Here's a joke for you, except there's no punchline. In today's ever-evolving digital landscape, the way we write is being quietly transformed, and it's not just a shift, it's a fundamental reimagining of how content gets made. As we navigate this complex tapestry, one thing becomes crystal clear: leveraging the right words has never been more pivotal. Whether you're a seasoned marketer or just dipping your toes in, this comprehensive guide will unlock the insights you need to truly elevate your craft and supercharge your results.

If that paragraph made you feel slightly unwell, good, that was the point. And you didn't even need an AI writing detector to know that no human wrote that. You felt it. It's a writing style that once seen, cannot be unseen. Perhaps more so to trained writers, but I think most people can feel what AI is putting down.

So to my fellow writing friends, or clients who find themselves needing to use AI to write, quite frankly because it saves time and time keeps on slipping… slipping, here is my roundup of top AI writing tells.

The 5 dead giveaways of AI writing

For the skimmers in the room, these 5 turn up so often in AI writing that spotting any 2 of them together is usually enough to know what you are looking at.

The glaring red flag words to ban outright

Sadly these words did nothing wrong except attract the deep undying devotion of ChatGPT, Claude and co. Because of how horribly, badly, these words are being exploited, I'd invoke a hardcore, mandatory ban on these beauties, today.

"It's not just X, it's Y"

Our LLM chatbot friends love a contrastive negation (I had to look that term up), which is a fancy way of saying the sentence tells you what something is not before it gets round to telling you what it is. "This isn't just a tool, it's a way of working." "It's not about the features, it's about the outcome." Chatbots go for this because it sounds clever. But it's a trick rather than a thought, and the moment you can name it you'll see it in every LinkedIn post, every product description, and bus stop ad for the rest of your life. Sorry about that. It's not human. It's AI.

The word "delve"

Here's a word doing a remarkable amount of incriminating work. "Delve" was always a slightly bookish word that most people went their whole lives without writing down. Then ChatGPT arrived and started delving into everything, to the point where academic researchers noticed the word genuinely spiking in published papers after late 2022. Me mate 'delve' does not travel alone. Wherever you find "delve" you're sure to meet "navigate", "underscore", "leverage", "harness" and the rest of the hot air verb family.

"In today's fast-paced world"

Or "in today's digital landscape", or "in the ever-evolving world of", take your pick, they all stand out as glaring examples of AI prose. We call this 'throat-clearing', the sound a chatbot makes while it tries to sound interesting. If this device missed the boat as a throat-clearer (love that term!), it's often roped in to provide the 'and so, in summary' bit that AIs like to close their essays on. At Koinly, which 'delves' into the fast-paced world of crypto, our writing team had to ban these sentences as early as 2023. AI loves this construct for tech and futurism stories. Personally, 'in today's fast-paced world' has become an inside joke with fellow writers. Needs a t-shirt.

The rule of 3. Over and over and over

Three is a satisfying number and good writers use triplets all the time, but a chatbot uses them compulsively, so you get three adjectives where one would do. Three examples, three little clauses, paragraph after paragraph. Like "clear, concise and compelling", "faster, smarter and more reliable". One triplet is rhetoric. A triplet in every single paragraph is a nervous tic. Watch. For. This.

Bonus: the "in conclusion" that concludes nothing

You reach the end of an AI masterpiece to discover a final paragraph helpfully titled: "In conclusion". And in conclusion, this final para says nothing. A chatbot writes this way because it was trained on the classic five-paragraph essay structure that traditionally ended with a summary. I have a hard rule against the use of a summary paragraph, it's a massive AI giveaway.

Sticking around for more gems? Here's Everwilde One's full list of words to watch, if not ban, for AI writing:

The words that commonly show up in AI writing

Inflated verbs

Like any novice writer, AI goes for fancy verbs that come off as hot air. There is probably good, natural use for some words such as 'drive' and 'facilitate' for instance. And sometimes, as a human, you have a penchant for certain words. Frankly, the word frankly is one of my faves. But I would flag these for being overused, so if they're in your linguistic toolbox, use 'em lightly if at all.

Inflated adjectives

As with their favourite verbs, AI wordsmiths love adjectives that grade something as impressive instead of telling you what it is. Aka, filler.

Abstract nouns

Nouns that float somewhere above the actual subject. "Landscape" and "realm" are the worst of them. So vague, so AI.

Transitions and connectors

The words that bolt one sentence onto the next. Human writing uses these too, but sparingly, and the tell is the frequency: a draft that opens three paragraphs in a row with one of these was assembled rather than written. A lovely tell.

Hedges and softeners

Words that quietly drain a sentence of its nerve. A chatbot hedges because it has been trained to avoid being wrong, but a writer who knows the subject is willing to just commit.

The throat-clearing openers

Please delete these moving forward. Instead, use the BLUF principle to get the most important information up, first.

The writing style and sentence structures AI uses

These sentence patterns are probably AI's attempt at sounding conversational. Bless. Unfortunately due to overuse, and frankly uselessness if you're trying to be clear and plain, like how we do it at Everwilde, these sentence patterns are a red flag for bad bot writing.

