Who does ChatGPT trust? Examining the news platforms AI cites most frequently

The debate over bias in tech platforms isn’t new. For years, companies like Google and Meta have faced accusations of favouring certain political ideologies. Now, as AI platforms like ChatGPT become integral to online interactions, questions arise about their impact on information flow and media visibility.

This edition of Franc Talking dives into ChatGPT's referral data from the past six months, examining which websites users are most frequently directed to—and what it might reveal about perceived biases, if any, in AI.

After Donald Trump’s first presidential victory, Google faced significant criticism in 2018 when a leaked video surfaced showing co-founder Sergey Brin visibly dismayed by the election outcome. Several of his comments in the video provoked outrage among conservative commentators, fuelling accusations of political bias within the company.

A 2021 study by NYU's Stern Center for Business and Human Rights found no reliable evidence that social media platforms systematically remove or suppress conservative content due to ideological bias. Instead, content removal typically results from violations of platform policies, such as spreading misinformation or hate speech, which apply to all users regardless of political affiliation.

In an era where disagreements are often dismissed as "fake news," avoiding polarisation becomes increasingly difficult. This dynamic fosters a divisive environment where individuals are pushed to align with one side or the other, creating an "us versus them" mentality.

Filter bubbles and echo chambers are real phenomena that can surround individuals with information reinforcing their existing beliefs, potentially increasing polarisation. This effect is driven by both human behaviour and the design of social media algorithms and tech platforms.

How long will it be before AI faces similar accusations of influencing users and favouring certain political ideologies? As adoption grows and examples emerge of platforms appearing to 'favour' specific viewpoints, it will be crucial for their creators to explain the reasoning behind these responses and demonstrate their neutrality and balanced approach to presenting information.

Which websites does ChatGPT link to?

Globally

Data from the past six months reveal the most frequently referred websites from ChatGPT. While many of these sources are categorised as left-leaning, there are exceptions, such as sociedademilitar.com.br, which is identified as a conservative right publication.

USA 

The majority of ChatGPT referrals in the United States are directed toward centre-left publications, with The New York Times being a prominent example. Interestingly, despite filing a lawsuit against OpenAI over alleged content crawling without a commercial agreement, The New York Times remains one of the most frequently referred sources."

UK

The referral patterns in the UK mirror those seen in the US, with a significant share directed toward left-leaning publications. Notably, theguardian.com and independent.co.uk account for nearly half of all referrals, showcasing their prominence in the media landscape.

Other notable sources include bbc.com, which, despite its attempts to remain impartial, is often perceived as leaning slightly left, and standard.co.uk, a centrist outlet with a strong regional focus.

Brazil

Traffic patterns in Brazil reveal a notable contrast to those observed in the US and UK, with a stronger emphasis on centre-right publications. Sociedademilitar.com.br, a conservative-leaning platform, leads the referrals with 21.7%, highlighting its prominence in the political discourse. Other significant sources include brasil.perfil.com (17.5%) and diariodobrasilnoticias.com.br (13.9%), which further reinforce the centre-right leaning trend.

Additionally, niche publications like comprerural.com (6.9%) and terrabrasilnoticias.com (10.5%) reflect the diversity of topics resonating with ChatGPT users in Brazil, ranging from rural and agricultural news to broader political coverage. Interestingly, marieclaire.perfil.com also appears, showcasing the mix of lifestyle and political content referred by ChatGPT.

Germany

The German segment shows a more centrist tendency, with significant traffic directed to mainstream platforms like t-online.de (24.3%) and dw.com (12.2%), which are known for their balanced reporting. However, there are notable conservative influences from publications such as bild.de (8.1%), a historic outlet often associated with right-leaning perspectives.

Other key sources include spiegel.de (8.1%), a left-leaning publication recognized for its investigative journalism, and haberturk.com (8.1%), reflecting some international interest. Additionally, regional sites like bz-berlin.de (6.8%) and chiemgau24.de (17.6%) indicate a strong preference for local and regional content among users in Germany.

France

As expected, the French media landscape shows a strong preference for centre-left publications. Leading the pack is lemonde.fr, which accounts for 21.4% of referrals, reflecting its reputation as a trusted and influential outlet in French journalism. Following closely is bibamagazine.fr (19.6%), highlighting a mix of lifestyle and news content that appeals to a broader audience.

Regional and national outlets like france3-regions.francetvinfo.fr (12.5%) demonstrate the importance of localized reporting, while global sources such as bbc.com (8.9%) and theguardian.com (8.9%) also feature prominently, indicating an interest in international perspectives.

Smaller yet significant contributors, including publicsenat.fr (5.4%) and linternaute.com (5.4%), underscore the diversity of sources shaping French audiences' views. The inclusion of nytimes.com (5.4%) and express.co.uk (5.4%) highlights some engagement with foreign, English-language publications.

What does this mean?

Does ChatGPT favour left-leaning publications?

No, ChatGPT does not favour any particular political leaning. Any apparent tendency for referrals to left-leaning publications is more likely attributed to user behaviour than an intention to direct traffic towards such sources.

ChatGPT’s primary demographic—predominantly younger (18–34 years old), 55% male, and university-educated—may naturally gravitate towards certain types of content.

Additionally, widely recognised and highly indexed publications, which may include left-leaning outlets, tend to appear more frequently in responses due to their online prominence.

From a personal perspective, whenever I’ve questioned issues like Brexit, ChatGPT has consistently provided balanced answers, weighing the pros and cons of each side, despite being aware of my political leanings and beliefs

Why do publications, such as the Nytimes.com, feature so highly?

Training data plays a pivotal role in shaping ChatGPT's outputs. Publications with strong international reputations, extensive coverage across diverse topics, and a foundation of trust built over decades—if not centuries—are naturally referenced more frequently. Their widespread recognition and influence make them cornerstones of information within AI models.

The New York Times is a particularly strong example. Despite the absence of a commercial agreement with OpenAI and ongoing legal action alleging unauthorised use of its articles for AI training, the publication remains highly prominent. It has also blocked OpenAI’s crawlers, restricting future access to its content. Yet, The New York Times continues to be a top destination from ChatGPT outputs.

This prominence is driven by several factors:

• Historical Inclusion: Content from The New York Times was likely part of earlier datasets before access restrictions were implemented.

• Indirect References: The publication is frequently cited or summarised by other accessible sources, amplifying its visibility.

• Reputation: As a globally respected source of high-quality journalism, The New York Times maintains its influence in discussions and responses generated by ChatGPT.

These dynamics underline how legacy publications, even without active partnerships, can retain significant visibility within AI-driven platforms due to their historical, indirect, and reputational strength.

Methodology used:

Using data from Allsides.com, any reference to “left” or “right” was backed up by the labelling in the platform. An example for The New York Times is below.

Data was extracted using the excellent SimilarWeb analysis tool, which I thorough recommend if you work in marketing.