Text quality analysis
Check waffle, keyword density, readability and clichés online and free — right in the browser, with no sign-up and without sending the text to a server. Tidy the text up before you publish it.
What the measures mean and which values count as normal
The tool does not grade the text — it shows five measurable signs of where the text reads heavily. Below is where each number comes from, which boundary counts as normal and what to do if a value falls outside it. The measures apply to text from 50 words up: on a shorter fragment the frequencies are too noisy, and the tool says so honestly.
Waffle
The share of stop words — prepositions, conjunctions, particles, fillers, intensifiers and empty words — among all the words in the text. The language needs these words, but when there are too many, a sentence carries less meaning than it takes up space.
Normal is up to 15%. From 15 to 30% is worth re-reading; above 30% the text is diluted.
Cut the intensifiers («very», «really»), the filler phrases and the connectives that add nothing. The breakdown by group under the measure shows which category the text has most of.
Keyword density
How often the same words repeat. Academic density is the total share of repeats by word stem, stop words excluded; classic density is the square root of the most frequent word's count. High density usually means a poor vocabulary rather than keyword stuffing.
Normal for academic is up to 8%, worrying past 12%. For classic, normal is up to 3.
Replace some repeats with synonyms or pronouns, and delete the rest: usually half the repetitions sit in phrases that can be cut entirely.
Readability
The Flesch index in Oborneva's adaptation for Russian: calculated from the average sentence length and the average word length in syllables. The higher the value, the easier the text reads.
Normal is from 60 up. Values from 30 to 60 need attention; below 30 the text is heavy for a wide audience.
A low index is not cured by «simplifying the thought» but by shorter sentences and by replacing long verbal nouns with verbs: «carry out a check» → «check».
Sentence length
The average number of words per sentence. Separately, the tool highlights sentences longer than 30 words — those are where a reader most often loses the thread.
Normal is up to 20 words on average; past 25 the text takes effort to read.
Split the highlighted sentences at clause boundaries. An even rhythm matters more than the average: a text of uniformly short phrases also reads badly.
Clichés
Recognisable empty constructions — «in today's world», «it is important to note», «plays an important role». The same list holds the turns of phrase typical of text generated by a model and never edited.
Normal is zero. One or two occurrences are tolerable; more than two are noticeable to the reader.
A cliché can almost always be deleted without losing meaning: cross the phrase out and re-read the sentence. If the meaning did not change, it was surplus.
Frequently asked about text analysis
Does the text go to a server?
No. The whole analysis runs in the browser: the text is not sent to our servers, not stored and never reaches the logs. You can close the page — nothing will be left.
Why do the values differ from other services?
Stop-word lists and formulas differ between services, so no two tools will agree digit for digit. Under each measure there is a «How it is calculated» block with the exact formula — it shows precisely what is being measured.
Why are the measures hidden on a short text?
Below 50 words the frequency measures turn to noise: one repeated word pushes density into the red, and one long sentence does the same to readability. Showing such numbers as a result would be a lie, so the tool warns that the text is too short.
Do I have to get every measure into the green?
No. This is diagnostics, not a grade: a legal document will score low on readability by the nature of the genre, and a short news item will show high density because the topic repeats. Look at the measures that missed the norm by a lot, and at the highlighting in the text.
Does this help check a text written by a model?
Partly. Clichés and repeats are typical traces of generation without editing, and they show up in the «Clichés» and «Keyword density» measures. But the measures judge the quality of the text, not its origin: an edited AI text will look like a good human one, and that is fine.
Check quality on a stream, not one text at a time
In the workspace the same measures are calculated for every piece in the content plan — before publishing and without copying anything into a form by hand.