Not a stakeholder, not just a media source, but an influencer.

Like many, I’m spending copious hours trying to make sense of AI, its risks and benefits, and how it will continue to shape our function and profession.

A debate that’s caught my attention is whether to treat AI as a stakeholder.  A recent Page Turner blog post explains why that’s a false proposition. Here I build on the author’s concept of AI as a media source and suggest that it also has the power to persuade and shares other characteristics of an influencer. These five points expand on my thinking about LLMs:

1. Its gatekeeping and prioritization functions ascribe it influence. LLMs don’t simply round up links to information; they serve curated overviews and digests based on what they find, choose and synthesize. This includes information that’s been sourced by their own agents to fill in knowledge gaps, as well as ideas that are altogether left out, such as newsroom content that’s blocked due to pending lawsuits. LLMs may sift through a lot of substance to generate a probabilistic average of the information, but it often serves answers back as their own, which can create a perception of authority or influence. The process alters how information is created and determines what content people see and whose voices are excluded – another form of influence.  

2. People are using LLMs for discovery research and may be treating the answers as independently verified, trusted opinions.  Our institutional websites used to be our digital front doors to the world. Audiences are now turning to AI for recommendations that contrast us with our competitors and thereby feel credible. Of the 900 million people who use ChatGPT weekly, over 50% use it to seek information. The outputs they receive can influence which organizations emerge in a person’s short-list of viable options. As evidenced by the rise in branded search (where people turn to search engines to look up our organizations by name), studies show that website visits happen further down the conversion funnel, once minds are partially set, to validate decisions or gather the remaining information needed for a final selection. 

3. LLMs, like influencers, reward recency, and face revenue generation pressures. Muck Rack’s “What is AI Reading” study showed that in LLM results, 57% of journalistic citations (where publish dates were known) were published within the last 12 months. Influencers, similarly, rely on current trends and the zeitgeist to stay relevant.  Recent reports suggest that Open AI won’t be profitable until 2030 and faces a $200B shortfall, which it could raise through continued growth or partnerships, tiered subscriptions, payment plans based on token usage, and the creation of custom models.  Some suggest that LLMs could incorporate paid advertising, whereby answers include sponsored content.  Such a model would not be unlike paid social media influencers who endorse brands in exchange for a fee to reach new audiences and impart their endorsement to boost credibility. 

4. LLMs, like paid influencers, promote content that’s biased. The answers that LLMs return can exhibit bias, especially if their training data was flawed or incomplete.  Much has been written about models being trained disproportionately on Western culture and the English language, which may be favoured in results. LLMs often generalize, ignore nuance, may optimize output that agrees with our line of questioning, and may favour information placed at the start or end of our prompts. Being paid to endorse a product or service on social media is also a form of bias because it reflects a disproportionate preference, or showing of favour for something.

5. AI agents and chatbots, just like paid influencers, are sought out for advice and are considered official representatives of our brands.  Studies show that audiences consider paid influencers and celebrity endorsers as official brand spokespersons. In good times, this relationship leads to a transfer of trust and may deepen brand appeal. When a paid celebrity endorser makes an ethical misstep that’s out of alignment with a brand’s values, the reputational damage can be immediate and severe, especially if it’s met with inaction by a firm. Similarly, AI-enabled customer service agents and chatbots are created to provide advice and are considered an official source of information. Users are bound to ask them unanticipated questions that probe on perceived shortcomings of institutional judgement. Great care needs to be taken to build in the parameters to produce nuanced and contextualized but truthful responses.  

Sources:

Dunleavy, K. (2026, May 7). How AI citations have changed in the last 6 months. Muck Rack. https://muckrack.com/blog/what-is-ai-reading-new-insights-may

Freberg, K., Graham, K., McGaughey, K., & Freberg, L. A. (2011). Who are the social media influencers? A study of public perceptions of personality. Public Relations Review, 37(1), 90–92. https://doi.org/10.1016/j.pubrev.2010.11.001

Güvençer, E. (2026, May 5). Your AI agents are your new spokespeople. Are you briefing them? PRWeek UK. https://www.prweek.co.uk/article/1956962/ai-agents-new-spokespeople-briefing-them

Hall, B. (2025, November 10). Cognitive bias patterns in LLMs. USC AI Beat. University of Southern California Libraries. https://libguides.usc.edu/blogs/USC-AI-Beat/bias-patterns-llms

Lichtenberg, N. (2025, November 26). OpenAI won’t make money by 2030 and still needs to come up with another $207 billion to power its growth plans, HSBC estimates. Fortune. https://fortune.com/2025/11/26/is-openai-profitable-forecast-data-center-200-billion-shortfall-hsbc/

Malik, A. (2026, February 27). ChatGPT reaches 900M weekly active users. TechCrunch. https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users/

OpenAI. (2025, September 15). How people are using ChatGPT. https://openai.com/index/how-people-are-using-chatgpt/

Peters, U., & Chin-Yee, B. (2025). Generalization bias in large language model summarization of scientific research. Royal Society Open Science, 12(4), Article 241776. https://doi.org/10.1098/rsos.241776

Silliman, E., Boudet, J., Robinson, K., Oppong, D., & Shah, N. (2025, October 16). New front door to the internet: Winning in the age of AI search. McKinsey & Company. https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/new-front-door-to-the-internet-winning-in-the-age-of-ai-search

Zewe, A. (2025, June 17). Unpacking the bias of large language models. MIT News. https://news.mit.edu/2025/unpacking-large-language-model-bias-0617