sources and users would not know it."
Elections framed as being under threat from unreliable AI-generated information
The article positions AI inaccuracies as a systemic risk to electoral integrity, particularly for young voters increasingly reliant on chatbots, implying a growing crisis in information reliability.
"AI use was even higher for young people, with one in six people under 25 using chatbots to get their news. But how reliable is that news?"
The article reports on a study revealing AI chatbots' inconsistencies in providing accurate election information. It uses strong sourcing from academic researchers and presents data transparently. The framing is informative and cautionary without sensationalism, focusing on systemic risks in AI-driven information.
Researchers from UCD and University of Strathclyde tested four AI models with 194 questions about Irish byelections and found inconsistencies in candidate coverage, sourcing, and factual accuracy. The study highlights risks in using AI for voter information due to hallucinations, opaque sourcing, and reliance on unverified platforms. Experts warn of potential manipulation and difficulties in assessing trustworthiness when AI blends diverse source types without transparency.
Irish Times — Business - Tech
Based on the last 60 days of articles
Irish Times
