The tells of unedited AI writing are a small, recognizable set of patterns: compulsive hedging, filler phrases like "it's important to note," lists where prose belongs, transitions that connect nothing, inflated vocabulary, suspiciously symmetrical structure, and false balance that refuses to take a side. None of them is an error. Each is a default (the safest, most average choice) and together they produce prose that reads as machine-made even when every fact in it is correct. The good news: once you can name a tell, you can remove it, usually in a single editing pass.

This is a recognition skill, and it's downstream of the larger discipline in quality and editing. Editing explains why the human pass matters; this piece is the field guide to the specific patterns that pass should hunt. Learn the fingerprint and you stop reading hopefully and start reading diagnostically, spotting the tell, then reaching for its fix.

Why these patterns read as generic

Every tell traces back to one cause. A language model generates the most statistically likely continuation, which means it gravitates toward the most common phrasing, the most balanced position, and the most uniform structure. That's the average of everything written on the subject, and average, by definition, has no distinguishing features.

So the tells aren't random quirks. They're the visible symptoms of writing-by-average: hedging because the average claim is qualified, filler because the average article pads, false balance because the average piece avoids commitment. Removing the tells isn't cosmetic. It's the act of pulling a draft away from the average toward something a specific person, with a specific point of view, actually wrote.

What are the most common AI writing tells?

Compulsive hedging

The draft qualifies everything. "This can sometimes be a factor that may, in certain cases, contribute to…" Every claim arrives wrapped in cushioning (can, may, might, often, generally, in some cases) until it asserts nothing you could disagree with, or quote.

Why it reads wrong: hedging signals an author unwilling to commit. It also kills citability: a hedged sentence can't be lifted into an answer because it doesn't actually say anything, which matters for getting cited by AI engines.

The fix: where the evidence supports a direct claim, make it. "This may be a contributing factor" becomes "This is usually the cause." Keep hedges only where genuine uncertainty exists, and then say why it's uncertain instead of just softening the verb.

"It's important to note" and other filler phrases

A family of stock phrases that announce themselves before saying anything: it's important to note, it's worth mentioning, when it comes to, in today's fast-paced world, at the end of the day, that being said. They're throat-clearing: sounds before the signal.

Why it reads wrong: these phrases are verbal tics shared by millions of generic drafts. Their presence is one of the most reliable surface tells there is.

The fix: delete the phrase and keep the sentence. "It's important to note that latency matters" is just "Latency matters." If the point that follows is actually important, it doesn't need an announcement; if it isn't, cut it too.

Listy sameness

The draft renders everything as a bulleted list, including ideas that are connected, sequential, or argumentative and should be prose. Three bullets of two words each where a sentence would do; a list of "key considerations" that are really one continuous thought chopped into fragments.

Why it reads wrong: over-listing flattens relationships between ideas. Prose carries logic (because, therefore, but) that a list strips out, leaving a pile of disconnected nouns.

The fix: reserve lists for genuinely parallel, scannable items (steps, options, criteria). When points build on each other or argue a case, write them as sentences and let the connective tissue do its job. A good piece alternates; an AI draft defaults to bullets.

Empty transitions

Connective words that perform connection without making one: Moreover, Furthermore, Additionally, In addition, On the other hand, As such. They sit at the head of a paragraph implying a relationship to the last one that often isn't there.

Why it reads wrong: real transitions earn their place by showing how one idea follows from another. Empty ones are scaffolding the model adds because that's what the average paragraph starts with.

The fix: cut transitions that only signal "here is another sentence." Keep the ones that mark a real turn (a genuine contrast, a real consequence) and make the relationship explicit instead of decorative.

Inflated vocabulary

The reach for the fancier word when the plain one is clearer: utilize for use, leverage for use, facilitate for help, delve into for explore, myriad, plethora, robust, seamless, in order to. The register inflates without adding meaning.

Why it reads wrong: inflated vocabulary signals effort to sound authoritative rather than be clear. A handful of these words is a near-certain tell.

The fix: prefer the plain word every time it carries the same meaning. Utilize → use. In order to → to. Facilitate → help or let. Clarity reads as confidence; inflation reads as padding.

Symmetrical structure

Every section is the same shape: an intro sentence, three bullets, a wrap-up sentence, repeated identically down the page. Paragraphs run the same length. Sentences hit the same rhythm. The whole piece is mechanically even.

Why it reads wrong: real writing is lumpy. Some points deserve a paragraph, some a sentence, some a whole section; a human writer varies emphasis with importance. Perfect symmetry is the fingerprint of generated-by-template.

The fix: vary deliberately. Let an important point sprawl and a minor one be a single line. Mix short sentences with long ones so the prose has a pulse instead of a metronome. Asymmetry is what attention looks like on the page.

False balance

The draft presents two sides of everything ("on one hand… on the other hand") and declines to land. It lists pros and cons, then concludes that "it depends" or "the right choice varies." It never says which option is actually better.

Why it reads wrong: refusing to take a position is the most generic move available. Readers come for judgment; false balance withholds the one thing they can't get from the raw facts.

The fix: take the position the evidence supports. Present the trade-offs honestly, then say which way you'd go and why. "It depends" is only acceptable when you also explain what it depends on, turning a non-answer into a real one.

The find-and-fix table

A scannable reference for the edit. Search for the tell on the left; apply the fix on the right.

Tell	What it looks like	Find	Fix
Hedging	Every claim qualified into mush	can, may, might, often, generally, in some cases	Make the direct claim where evidence supports it; keep hedges only for real uncertainty
Filler phrases	Throat-clearing before the point	it's important to note, it's worth mentioning, when it comes to, in today's…	Delete the phrase, keep the sentence
Listy sameness	Everything bulleted, including connected ideas	Stacks of short bullets where logic should flow	Convert argument and sequence back to prose; reserve lists for parallel items
Empty transitions	Connectors that connect nothing	Moreover, Furthermore, Additionally, As such	Cut decorative transitions; keep only real turns
Inflated vocabulary	Fancy word for a plain meaning	utilize, leverage, facilitate, delve, myriad, robust, seamless, in order to	Swap for the plain word: use, help, explore, many, to
Symmetrical structure	Every section the same shape and length	Identical paragraph and section templates	Vary emphasis and length with importance; mix sentence rhythm
False balance	Two sides of everything, no verdict	on one hand… on the other… it depends	Take the position the evidence supports; if it depends, say on what

A quick tell-hunting pass

Run this as a focused sweep, separate from your accuracy and structure passes:

Search the draft for the filler phrases and inflated words above and delete or swap each one.
Read every paragraph opening. Is the transition earning its place or just decorating?
Find every hedge and ask: is this genuine uncertainty, or reflex? Make the reflexive ones direct.
Scan the shape of the page. Are all the sections the same size? Break the symmetry.
Find every "on the other hand" and check that the piece eventually lands somewhere.
Read one section aloud. If it sounds like a metronome, vary the sentence lengths.

The tells travel together, so removing one often reveals another underneath. That's expected; keep sweeping until the prose stops sounding average and starts sounding like someone in particular.

Where to go next

Removing tells is one pass inside a larger craft:

Quality & editing: the human-in-the-loop: why the human pass is the highest-leverage step, and the failure modes it targets.
An editing checklist for AI drafts: the full run-it-every-time checklist, with voice and cutting passes that catch these tells in context.
How to spot AI-written content: the same patterns, read from the outside, for evaluating work you didn't write.

The tells aren't a moral failing of the tool; they're just the default it reverts to. Editing is how you override the default, and once you can name each pattern, overriding it is mostly a matter of looking.

Common AI Writing Tells (and How to Remove Them)