An AI tell-list checker points at paragraphs that were assembled.

The entries are phrases and shapes rather than crimes: stock transitions that announce a point instead of making one, verbs inflated to sound consequential, everything operating within a landscape of something, the symmetrical three-item list that fits any subject.

Add the punctuation habits. Em dashes used as glue between half-thoughts. Colons introducing a list that could have been a sentence. Sentence openers repeating within a paragraph because each was generated from the same template.

The checker counts these per draft and reports density. One occurrence is nothing; a cluster in every paragraph is a signal that the section needs rewriting rather than patching.

It cannot tell you who wrote the text. Plenty of people write in these registers, especially when writing to a formula: agency boilerplate, technical documentation, SEO copy that was templated long before generative models existed.

It also cannot tell you whether the text is good. A draft can be free of every tell and still be wrong, dull or unsourced, and a draft heavy with them can still be accurate.

What it can tell you is where to spend editorial attention. A paragraph with four stock phrases is usually a paragraph with no specific content, and that is a diagnosis you can act on.

Density of listed phrases is the headline number, but the neighbouring checks matter more for editing.

The wrong response is a find-and-replace against the phrase list. Removing delve does not produce a sentence with a point; it produces a sentence with a hole where a verb was.

The right response is to ask what the paragraph claims and write that. Most tell-list paragraphs dissolve at this stage because they were assembling context rather than making an assertion, and an assertion is shorter.

Keep the specifics. A sentence with a number, a name or a consequence in it is almost never on the list, because those words come from the world rather than from the distribution.

Every brand has its own tells. Phrases that were fresh two years ago, a signature construction everybody on the team overuses, a competitor's vocabulary that keeps creeping in.

A house list is worth maintaining because the published list is generic by design. Add your entries with the same discipline: a phrase, an example of the better version, and a note on when the phrase is legitimate.

Review it twice a year. Lists grow and rules that once caught something useful start flagging standard industry vocabulary.

Run the check on the first draft to see what the draft is doing, then again after editing. The second number is the one that matters, because a draft that started dense and ended clean is a draft someone worked on.

Do not gate publications on the tell-list alone. Gate on unsourced claims and factual risk; treat style density as an editorial note, since the acceptable register varies by format.

Pair it with a named reader for anything published under a company's name. The numbers narrow the reading; they do not replace it.