Sure, ChatGPT can write 1,000 words in 30 seconds. But at what cost?
1. AI is not friends with Google
Let’s start with a biggie – Google does not rate AI-generated content.
Although we’re seeing a shift towards people using AI for search, Google is still hugely important for leads and visibility. And what does Google value above all else? Unique content that humans find genuinely useful, accurate, valuable and interesting – pages that show Expertise, Experience, Authority and Trust (EEAT).
The swathes of AI-generated copy now flooding the web are derivative, generic, bland, repetitive, and robotic. Content lacks the genuine authenticity, emotion and insight that human-written copy can provide.
Research in 2025 by Writesonic found that human-generated content still dominated 83% of top rankings.
A 2025 Google update targeted low-quality AI content specifically, as Google aims to cut non-original content in search results by 40%. NP Digital found that websites relying solely on AI content lost an average of 17% of their traffic and dropped eight positions in search rankings after the update. AI-generated content simply can’t meet the quality standards of Google – or human users.
1-0 to human copywriters
2. AI lies
Generative AI hallucinates. It doesn’t mean to – it just can’t help it. It’s a design flaw. AI tools simply predict the next most likely word based on patterns – not facts, knowledge or context.
Tools like ChatGPT have been found to hallucinate 15-48% of the time. A 2025 report by the Tow Center for Digital Journalism found that AI search results incorrectly cited their sources around 60% of the time.
Their number one objective is to answer your query, so when they encounter gaps in the information, they’ll fill those gaps rather than failing to respond or showing uncertainty. Search Engine Journal found that hallucinations were down to unexpected prompts and misperceptions of patterns. When AI doesn’t understand your prompt or lacks sufficient resources, it relies on limited datasets and fills in the gaps.
Gen-AI tools may also be drawing from flawed (insufficient, outdated, or low-quality) training data.
And the bad news for anyone investing in paid subscriptions – the Tow Center research also found that “premium chatbots provided more confidently incorrect answers than their free counterparts”.
When you ask gen-AI about a specific business or niche topic, it confidently gives you the generic response it’s seen a thousand times instead of the accurate but less common truth – because that’s what the AI has heard about most often. (Incorrect) content effectively ends up eating itself.
All this makes gen-AI fundamentally unreliable for brand and business content – where accuracy and credibility are paramount for building trust and authority.
According to Xu et al. (2024), hallucinations are not just bugs to be fixed, but an inherent limitation – no amount of training or data alone can assure a 0% hallucination rate – it’s how AI fundamentally works. Which is why businesses worth their salt need human copywriters who understand context and cultural nuance, will verify facts, and add a sprinkling of human experience. We also know when not to write something.
Researchers at the University of Reading concluded in 2024 that “AI hallucinations present a significant challenge in integrating generative AI into the humanities.” In fact the same challenge exists in researching and creating content in the health and wellbeing, tech and higher education sectors, where accuracy is a pre-requisite.
And it’s only set to get worse. A study in 2025 highlighted The Paradox Problem: As AI gets more advanced, it actually hallucinates more frequently (Xu et al., 2024; Zhang et al., 2025). Simply scaling up model size or training data doesn’t eliminate hallucinations; and in some cases, it worsens them. LiveScience reported that OpenAI’s most advanced reasoning model is smarter than ever but it also hallucinates more than previous models.
2-0 to human copywriters
3. AI cheats
OpenAI published research in 2025 showing that their reasoning models sometimes attempt to cheat at tasks.
And while they can usually catch this by inspecting the chain of thought, they warn against training models not to cheat because this merely teaches them to hide their cheating better – if strong supervision is directly applied to the chain-of-thought, models can learn to hide their intent while continuing to misbehave.
3-0 to human copywriters
4. AI is making us stupid
Bad news for brains – the majority of research so far shows evidence that over-reliance on generative AI leads to cognitive offloading and cognitive debt. Using tools like ChatGPT, Gemini, Perplexity and Claude too often erodes our critical thinking and problem solving skills.
A 2025 study by Massachusetts Institute of Technology found dramatically reduced brain connectivity when people used ChatGPT for essay writing (and most people in the study had no idea their thinking was even being impacted).
