Composition

Unpromptedly narrating what the text or the model itself is doing, deciding, planning, or about to do, instead of just doing it and producing the output. The reader gets a voiceover of the writing's own moves and the model's deliberation. Literal chain-of-thought residue that has no place in the final output, serving solely to pollute and bloat the content.

“What that changes in the design is smaller than it might appear, and what it changes is worth being precise about.”

A point (often, but not only, a question) preceded by a paragraph of its own evidence, so by the time the model spits it out it has already been made two or three times over. Increasingly common and very much linked to reasoning leaks.

“Is next-day delivery available in this region, and what's the current rollout status? One internal doc says it launched in the spring. Another says it's still being tested. A teammate mentioned it was paused in a message last month. Every other region has a clear answer, and if this one doesn't, that's a conversation with the partner rather than something we can fix ourselves.”

Opening with what the output is about to do, a preface to the point, or a restatement of the prompt, instead of delivering the point. Includes structural announcers that name the count or shape of what follows ("Two constraints shape the design"), throat-clearing frames ("The more important point is..."), and paraphrasing the question back. The sentence sets up the answer instead of being the answer. Sits in the signposting family with compulsive-counting (states the number) and enumerated-prose (the "The first... The second..." delivery); this is the announcer that precedes them.

“Two constraints shape the design.”

Stating a minor or uncontroversial point just to defend it as if anticipating an objection nobody was going to raise.

“We are setting this out in full rather than quietly changing the recommendation, because the failure mode is the reason it matters.”

Closing by restating the answer and looping it back to the original question, instead of stopping once the answer is given. The reply has already delivered the point, then bolts a summary of itself back onto the ask ("So, to answer your question, X does Y").

“So, to answer your question: yes, the employee can be added to the app.”

The model reuses one of its own words or phrases from earlier in the same document as if paying it off, when it's really the same narrow vocabulary surfacing again under sustained topic pressure.

“quietly become the real source of truth ... nothing you've ever believed can quietly disappear”

The ending stacks clause after clause instead of landing one point like the model is reluctant to actually stop.

“But you can't optimize what you're mismeasuring, and a wrong metric is worse than no metric because it actively steers. If you only change one thing: stop measuring individuals by output volume, start measuring the system's ability to deliver working software, and ask your engineers what's in the way. The last one is free, and it will tell you more in an afternoon than a quarter of velocity charts.”

"What I'm going to tell you; what I'm telling you; what I just told you" applied at every level of the document. Every subsection gets a summary. Every section gets a summary. The document itself gets a summary.

“In this section, we'll explore... [3000 words later] ...as we've seen in this section.”

ESPECIALLY COMMON IN TECHNICAL WRITING. Rapid-fire listing of historical companies or tech revolutions to build false authority.

“Apple didn't build Uber. Facebook didn't build Spotify. Stripe didn't build Shopify. AWS didn't build Airbnb.”

Making a single argument and restating it in 10 different ways across thousands of words. The model pads a simple thesis to feel "comprehensive" by rephrasing the same idea with different metaphors, examples, and framings. An 800-word argument becomes 4000 words of circular repetition.

“The same point, restated eight ways across 4000 words.”

Repeating entire sections or paragraphs verbatim within the same piece. This happens when the model loses track of what it has already written, especially in longer pieces. A dead giveaway of unedited AI output. Less common nowadays thanks to 1M context windows.

“The same section appeared twice, word-for-word identical.”

The rigid formula where AI acknowledges problems only to immediately dismiss them. Always follows the same flow: "Despite its [positive words], [subject] faces challenges..." then ends with "Despite these challenges, [optimistic conclusion].".

“Despite these challenges, the initiative continues to thrive.”

Explicitly announcing the conclusion with "In conclusion", "To sum up", or "In summary". Competent writing doesn't need to tell you it's concluding, it's obvious and the reader can feel it. AI signals its structural moves because it's following a template, not writing organically, it doesn't know when its actually going to end.

“In conclusion, the future of AI depends on...”