I am a huge proponent of content marketing. I believe that there is still power in sharing your knowledge for free to build know, like, and trust, which become the foundation for people wanting to do business with you and your company.
I am also an avid fan of using AI in the process. AI can summarize long articles and give you bullet points to help you outline what you want to say. It can proofread and even rewrite your text for you.
What I am not a fan of is what is known as AI slop. That is what I call writing an article in AI that usually yields low‑quality, mass‑produced, AI-generated text. It values speed over substance, and output over originality, coherence, or real human effort.
You may think people might not notice, but it’s almost subconscious; we can detect the difference.
AI Slop Detection
Most AI detectors rely on perplexity (not the AI tool) and burstiness.
Perplexity is a measure of how “surprised” a language model is by a piece of text. Low perplexity means a detector model finds the text very predictable. High perplexity means it finds the text unusual or harder to predict.
Burstiness is a measure of how much the structure of your writing varies over time, especially sentence length and complexity, rather than consistency from sentence to sentence.
Most AI detection tools assume that AI-generated text is more predictable and less “spiky” than human text, but this approach is becoming harder to apply as AI models and LLMs improve.
Research groups and universities often view these tools as unreliable. They tend to favor certain writing styles and can be tricked by paraphrasers, translators, or even simple human changes.
Pangram

Screenshot
I was listening to a Verge podcast about AI detection, and they interviewed the CEO (Max Spero) of a company called Pangram.
Pangram is a 3-year-old company that approaches AI detection differently. It stands out because it dives into more complex patterns and uses active learning to train AI writing, specifically on what it means to “tell.”
I took the plunge and subscribed to the free trial. It’s $20 a month to purchase.
I plugged my latest blog post into its interface, and it said 100% Human Written (which it was). It also said that 65% was High Confidence and 35% Medium Confidence. I do use AI for research and then rewrite it in my voice.
Then I checked a client (that I know doesn’t use AI), and it came back with 100% Human Written and 100% High Confidence
Again, I checked a client (that I know DOES use AI), and it came back with 100% AI-generated and 100% High Confidence.
So I am pretty convinced that the program does what it promises.
What AI Does to Text
In the 100% AI text example I noted above, Pangram gives three pieces of evidence that validate its findings. It found 10 instances, 3× above typical. Human-written text of this length normally contains 4 AI-associated patterns. This document significantly exceeds that baseline.
These patterns aren’t used to determine Pangram’s AI score; they help you see why AI text often reads differently.
- Triads
Grouping ideas in threes, a common AI rhetorical pattern
“…clarity, data, and systems.” - AI Phrases
Word patterns that appear far more often in AI-generated text
“The question is not whether…” - Em Dash
Overuse of em dashes where human writers typically would not
10x more likely & 7 found —
It had 3 sections with 3 paragraphs and 3 quotes (overly symmetrical). I also played back the blog post with my Mac voice (which I do with my own blogs every week to proof and make sure it flows and makes sense).
- Text was sing-songy. Meaning it had the rhythm of a drone versus a conversation.
- The points were unclear and lacked a clear problem-solution modality.
- It has a very generic sentence structure with words most people don’t generally use.
Granted, the above is my opinion and not from any tool, but I have years of experience in producing audio and video, so I have a keen sense of context and flow.
Why Human Writing Is Better than AI
You may be saving time and costs by using AI to create your text, but you may be sacrificing reader comprehension and engagement.
Pangram explains that AI has a “shockingly uniform voice” compared to a trained and prolific human author. AI models are optimized to produce acceptable and average‑sounding text.
Human authors often have a unique voice through phrasing and stylistic quirks. Human text just has a different texture that increases comprehension and readability.
AI writing tends to suppress odd or risky phrasing or words and stick to safe, middle‑of‑the‑road phrasing that makes it sound smooth but bland. The real telltale sign of AI-written text is that it just sounds machine-generated.
The bottom line is: “Human writing beats AI because it has a distinctive voice and real lived experience behind it, while AI tends to sound uniform and generic.” (P.S. That was written by AI.)
Closing Thought
I switched to the Verge Podcast because it was more human than another AI education podcast I was listening to. It was clear the host was reading AI-generated news stories and then injecting his personal commentary.
My goal is to learn as much as I can. I know I have biases, but my Pangram test validated what I was feeling about AI text: It is harder to comprehend and retain.
I hope you feel that my posts are written with a style and come from a place of experience that helps you make the most of these new marketing trends and tools.
______________________________
Comment below and share your thoughts, ideas, or questions about business-to-business sales and marketing today! Do you have a sales or marketing communications strategy that works for you? What tips or techniques can you share that work for you and your business?
To learn more about this and other topics on B2b Sales & Marketing, visit our podcast website at The Bacon Podcast.





