There are a lot of big feelings about AI in marketing, both for and against. On LinkedIn, AI is either the anti-Christ or the key to delivering humanity into its next golden age. There’s not a lot of nuance between the two. Beyond some high-profile disasters, there aren’t a lot of real-world examples of when AI doesn’t deliver on its promises. As somebody who uses AI a lot, I am intimately aware of its shortcomings.
To be clear, I’m not arguing that you shouldn’t use AI. Just that, as the AI bandwagon rumbles through (and over) marketing teams, you need to be clear-eyed about the limitations. I’m not going to focus on ethical, environmental, or copyright issues, as those have been covered in great detail elsewhere. Instead, I’m going to hone in on marketing, as that’s my bread and butter.
Trust
Marketers usually wind up being early adopters of technology. They have to be to stay on the cutting edge and gather attention and conversion for their clients. But while the tech world and a majority of marketing teams charge ahead with AI, there needs to be some attention paid to the GAP. That is, the trust gap between early AI adopters and the rest of the public.
Most marketers are gung-ho about AI, with a few reservations. The public is a different matter entirely. Just take a look at this recent polling data from Axios:
- Only 18% of young people ages 14–29 say they feel hopeful about AI.
- Over 70% of all Americans think AI is advancing too quickly; 68% of Republicans and 77% of Democrats agree.
- Negative views of AI have risen from 34% three years ago to just over 50% now (YouGov).
Here is the big one for marketers:
69% of digital consumers admit they have far less confidence in material produced by AI compared to that created by humans.
AI isn’t going away; it’s already embedded in many marketing teams. But marketers will need to work hard and carefully to bridge the trust gap.
Inconsistency and brand drift
Here’s the bottom line: AI STILL HALLUCINATES.
It doesn’t matter how good your prompt is. It doesn’t matter how fancy your workflow is. AI still makes a lot of mistakes (even NotebookLM, which is supposed to be hallucination-proof). This may be something that never goes away, no matter how good AI models get, because of AI’s probabilistic nature.
In other words, AI is a really good guessing machine that takes a very educated guess at what you probably wanted in answer to your prompt. If you input the EXACT same prompt 50 different times, you will get 50 slightly different answers. And that inconsistency is an AI limitation that any content, product, or brand team has to contend with.
Branding, especially, is where this can get out of hand. Sure, you can give AI your brand guidelines, tell it what colors to use, give it your in-house brand voice style sheet, and set some limits and rules; it will greatly improve output. It can get close to being on brand. But with branding, close isn’t really good enough.
Workslop
Sometimes it feels like my job is turning into an “AI-corrector”, plucking em dashes out of copy and fixing brand voice to align with the actual guidelines, not Claude’s best guess. I’m not alone.
An estimated $9 million was spent last year on fixing poorly made AI content. That’s pretty far from AI’s big promise of making things faster and more efficient. In reality, it often gums up the works and makes things harder (and more expensive) than they would be if starting from scratch.
Brain fry
One of my biggest issues with gen AI is that we (collectively as the whole of humanity) don’t yet fully understand what using it does to our brains. The Harvard Business Review has identified a new type of burnout caused by excessive use of generative AI. They call it “brain fry”:
- Workers with high AI oversight expended 14% more mental effort and reported 12% more mental fatigue.
- Workers experiencing AI brain fry report 33% more decision fatigue than those who don’t.
- Brain fry workers scored 11% higher on minor errors and 39% higher on major errors.
There has also been much made about “cognitive surrender” lately, in which people defer to AI, essentially letting it think for them. A study from MIT seemed to confirm this idea.
The AI-drafting group showed minimal brain activity and poor recall compared with those writing from scratch. The second, from-scratch group had the highest brain activity. Notably, the third group, which used AI only for research, not drafting, maintained high brain function. This indicates that brain health depends on applying AI as a tool rather than a replacement for thinking.
AI smooths out creative friction
Sometimes the struggle is part of the process. Agonizing over that headline, rewriting the CTA for the twelfth time, going back and forth about where to put a testimonial on a landing page; these are points of creative friction that often make a piece of collateral better. And gen AI is threatening to make it extinct.
Friction is an essential part of creativity. By smoothing out the lumps and making things easier and faster, I wonder what we are losing.
Working through the limitations: What to do instead
I’ve made no secret of my issues with AI. But I know that’s not going to convince anybody to stop using. Heck, I’m not going to stop using it, either. The point I’m trying to make is that we have to be thoughtful about all this stuff, or trouble follows.
We need frameworks on how to work and collaborate with AI. Blue Star recently put together a working manifesto on this. It does a great job of explaining our AI philosophy, but I wanted to put a few finer points on it:
Use AI to support, not replace processes: Sure, you can make a blog post in 15 seconds with Claude or ChatGPT. But will that content actually convert? Probably not, since human-written content outperforms AI slop by a factor of 5. There’s that trust gap.
One way I use AI to support my writing process is to give it my research notes and have it organize them into useful categories: statistics, definitions, benefits, features, and pain points. This is something I would do anyway, and AI does it way faster. To be clear, it isn’t always perfect. I tend to use it more as a reminder of what I read across 50+ pages of notes. It’s not replacing anything, just giving me another tool in my belt.
Use AI to communicate more effectively with creatives: Another use case for AI is mocking up design elements. I am a writer. I suck at design. That being said, I often have an idea for a graphic element, but don’t have the design language to describe it. Rather than spend an hour or more going back and forth with a designer, I’ll knock out a rough sketch in Claude and ask our designer to make something better.
Recently, I’ve also been using Claude to create simple graphics (think flowcharts and the like) for technical blog posts. This saves the designer time by not having to learn niche subjects I already know, and saves the client money by avoiding the need to pay more for an add-on graphic.
Become a centaur chessmaster: This concept comes from chessmaster Garry Kasparov. A player uses a computer guide to play against an opponent in real time. The essential point is that by working with AI to the best of its capacity, with the best of your human experience, you can create outputs that are better than if you simply rely on an agentic workflow or a human copywriter or designer (or chess player).
In other words, it just means “better together”. And that’s really the key point here: if you’re going to use AI, use it to do better work than you would otherwise. The world doesn’t need more slop. Use AI to think in ways that your competitors aren’t.
Need help making sense of AI limitations in marketing?