In case you hadn’t noticed, generative AI is everywhere. It’s transcribing audio files. It’s creating video thumbnails. It’s predicting how a Wes Anderson-made Star Wars might look. Its presence is felt in virtually all avenues of digital production. We in the creative industries are frequently reminded of this technology’s game-changing nature.
As, of course, are our clients.
Affordable, tireless, and informed by more data than a learned specialist can accrue in a lifetime, it’s easy to see the allure of generative AI to cash-strapped SMEs that need a steady stream of web copy.
We became curious about how many of our SME clients used AI copywriting tools. If so, how are they working out? We decided to find out. We surveyed 76 SME marketers and the results were so interesting that we figured they merited a blog post.
Unsurprisingly, we found that a 77% majority had already used AI platforms to generate marketing content. More surprisingly, the majority did not use the platforms that were aimed specifically at marketers.
Almost half (43%) pay a monthly fee for their use, with 29% paying under £40 per month and 14% paying above £40.
Perhaps our most significant finding was how few respondents were prepared to publish pure AI content with 85% saying they needed to edit AI copy before publication. When we asked how long they typically took, the results were also revealing:
This creates issues with scalability. For marketers to produce AI at scale, they need to allocate a substantial budget for human editing. Otherwise they risk the reputational damage that comes with publishing sub-standard work.
AI-generated copy broadly seems to need some engineering to be fit for publication. This can range from a few minutes to over an hour. But what exactly is wrong with the content generated?
Let’s look at some of the most common culprits according to our respondents.
Digital copy can be many things. For many SMEs and marketers it helps to maintain a brand’s authority on a given subject. This applies to everything from the contents of a product page to thought-leadership posts.
Over 29% of our respondents complained of factual inaccuracies within AI-generated copy.
Their experiences have been validated by some recent high-profile AI inaccuracies.
Generative AI models are trained on vast datasets. As of 2023, Open AI has given Chat GPT access to the entire English-language internet. Nonetheless, these models struggle to distinguish between reliable and misleading data sources.
Remember when Google’s AI tool Gemini was mocked for suggesting that pizza would be improved by adding glue? Or that readers should eat at least one small rock per day? The former suggestion came from a tongue-in-cheek Reddit post and the latter from the satirical site, The Onion.
These are some of the most high-profile AI mishaps. However, all inaccuracies need to be sought out and corrected. Which can be time-consuming.
The biggest failing of AI content, according to 42% of respondents, was its stylistic flaws. As impressive as AI tools can be, their outputs often seem inauthentic. Just as ultra-glossy AI images can feel uncanny, AI copy often appears to misread the prompt’s stylistic expectations.
Marketers and SMEs typically use written content to engage with a particular issue and showcase their expertise. This requires a hefty word count.
This is necessary to engage with the subject in a meaningful way and inform the reader. It can also benefit important SEO metrics like dwell time and scroll depth.
Therefore, it’s frustrating when AI platforms generate copy that falls short of the required word count. No matter how many times they’re prompted.
Nine percent of respondents found that the copy generated by AI did not match the required word count. This necessitates the rephrasing or complete rewriting of prompts. Over and over again.
A detailed and specific content brief is the surest way to get great results from a human copywriter. Likewise, the clearer and more detailed a user’s prompt, the better the results from AI platforms.
At least, in theory.
In practice, many respondents had to do a lot of prompting to get the desired results. 28% of respondents had to spend “a lot of time” writing and rewriting prompts to get the response they wanted.
This issue can create further operational bottlenecks when deploying AI at scale. When that happens, the time and cost incentives can quickly melt away.
Finally, 8.86% of respondents had difficulty with their AI platforms’ interfaces. Platforms felt “over engineered”, with too many features that they didn’t know how to use. If marketers found these tools too cumbersome, there’s little hope for small or microbusiness owners.
Readers will know that our offerings come from the squishy organic brains of human writers. As such, they may assume that we want to see AI copywriting platforms expunged from the Earth.
But the truth is we’re pretty sanguine about AI
ChatGPT, Bard and Jasper have been accessible to our clients for a while now. We’ve had to engage with the opportunities that they represent.
We’re all for AI. We’re confident that it’s not an existential threat to human copywriting. Although naysayers within the business have predicted that AI will mean the end of copywriting, our revenues have actually increased since its arrival. In fact, we may have underestimated the business opportunity that AI represents for us. After all, someone has to do all that editing to get the content into shape.
Generative AI is only going to become cheaper, more capable and more ubiquitous.
Nonetheless, its inherent limitations are just that.
Inherent.
Algorithms can’t learn to think for themselves. They can’t form opinions or reflect on the data that they assimilate. To use a crude analogy, they eat and selectively regurgitate. Thus, our clients will still need flesh-and-blood copywriters to lend written content a human touch. And we see that as an opportunity, not a threat.