- Human content powers AI search: Most AI-generated search results originate from human-written, expert-led content.
- Authenticity beats AI-generated content: Original insights build trust with journalists, audiences and AI platforms while generic AI content risks being ignored.
- EEAT drives visibility: Experience, Expertise, Authoritativeness and Trustworthiness remain critical ranking factors in GEO.
- Thought leadership earns citations: Authoritative content is more likely to be referenced by AI search engines and media outlets.
- Use AI to support, not replace: AI excels at research and analysis, but human oversight and content creation are essential for credibility, accuracy and GEO performance.
Large language model (LLM) ChatGPT boasts that it has approximately 900-million weekly active users worldwide. If just one LLM, albeit the most popular, is boasting almost a billion users a week, it is hardly surprising that more people are starting to roll their eyes at the flood of AI-generated content everywhere.
Africa’s technology communications partner DUO Marketing + Communications has drawn a line in the sand, telling clients: “Do not expect us to feed AI-generated slop to the media”.
CEO Judith Middleton says that if a PR partner positions itself as a guardian of brand trust and authenticity, and legal safety, it is obliged to deliver human-led strategy and human-generated content to a media landscape and social media environment saturated with AI content.
“Journalists trust that we are offering them genuine thought leadership, not aggregated beige content at best, or plagiarised ideas at worst,” she says. Beyond this, reports show that 86% of Google AI results come from human-written content. “Knowing this, why would you bother with AI sludge?” she asks.
Media backlash
The hard line represents a shift from the standard question of “the role of AI in PR”, which tends to look at automation and potential job losses, to “the role of PR in an AI-first information ecosystem”, a more nuanced approach designed to protect against brand dilution and reputational damage.
“It is tempting for business leaders to view generative AI as a quick win for efficiency, but the truth is that relying on it for public commentary can lead to an erosion of trust and potential media backlash,” says Middleton.
Tamsin Mackay, a prominent IT journalist, an international corporate trainer and PhD candidate researching the intersection of writing and AI says the outputs generated by LLMs are simply not writing.
“We all have a sense of what writing is. An LLM is not doing that, it’s predicting which word should follow on from the previous word. This is why a conclusion written by AI is generally fairly limp and soggy, like lettuce that’s been left in a cheese and tomato sandwich for too long.”
Severe consquences
However, beyond aesthetic flaws, that often only writers can identify, Mackay explains that the architectural reality of LLMs poses factual risks, and this is where the more severe consequences for brands might lurk.
“Recent academic and industry data points to the fact that AI is fundamentally an automated ‘pleaser’ rather than a factual researcher. In an accuracy benchmark by Stanford, the study found that hallucination rates across 26 top models range from 22% to 94%.
“However, in certain scenarios, GPT-4o’s accuracy dropped from 98.2% to 64.4%, and DeepSeek R1 fell from over 90% to 14.4%. When a false statement is presented as something another person believes, the models handle it well. When the same false statement is presented as something a user believes, their performance collapses ”
Speaking about a personal anecdote, which draws parallels with the recently spotted hallucinations in the Department of Communications and Digital Technologies’ draft National Artificial Intelligence Policy, Mackay says that even with the best intentions, humans don’t always “spot the slop”.
Satisfying the creator
“The software is mathematically optimised to satisfy the prompt creator. This means it will happily fabricate realistic looking data to align with a user’s bias. No C-suite, to my knowledge, would be happy with an overly eager Labrador giving it what it wants to read over well-researched, accurate information. And it’s not easy to identify.
“In a recent classroom experiment, I presented an AI-generated research text that contained entirely fabricated data. Out of the entire class, only one student spotted the hallucinations. The point is: if AI is going to be used for a task, it needs deliberate human supervision.”
What this means for brands
Middleton explains that the consequences of relying on, and publishing, AI content extend well beyond poor grammar or hallucinations.
“Search dynamics are undergoing a radical shift from traditional Search Engine Optimisation (SEO) to Generative Engine Optimisation, where platforms like Google use AI to answer user queries directly.
“The greatest irony is that the LLMs rely on good-quality content published on high authority websites. Think of traditional PR with a new superpower. Now, while search engines and AI platforms do not ban AI-generated content completely, they heavily penalise and downweight generic, unhelpful, or spam-heavy material.
“The technology itself proves that AI visibility depends on human expertise. Recently, Medium quoted a US-based SEO and GEO specialist agency which found that 86% of Google AI results come from human-written content. That figure is at 82% for ChatGPT and Perplexity.”
Understanding the EEAT framework
It is important for CMOs and other C-suite executives to understand that under Google’s core EEAT framework (Experience, Expertise, Authoritativeness and Trustworthiness), backward-looking AI synthesis is penalised.
In other words, explains Middleton, in the rush for speed and efficiency, AI generated content that a brand puts out is simply a rehash of what’s already out there, reducing any chance it will be cited in AI-driven searches as well as being published in mainstream media.
“Beyond this, and this is probably the most important point through the lens of public relations, there is an ever-present risk of unintentional plagiarism and copyright infringement as LLMs can read data behind paywalls. One only has to follow the high-profile legal battles involving major international publishers to understand how serious, and real, the risk to brands is.”
How AI is successfully used in PR
“We need to be deliberate,” says Middleton. “When we understand the structural flaws of a probability engine, we appreciate our role in the PR industry: to ensure we keep the brands we represent firmly on the side of trust. It requires a careful and considered approach to AI, underpinned by ethics.
“Certainly from our perspective, the DUO AI Code of Honour is an agency north star that enables us to augment workflows for the much-desired efficiency, while keeping authentic, human-led content front and centre of our outputs.”
Two important points
Middleton says that brands would do well to work with partners who understand where AI should, and should not, be deployed in communications. This is done by understanding two important points:
Where AI excels: Deep research synthesis, media coverage trend and sentiment analysis, workflow structuring, identifying key platforms that feed search engine datasets, and more.
Where AI harms: Content generation in the form of executive opinion pieces, press releases, signature industry columns, brand voice, tone and positioning, and more.
Middleton invites all CMOs and CEOs to look at their current communications and ask a simple question: Can I genuinely tell which of my brand’s thought leadership articles were written from real human insight, and which were quietly outsourced to a chatbot?
If the answer is yes, and it is AI-generated, ask whether you can stomach the risk. If the answer is no, it’s time to engage a communications professional.