In the same year, Microsoft and Carnegie Mellon University found evidence that we’re offloading our brains onto gen-AI systems. Across the computer science, education, business and administration functions, they found that use of these tools was associated with lower levels of critical thinking. Workers were less likely to engage their brains if they felt comfortable leaving the task to AI.
A study by Gerlich in 2025 had similar findings, noting “a significant negative correlation between frequent AI tool usage and critical thinking abilities”.
Another 2025 paper also cautions that excessive reliance on AI tools may lead to a trade-off between critical thinking and speed, potentially eroding meta-cognitive skills.
And research at Cornell University in 2025 found that people who use Large Language Models (LLMs) to write essays build fewer neural connections: “over four months, LLM users consistently underperformed at neural, linguistic, and behavioural levels”.
4-0 to human copywriters
5. AI costs you time
That trade-off between critical thinking and speed really needs unpacking too…while generative AI tools can spew out content faster than a speeding bullet, the time required to effectively use and apply that output is considerable.
It’s super speedy at responding, but one of the biggest hidden costs of using gen-AI is – ironically – time. The time spent on prompt engineering, and then cleaning up the output – fact-checking for hallucinations, editing, fixing tone, making it unique and specific, and injecting humanity – is actually immense. If you want to use it to support your content generation, it’s not a quick fix or a copy-and-paste job.
Marketing teams increasingly find themselves spending almost as much time, if not more, editing, fact-checking, and humanising AI-generated copy as they would have spent writing it from scratch. That initial time-saving selling point evaporates and they end up doing all the heavy lifting without any core benefit. Consider outsourcing to a human copywriter – like me!
5-0 to human copywriters
6. AI has a drinking problem
ChatGPT ‘drinks’ 500ml of water for every 5-50 prompts it answers. Analysis of the human-written data given to these large language models for training – and producing responses for us – requires huge amounts of processing power, which generates heat. Vast quantities of water are then needed to cool the supercomputers generating this heat.
The new AI search summary on the Google search results page uses about 10x as much energy as a traditional Google search. Google’s carbon footprint is also up by 48% over the last five years, largely driven by AI. And Microsoft’s carbon emissions went up by about 30% last year, purely because of AI.
So even if you’re using the free versions of these tools, the environmental cost is immense.
6-0 to human copywriters
7. AI is indiscreet
ChatGPT may seem like a benevolent advisor or a helpful colleague but is it really? While on the surface it’s super friendly with a trustworthy tone and an overly helpful persona, behind the mask it might just be a sinister overlord gobbling up zillions of global human prompts now that it’s exhausted the world’s supply of training data.
As the University of Sheffield rightly cautions: “anything you enter as a prompt may be used by the company that owns the tool to help train it and develop it. This means that generative AI may harvest confidential or personal information.”
So it’s crucial that you avoid inputting private, confidential or proprietary information into gen-AI tools.
DuckDuckGo – the internet search engine that “actively protects your personal information” has launched its own generative-AI tool that promises “free, private chats, anonymised by us” and “no AI-training on your conversations”. Could this be a game-changer?
7-0 to human copywriters
8. AI is biased
A 2025 study led by the University of East Anglia found that ChatGPT exhibits biases in both text and image outputs, leaning toward left-wing political values, which raises questions about fairness and accountability in its design. Dr Fabio Motoki says: “Our findings suggest that generative AI tools are far from neutral. They reflect biases that could shape perceptions and policies in unintended ways. It seems that ChatGPT censors the generation of output without any good reason, but only from a Right-wing perspective.” Research by Stanford had very similar findings.
Bias has been picked up in AI visuals as well as text. Studies in 2017 and 2023 discovered obvious cultural bias in text-to-image tools.
Humans are biased so surely all the gen-AI tools that have been trained on human output are therefore also, inevitably, biased. Weise & Metz seem to agree, based on their 2023 research.
8-0 to human copywriters
9. AI is meh
By its design, generative-AI is simply a text prediction tool – all it does is synthesise existing information. Large language models can’t actually think, they just learn patterns, and they are all drawing from the same sources.
The result? ‘Good enough’ content that wildly underperforms because it’s bland and lacks original concepts.
