AI VOCABULARY CHECK
NEVER SENT TO A DETECTOR
Try an example:
words from the AI lexicon
of the draft
lexicon rules triggered
Paste a draft and every one of the 14 lexicon rules runs against it, here, in this tab.
Scanned locally · nothing leaves the tab
Free AI vocabulary checker · runs in this tab

The words that make a draft read as machine-written.

Delve, leverage, foster, robust, seamless, tapestry. Paste your draft and every one of them comes back marked, with the reason it stands out and a replacement you can apply in one click. Your vocabulary range, your repeated words and the ones you lean on hardest are measured underneath.

The support reading

Vocabulary health, measured on the same paste.

Stock lexicon is the loud problem. The quiet one is range: a draft can be clean of every flagged word and still lean on the same six nouns from the first paragraph to the last. These figures update with the box above.

Paste a draft above and the range figures fill in here. Flagged words are reported at any length; these need 25 words.
Why these words

Three habits, and the words that give each one away.

The stock lexicon is not a random list of banned words. It falls into three groups, and knowing which group a word belongs to tells you what to replace it with.

Register, not rarity

None of these words are obscure. Delve, foster and robust have been in English for centuries. What marks them is the rate: a model reaches for them far more readily than a person writing about the same subject, so a cluster of them in one paragraph is unusual in a way the individual words are not.

A rating where a description belongs

Robust, meticulous, crucial and invaluable are verdicts. They tell the reader how to feel about a thing without describing the thing. Generated prose leans on them because a verdict is always available and a specific detail is not.

Metaphors running on autopilot

Ecosystem, landscape, tapestry and paradigm are borrowed nouns that name a mood rather than an object. Sometimes one of them is the right word. Usually it is standing in for a noun the sentence never gets around to naming.

Formal register learned from a textbook overlaps with the register models produce, and detectors cannot tell the two apart. Peer-reviewed work found 61% of TOEFL essays by non-native English speakers falsely flagged across seven commercial detectors, while essays by native speakers came through nearly clean. Who gets flagged, and why it happens covers the study in full.

The list, in full

The stock lexicon, one row per word.

14 rules cover these 24 entries plus their inflections. Leverages, leveraged and leveraging all belong to the row marked "leverage", and the adverbs (robustly, seamlessly, meticulously) belong to theirs. Every row was produced by running the phrase in the first column through the registry the box at the top uses, so what is printed here is what the tool will offer you.

Word or phraseWhy it reads as machine-writtenOffered instead
delve intoPost-2023 generated text uses this verb at many times the human rate. It is close to a fingerprint.dig intoexamine
leverageConsultant-register verb that models default to. "Use" says the same thing.usedraw on
fosterAlmost always replaceable with a concrete verb — build, grow, encourage.buildencourage
utilizeA longer word for "use" with no extra meaning attached.use
navigate the complexities ofA stock collocation that survives from prompt to output almost unchanged.work throughget through
unlock the potential ofA promise with no verb of its own. Say what the thing can now do.make use ofget real value from
robustA rating word standing in for a description. Say what makes it hold up.sturdysolid
seamlessMarketing adjective. If nothing broke, say nothing broke.smooth
meticulousSelf-praise adjective that models attach to any process noun.carefulexacting
holisticTapestry, synergy, paradigm: nouns that name a vibe instead of a thing.wholeend-to-end
tapestrySame rule as the row above.mix
synergySame rule as the row above.overlap
paradigmSame rule as the row above.modelpattern
ecosystemBorrowed ecology words standing in for the actual noun. Sometimes right — usually a placeholder.network
landscapeSame rule as the row above.fieldmarket
a myriad ofQuantity words that avoid naming a quantity.manya lot of
a plethora ofSame rule as the row above.manya lot of
game-changerA verdict borrowed from press releases. Name the change instead.turning pointa real shift
crucialRating adjectives inflate without informing. Cut, or replace with the stake itself.decisivenecessary
pivotalSame rule as the row above.central
paramountSame rule as the row above.the first priority
vitalSame rule as the row above.necessary
quintessentialSame rule as the row above.classic
invaluableSame rule as the row above.worth the cost

Lexicon is one of six tell groups in the registry. See the structural and cadence tells →

The workflow

Four steps, all of them on this page.

1

Paste the draft

No account, no upload, no word cap. The scan runs the moment the text lands, because the word list and the rules are already in this page.

2

Read the flagged words

Every finding shows the word as you typed it, how many times it appears, and one sentence on why that word reads as machine-written.

3

Swap the ones that are wrong

Single-word rules carry their own inflection table, so "Delving into" comes back as "Digging into" rather than "dig into". Replace one instance, or every instance of that word at once.

4

Check the rest of your vocabulary

Range, repetition and the words you lean on hardest are measured underneath. A draft can be clean of stock lexicon and still say "important" nine times.

Why the replacements fit

Replacements keyed to the form you wrote.

Find-and-replace on a word list breaks tense. The single-word rules each carry their own inflection table, so "leveraged" comes back past tense and "delving" comes back as a participle. The phrase rules offer one plain-English phrase, so read the sentence back after a phrase swap.

The applier cleans the seam either way: no doubled space, no space stranded in front of a comma, and the capital restored when the edit lands at the start of a sentence. Where a rule swallows a following preposition — "delving into" — the replacement carries it, so you do not end up with two.

Delving into the data
Digging into the data
She leveraged the archive
She drew on the archive
We had a myriad of options
We had many options
It will unlock the potential of your data
It will get real value from your data

AI vocabulary: common questions

Which words make writing look AI-generated?

The ones this page marks: delve, leverage, foster, utilize, robust, seamless, meticulous, holistic, tapestry, synergy, paradigm, ecosystem, landscape, myriad, game-changer, and the rating adjectives — crucial, pivotal, paramount, vital, invaluable. The registry covers 14 lexicon rules in total, some of them phrases rather than single words, and the full list is printed further down this page.

Is "delve" really a tell on its own?

One instance in a long essay proves nothing. Three in four paragraphs, alongside "leverage" and "robust", is a pattern, and pattern is what a statistical checker measures. The count matters more than any single word, which is why this tool reports counts rather than a verdict.

Should I remove every word it flags?

No. "Ecosystem" is the correct word in an ecology paper and "landscape" is correct in a geography one. The tool marks the word and gives you the reason; the judgement about whether it is carrying weight in your sentence is yours. Replacing a word that was doing real work makes the writing worse, not less detectable.

Will changing these words get me past an AI detector?

Vocabulary is one input among several. Sentence rhythm, structural habits like the "not X, but Y" pivot, and hedging all feed the same class of classifier, and no tool can tell you what a specific detector will output on a given day. What this page can tell you is exactly which words in your draft belong to the stock lexicon, and what to put in their place.

What is type–token ratio, and what counts as a good one?

Unique words divided by total words. It falls as a text gets longer, because common words repeat no matter how varied the writing is, so there is no threshold that means "good" across documents of different lengths. It is useful for comparing two drafts of the same piece, and misleading for anything else.

I am a non-native speaker and I was taught to write formally. Am I penalised for that?

By this tool, no. It names words and gives you the choice about each one. By commercial detectors, often yes. Formal register learned from textbooks overlaps heavily with the register models produce, and that overlap is a large part of why false positives fall unevenly.

Does my text get uploaded?

No. The scan runs in your browser and nothing is uploaded or stored.

Keep reading

Find the words.
Keep the sentence.

Free vocabulary check. No account. Runs in your browser.

Open the Full ReportSee the Whole Registry