Why this took me six months (and counting)
It took me 6 months to write about this topic because so much keeps changing in the world of LLMs and GenAIs that I have to keep updating the draft.
I am sure you have noticed that even regular emails today sound prim and proper, thanks to ChatGPT’s free version. Unless, of course, your organization has shut its doors on GenAI for security reasons. Then, you can stop reading here.
One of the reasons why it took me 6 months is that I wasn’t so sure this was a problem for B2B teams. I kept telling myself, ‘Sabari, you are being a purist. The benefits far outweigh the problems.’ Until now. Houston, we certainly have a problem!
Cringey AI vocabulary: ‘Fostering,’ ‘forefront,’ and other word salad
As someone who goes through 35,000 to 40,000 words a day reviewing, writing, and editing content, I can spot GenAI vocabulary pretty easily. Words like ‘fostering,’ ‘forefront,’ ‘delve,’ ‘dive,’ ‘akin,’ ‘ethos,’ ‘beacon,’ ‘tapestry,’ and ‘epoch’ make me flinch. Especially when I see these words in CXOs’ LinkedIn posts or emails, all I can think is, ‘I am sorry your comms team didn’t care enough.’
When I read articles with obvious ChatGPT (LLM) syntax such as ‘it is not about…but it is about’, it only tells me that someone has been lazy. They didn’t even bother cleaning up the first draft from LLMs.
If I can spot It, so can SEO bots
You could argue, ‘Sabari, it is just words. As long as it is grammatically accurate, why should I care?’
If I can tell, then SEO bots can most certainly spot them. Tools like Grammarly have started rewriting typically AI-constructed sentences in their paraphrasing services. If your marketing team relies on AI to create your content, it may only take another algorithm update from Google and Bing to send your website to oblivion overnight. And mind you, if your discoverability gets a hit, YOU DO NOT EXIST FOR YOUR CUSTOMERS. HubSpot recently lost 80% of its organic traffic – not because of AI-written content, but because of how brutal the SEO landscape is. The topic of discoverability in the era of Perplexity, voice UI, and new-age search shall be parked for another day.
More importantly, your audiences care. B2B decision-makers in their 50s grew up reading extensively, long before PCs were commonplace. They can quickly spot word salad regurgitated by LLMs.
One of our clients’ CROs was livid when a client executive from a $350M account responded with a stinker to our client’s automated email that reeked of non-personalized AI word salad.
Talking about LLMs, if GenAI still seems like magic, here is a two-sentence explanation of how they work:
1. They cannot think yet. Even the best of reasoning models don’t think in the real sense of the term.
2. They cannot think yet. Even the best of reasoning models don’t think in the real sense of the term.
Armageddon marketing: AI slop on a loop
That brings us to the impending Armageddon marketing. All the models released so far from Meta’s LLAMA, OpenAI’s many models, and Anthropic AI’s Sonnet (not going to even croak about that other one) and also those distilled on top of others like the Deepseek are trained on almost all the human knowledge available on the world wide web. They are now being trained on synthetic data. While SLMs (small language models) and RL (reinforcement learning) techniques are picking up now, what most models are becoming is AI slop trained on AI slop… garbage in, worse garbage out, and this garbage going back in on a conveyor belt.
I am not even talking about LLM hallucinations and factual errors; I’m talking about absolutely obtuse content being produced on marketing tools that rely on LLMs. All publications are quickly realizing how valuable their original information is and locking it down. The only things left to scrape will be AI slop and self-serving rubbish, which goes on a loop.
I recently had ChatGPT do an analysis of industrial IIoT sensors in the Bay Area and then went through each of its sources, and the majority of its sources were the blogs on the websites of industrial IIoT sensor sellers who were marketing themselves. So, chances are your website is being scraped and rewritten by your competitor while your social media team is working on an AI mush that mixes their site’s content built on top of your content.
Well, AI has not always failed. In the right hands, it has yielded significant scale and efficient operations. For one of the leading beauty and cosmetics brands, the ad operations team was optimized to half as their Meta ads were seamlessly adapted for different regional nuances, various attires, skin tones, and languages. This would have taken many iterations and a large team of graphic designers. However, when the brand book was codified and the AI trained through reinforced learning, it kept getting better.
Seven steps to save your marketing from AI slop
We helped some of our clients fix their marketing ops by salvaging their Jasper, Yarnit, and CopyAI pilots. Some of the lessons from these rescue missions are:
1. First, codify your brand’s tone, voice, and persona into the LLM
2. Scrub words that scream ‘LLM-generated’ through ML feedback cycles. Rinse. Repeat.
3. Curb your IT teams’ enthusiasm for automating everything. Some comms are best humanized. They may not appreciate the finer nuances in writing. And that is precisely why marketing must review, and account teams should have the final say.
4. Your SMEs and senior marketing colleagues who understand the domain are going to be far more valuable. Look at AI as a tool to make them more productive, not to replace them.
5. Keep AI away from the younger crop. Like how you would restrict your toddlers’ access to tablets/phones. It stunts their ability to learn, and they don’t pick up core skills, business acumen, and domain knowledge that would make them valuable to your business and clients. Without these, they just cannot make the AI work.
6. Consider a tiered approach to AI implementation, i.e., some of the low criticality copy and content, say internal comms, can be managed with AI.
7. However, a strict no-go zone should be carved out for client correspondence, thought leadership, and executive communications. This can become a significant competitive advantage in an ocean of AI-generated noise.
If you have any questions or ideas on this topic, I would love to hear from you. If you disagree with anything above, we must certainly speak.
The bottom line is that good talent is going to be more in demand because they can use great AI tools and deliver exceptional values. Technology is now table stakes. Talent is back at the center of competitive differentiation.
P.S.
If you received this version, you probably didn’t receive the other version that speaks about how AI is reducing the time-to-market of dev teams through agentic AI. Companies are shipping client-facing applications without a single developer being involved. Smart kids are building front-end apps by connecting AI agents with Replit that talk to each other to produce an entirely hands-off CI/CD pipeline. Of course, if you end up with spaghetti code or tech debt from this hyper-fast cycle, it becomes a gargantuan mess to clean up.