The giveaway signs of AI-generated content change every day. You might have your ban list compiled, but you need to keep updating it. As the models adapt, they find new ways to pretend to be human. And that involves new signature words.

If you're using ChatGPT Claude and Gemini regularly you might have already spotted them. If you're still amazed just by the fact they can write content for you, you might be falling into traps.

Check out this list, understand what's going on and don't let your content be let down by these new favourite phrases of large language models (LLMs). Use it alongside my previous list from February.

Here is a comprehensive list of the words that keep being added and what's wrong with each one.

LLMs love these words and phrases. Don't get caught out.

This adverb has become the LLM's favourite way to add drama to anything happening slowly. Watch for "quietly building," "quietly dominating," "quietly transforming," because the word adds no information and points straight to an AI hand on the keyboard.

Models tend to use "shift" whenever a change happens, big or small. If your draft contains "a shift in thinking," "the shift towards," or "this represents a shift," replace it with the specific verb that describes what changed.

When AI wants to reinforce importance, it tells the reader something "matters." Look for "this matters because," "what matters here," and "this difference matters," or even just “this matters” on its own. None of these are needed if what you’re saying really does matter.

This verb has become a stand-in for any abstract influence the model can't define. "Shapes the future," "shapes how we think," "shapes the conversation," all sound meaningful but tell you nothing about what is happening.

Watch for any sentence where a message, point, or idea "lands." The models use this verb constantly as a softer alternative to "works" or "resonates," and it's become so common that it now reads as filler.

This word creeps into AI content everywhere, often paired with "matters" to insist that something matters. No one talks like this in real life. Search your draft for "actually" and delete all instances except for where it actually helps you make your point better.

The word itself is fine, but watch for it being used as an intensifier in front of any noun the model wants to inflate. "Real growth," "the real reason," "real value," and "real impact" all point to the AI reaching for emphasis it hasn't earned through the content.

LLMs have started attaching "earn" to anything abstract a person can possess. "Earn the right to," "earn attention," "earn trust," "earn your audience," all sound principled but the verb is doing rhetorical work the rest of the sentence should be doing.

This phrase has become a tell across AI-generated business and personal development content. "Do the work," "trust the work," "the work itself," all carry a vague reverence for effort without specifying what that effort involves.

Watch for the metaphorical use of "hold." "Hold space," "hold the line," "what you hold," and "the grip it holds on you" are abstract usages that the models use when they want emotional weight. One use in your article? No problem. Any more? Get the red pen out.

Models might say "pull" whenever they want to describe a force without naming it. "The pull of," "feel the pull towards," "a stronger pull," all sound evocative but they sidestep saying what the motivation is. If you can be more specific, do so.

This verb is everywhere in AI-written business content. "Decisions compound," "small actions compound," "the benefits compound," all sound mathematical and impressive, but the word has become a default substitute for "add up" or "grow." Avoid it slipping in arbitrarily.

When used as a noun, "signal" has become a giveaway. "Send the signal," "the signal that," "a strong signal," all sound strategic, but they often replace concrete description with something more abstract.

This phrase started in culture and has been adopted enthusiastically by AI models trying to sound contemporary. Watch for "built different," "built for the bold," "built to last," all of which now read as stock LLM positioning rather than considered writing.

Still the biggest giveaway: contrast reframe sentences

Contrast reframes deserve their own mention because they survive even the most carefully-crafted prompt. The structure looks like this: "It's not about X, it's about Y." Or "This isn't a tactics problem, it's a positioning problem." Or "It's far more than X. Focus on Y."

The pattern works by dismissing one thing in order to introduce another. It feels punchy at first. It creates the illusion of insight by setting up a fake opposition and resolving it in one move. LLMs have learned that humans respond to this technique, so they produce it constantly.

The fix is to delete the dismissed half. The contrast. Say only the thing you mean. Be more direct with your words and your readers will thank you. Once you start spotting contrast reframes, you'll see them everywhere. Don’t make them part of your content.

What separates human writing from AI writing in 2026

Your ban list should be a living document. Every model update brings new signature phrases and retires old ones. Keep watching, keep updating, and keep editing. Don't treat your audience members like they're stupid. They're not. AI is a tool that can expand your work. But letting it add its signature phrases will undermine your professionalism.

You don't need to declare an outright ban on every single word in this article. Just because a few of these words appear in a LinkedIn post, it does not automatically mean it was written by AI. But your readers are smart, so treat them like they are. When you use too many of these words in a way that sounds contrived or formulaic, you undermine your skills and underestimate their intelligence. And no one wants that.

Catch my free workshop , how to write with AI.