Back to Main
This Week in AI — 📋 An updated look at AI tells
AI | Content Marketing

This Week in AI — 📋 An updated look at AI tells

September 30, 2025 | By: Dave Isaac

Each week we’ll gather headlines and tips to keep you current with how generative AI affects PR and the world at large. If you have ideas on how to improve the newsletter, let us know!

What You Should Know

 

Revisiting the List of AI Tells

By now, you’ve surely seen the slander against em dashes. Since ChatGPT started overusing the punctuation close to a year ago, many folks have claimed any em dash is a surefire sign of AI writing. (“ChatGPT hyphen?” Tell that to Jane Austen.) Of course, that’s not true. I type Alt+0151 (or Option+Shift+dash on my MacBook) manually, thank you very much.

We’ve seen plenty of columns passionately defending the em dash, and while it has earned the most attention of any AI tell, there are plenty of other, more obvious signs that AI was used for writing.

Identifying them isn’t about deciding what is AI vs. what is human or shaming anyone. Any AI tell is rooted in human writing and can be found in 100% human-written text — guess where AI learned those moves.

We look for AI tells to reduce excessive use of words, phrases, punctuation, and style choices that sometimes appear in every sentence or paragraph, often out of place, and can distract from a message or point.

Eliminating AI tells also makes the work your own. They can signal a lack of effort or quality control for anyone who is aware of them. For those who aren’t, they can make a quote, a news announcement, a social post, and other writing feel generic — the exact opposite of what we’re trying to do as communicators. Think of AI tells as placeholders where you can make the text your own.

The biggest tell — which isn’t nearly as common as it used to be, but remains just as damning —  is starting content with “In the ever-evolving world of…” The contrasting structure of “it’s not X, it’s Y,” or “not only X, but also Y” is another big one, sometimes appearing in every paragraph.

AI tools also seem to have a new favorite punctuation mark: the colon. It commonly shows up in rhetorical short-form that’s meant to mimic spoken emphasis, like “Here’s the thing:,” “Think about it:,” or “There’s a catch:.”

AI is known for uniform sentence structure, typically several long sentences strung together. Since GPT-5 was launched, there seems to be a run on short, three- or four-word sentences so the text is meant to look punchy and staccato style. Some new favorite words have popped up too, like “stark.”

With each new model from OpenAI, Anthropic, Google, and others, there will undoubtedly be a new wave of AI tells. It’s something communications pros need to stay on top of so their work doesn’t look like it was purely outsourced to AI. That’s not to say these words, phrases, or styles can never be used, but there needs to be thorough oversight of AI-generated copy to ensure it still reads like it was produced by a human.

We’ve updated our public list of AI tells, which you can use as part of a prompt or for GPT or Project instructions to weed out some of the worst offenders before AI gets to work (though you’ll still need to edit). If there’s something you think should make the list, let us know!

Elsewhere …

🗑️ How to avoid workslop

What’s happening: Last week, stories about “workslop” — AI outputs that lack substance, according to research from BetterUp Labs and Stanford Social Media Lab  — went viral. Non-power users of AI may have been turned off, but they shouldn’t be because the technology isn’t the problem. You just need to know how to use it well to protect your time and others’.

It’s everywhere: According to the research, 40% of U.S. adults surveyed said they’d encountered workslop within the last month, and that it slowed down their workday as a result, requiring nearly two hours to deal with each instance. Receiving a vague AI output full of empty words instead of clear direction or context can be maddening … yet 18% admitted sending workslop themselves.

Be diligent: The best way to overcome workslop is to be thoughtful about how you’re using AI. Read an output before you share it with a colleague and ask yourself if you’d find it helpful if you were on the receiving end.

If it seems vague or unhelpful, reconsider your prompts and whether you were specific enough in what you were asking for. If you know it’s a weak output but feel stuck, share your prompts with that colleague and ask for help on how to get to a more quality output.

Saving time might be the most-cited benefit of using AI, but if you don’t read what it’s generating, you could actually add time instead.

Quote of the Week

“The broad vision that we’re working towards is one where the JPMorgan Chase of the future is going to be a fully AI-connected enterprise.

“Every employee will have their own personalized AI assistant; every process is powered by AI agents, and every client experience has an AI concierge.”

— Derek Waldron, Chief Analytics Officer at JPMorgan, to CNBC in an exclusive interview

How Was This Newsletter?

😀 😐 🙁