The rhetorical question as a section break

We've covered contrastive negation (not x, but y), and the rule of 3, so let's continue with another famously AI sentence structure, rhetorical contrast. "So what does this mean for your business?" "The result?" "Why does this matter?" A chatbot loves to pivot between ideas with a question, because a question is an easy hinge to swing a new section open on. Used once, in the right place, it is perfectly fine. Used as the standard joint between every single section, it reads as an AI template. It's the pattern, not the individual words, that paint the picture.

Setup-payoff fragments

Ever get the sense that Claude is all about the dra-mah!? Mine is a pro at chucking in thespian-grade nuggets like, "the result? A mess." "The catch? There isn't one." "Our advice? Don't." It is the written version of a dramatic pause, and a chatbot loves it because it feels punchy and confident, but a whole piece built out of these stops feeling like an argument and starts feeling like a series of small magic tricks performed back to back. My current favourite, as seen on LinkedIn, is the heart-to-heart use of 'honestly'. And honestly? I used AI to come up with this point of view. Let that sink in.

Uniform cadence

This is the most durable tell of the lot, and the hardest one to fix with find-and-replace, because it is not about any word at all. Chatbots tend to produce sentences of a very similar length, one after another. Human writing moves around. I'm famous for never-ending sentences that are a bit rambly, partly because I write the way I speak. That variation has a technical name, burstiness, and it is what most AI writing detectors are really measuring. Variety is a spice, right?

AI writing formatting habits

These are the tells that live in the punctuation and the layout rather than in the words themselves. Let's start with the tell-iest of all tells, the classic, the unstoppable, the em-dash.

The em-dash

The famous one. The one everybody points at. Chatbots lean on the em-dash far more than the average writer. Probably because the em-dash is genuinely good as a linguistic tool. But that time is over. Whether it's needed or not, at Everwilde One we avoid em-dash completely. There is always another way of saying something without having to deploy an em-dash.

Title Case On Every Header

A chatbot tends to capitalise every significant word in a heading, the way this very subheading is capitalised. It's a small thing, but using title case all the way down a page adds to that AI generated feeling.

Of course, every brand makes its own styling choices and for some, title case has a place. But if I were coaching a content team on how to make their work feel less AI-generated, I'd almost always recommend sentence case for headings.

A bullet list for absolutely everything

Some ideas genuinely are a list, and this post uses plenty of them, but most ideas are a paragraph. A chatbot breaks prose into bullets by reflex, because lists are easy to generate and they look organised and tidy, so when an argument that should have flowed as connected sentences instead arrives as seven disconnected fragments, that reflex is what you are looking at. Recently I was building out a DeFi topic cluster for a client. The subject matter was heavy on detail. In a series of deeply technical H3 sections, ChatGPT kept interrupting a single thought with bullet points in an attempt to "chunk down" the information. It went a little something like this. I'll use fruit.

If ever you find yourself with a hankering for a tropical fruit salad including

  • mangosteen
  • durian
  • dragon fruit

you should plan a trip to Bali for the ultimate foodie experience.

Stunning. Please don't.

Suspiciously perfect

Real writing carries small inconsistencies, the little fingerprints of the person who made it. Perfection is a tell. This is crazy advice, but if you want to sound human, sprinkle in a mistake or two. Mistakes can be endeering, like bambi.

Add these to your 'how to write' skills

This is the list I use in my brand tone of voice Skills at Everwilde One. It's a roundup of the words, sentence structures, rhythms and devices that quietly mean a piece was almost certainly written by a chatbot. As I've said, not all of the words are tells in their own right, and some industries or contexts rely on vocab from our banned list. So use this as a foundation, then adapt it to suit your brief.

A word of warning: Simply adding 'no filler or fluff' to your writing prompt is not enough. Nor is copy-pasting a list of words to 'watch out for'. The prompt, or better yet the Skill source behind it, is like copper. It needs regular polishing or it starts to tarnish. And don't assume your AI colleagues actually read the instructions you give them. Insist on it. Refine your prompts constantly. Nurture them a little. That's the real trick to keeping AI output sounding human, or at least as human as possible.

Free resource

The Everwilde One AI writing tells Skill

This is the actual editorial Skill behind this post. The full list of banned words, the sentence structures to avoid, the cadence rules and the pre-publish checklist, written as a reusable Skill file. Download it, copy it into your own AI tool's instructions, and adapt it to your brief.

Download the Skill (.md)

Sources and further reading

Bloomberry's database is a gold-mine (and a great example of content as a product). I've bookmarked it and a few other resources. You may find these helpful too.

Michelle Legge
About the author

Michelle Legge

Founder · Everwilde One

Two decades across brand storytelling, SEO, content strategy and now GEO. The full arc of digital, not just one chapter of it. Specialist in AI visibility, entity strategy and Search Everywhere Optimization. EMEA-focused, globally experienced.

LinkedIn Full profile About everwilde one Email

Not showing up on ChatGPT?

Probably because how you describe yourself is inconsistent or incomplete. Everwilde One's AI Search Entity Builder is a free tool that helps you fix that. 3 minutes to map out what your business looks like from an entity point of view. No catch.

Or, if you'd rather talk it through, book a 30-minute call. I'll tell you what I'd do.

Help me define my AI search entity
← Back to all posts