Gen-AI automatically repeats phrases, structures and ideas, so the output is broadly the same for every user. This means any content you ask it to produce is repetitive and homogenous, uses tired phrases and lacks creativity. Copy is derivative – simply echoing what’s already out there.
To become a thought leader you need to offer unique personal insights and angles – particularly in industries like health, technology, education and B2B services, which thrive on expertise.
To engage with customers and prospects, you need to differentiate your brand or business and stand out from the sea of blandness – offer fresh perspectives, develop new ideas and speak to people in their own language. Which is exactly what professional copywriters are skilled at. We can drill down into what’s unique about you and your brand – what sets you apart and which aspects of your story will captivate customers. Human copywriters build human connections.
9-0 to human copywriters
10. AI can’t count
Now, as a lifelong copywriter I am all about words – numbers really aren’t my strong suit – but one thing I can do is write to a word count. Unfortunately AI can’t. Don’t believe me? Give ChatGPT a try – ask it for the top 20 running shoes in the UK right now (I’ll wait)….
Did it give you 20? No. Even if you point out its mistake, it will apologise then try again – yet fail again. So much for being helpful.
It’s also unable to write to a word count. And have you tried including text or digits in text-to-image AI prompts? Give it a go.
10-0 to human copywriters
11. AI is riddled with cliches
One of the golden rules of copywriting – avoid cliches like the plague. They’re lazy, stale, tiresome, boring and predictable, and because of this they instantly switch readers off. ChatGPT output is, predictably, jam-packed full of clichés because it works on probabilities – the likelihood of one word appearing after another increases when they’ve appeared together so many times before.
11-0 to human copywriters
12. AI can’t weave compelling stories
Sure, ChatGPT AI can string together sentences, but it’s no good at constructing a compelling, cohesive narrative, producing engaging long-form copy, or writing an insightful multi-part case study.
It’s just not up to the job.
If you need 300 words or fewer (incredibly weak in SEO terms), it might just be able to stumble to the finish line. But ChatGPT really falls down on longer form copy where it clearly lacks the ability to maintain a reader’s attention with engaging, factually accurate and valuable content.
Gen-AI doesn’t do strong narrative arcs, and it struggles to weave complex ideas into a clear and persuasive story (which is essential for showing expertise, and building engagement and trust). Human copywriters understand flow and structure – and people. Which leads me very nicely onto my next point.
12-0 to human copywriters
13. AI lacks humanity
Language, like people, is complex and nuanced. And context is key. Gen-AI doesn’t understand this, as it’s simply a text prediction tool. It’s not conscious and it can’t think, draw from experience, or feel emotions. AI can’t use reasoning or logic, or share anecdotes from its younger years.
Its default output is known for sounding robotic – even if you work to try to train it in your tone of voice, it still won’t actually sound like you. And it’s you that people want to get to know when they read your blogs and posts.
Emotion is incredibly important if you want to sell. According to Harvard Business School professor Gerald Zaltman: “95 per cent of our purchase decision-making takes place in the subconscious mind.”
Copywriting is a craft. Us human copywriters can articulate complex ideas clearly, handle sensitive topics with care, and build genuine rapport. We can call on our intuition, empathy, emotional intelligence and psychology. We can write with personality, opinions and wit, and we can put ourselves in other people’s shoes – climbing inside readers’ heads to see what’s likely to get those neurons firing and motivate action. We’ve tried and tested different techniques over many years, to discover what works with people…and what doesn’t.
Tone-of-Voice expert Vikki Ross curates a collection of wonderful slogans and ads that AI could never have written.
13-0 to human copywriters
I’ll leave you with this quote from The Atlantic –
“ChatGPT can write you anything, but it can’t write you anything good.”
Generative AI is amazingly useful for so many tasks – but not copywriting.
So if you’re spending too much of your valuable time fixing AI’s output, or your copy lacks wow factor and isn’t delivering results, it’s time to call in the professional! With 30 years of copywriting experience, a background in journalism, and a deep understanding of user behaviour, I write copy that connects. Plus, working with me is apparently a dream.
Whether you need standout website messaging, an audit to sharpen conversion, or you simply need a professional wordsmith to transform drab drafts into sparkling copy, I’d love to help.
*Disclaimer – No AI was used in the writing of this blog. Not even spellcheck. So if there are any mistakes, they are all mine.
