A Grok style analyzer without scores

This route submits a passage to a classifier and returns prose findings about how it reads. The governing instruction asks for a concise result with clear findings, restricted to the qualities implied by the tool name, and it forbids certainty about authorship.

What such a system establishes is a reading. A reading can be checked by returning to the text and looking for what was described, which is a real form of verification and the only one this route supports.

What it does not establish is a quantity. No arithmetic is performed on your text beyond the live character and word meter. The route is remote, bounded at 1 to 12,000 characters, and metered at 5 runs per rolling 24 hours.

The competing set divides cleanly by mechanism, and the division is worth understanding before choosing a tool for a given question.

Calculators

One analyzer computes Gunning Fog, Flesch-Kincaid, SMOG, Coleman-Liau and the automated readability index, plus syllable counts, a parts of speech breakdown and a punctuation frequency table, and plots Fry and Raygor graphs.

Published formulas

Another states its arithmetic openly: lexical density as content words divided by total words, lexical diversity as unique words divided by total words, each expressed as a percentage.

Classifiers

A third identifies over twenty-five named tones at sentence and document level. It discloses no word limit, no accuracy metric, no confidence value and no stated limitation anywhere on its page.

This route

A classifier that publishes its limits. Findings in prose, no label set, no index, and an explicit statement that a single run is not a measurement.

The selection rule

If the requirement is a number a standard specifies, use a calculator, because a figure produced without arithmetic is not the figure the standard means. If the requirement is what a reader will notice, a description carries more.

The procedure exists because the output requires verification and the allowance does not permit repetition.

Step 1

Submit a homogeneous sample of adequate length. Rhythm and recurrence are properties of a passage, so a paragraph is the practical minimum, and mixing sources yields findings about the seam.

Step 2

Constrain the analysis in the 300 character instruction field. Name an angle or an audience. A request for a numeric grade returns an estimate unsupported by computation and should not be made.

Step 3

Verify each finding against the source before acting. A described recurrence is a hypothesis; the document search that confirms it is the evidence. Discard anything you cannot locate.

A formula is deterministic by construction: the same input yields the same value on every execution, which is what makes an index citable in a standard. A generated analysis carries no such property. Two submissions of identical text may emphasise different observations or express the same observation in different terms.

This does not make the output unusable, and it does determine how it should be used. A finding is a direction to look, and its confirmation lives in the text rather than in a second run. Re-running to see whether the wording settles spends the allowance without adding evidence.

No competing model-backed page in this result set states this. The most prominent of them publishes no accuracy figure, no confidence value and no limitation of any kind, which leaves a reader to assume a stability the mechanism does not provide.

The instruction governing this route states that the model must never claim certainty about authorship, plagiarism, or which AI model wrote a text, and must describe observable signals and limitations instead. That is a property of the route, not an editorial preference expressed on this page.

The constraint is defensible on its own terms. Stylistic uniformity, stock connectives and unvaried sentence length are observable features of a passage. None of them identifies a writer, because each is produced routinely by humans writing under time pressure, to a template, or in a second language.

The practical effect is that findings are actionable rather than accusatory. A flat rhythm is worth correcting whoever produced it, and the correction does not depend on settling who did.

No, and the page declines the question rather than hedging it. The route is instructed not to, and the instruction reflects what the evidence supports.

A related product in this category matches a writer against famous authors. That is authorship inference presented as entertainment, and it is precisely the class of claim this route is built to avoid.

Where a detection opinion is genuinely required, that is a separate route on this site. It reports signals rather than proof, and it neither promises nor predicts the verdict of any third-party detector.

No readability index of any name. No lexical density or diversity figure. No syllable count, no parts of speech distribution, no punctuation frequency table, no word cloud and no readability plot.

No label set. There is no enumerated tone taxonomy, no percentage attached to any category and no sentence-level annotation, because the output is a single prose block.

No rewriting, no grammar correction, no file intake, no download, no batch queue and no programmatic access. The only computed values on the page are the live character and word counts, which update as you type.

The characteristic case is familiarity blindness. A writer has read a passage often enough that its regularities no longer register, and what is needed is an external reading that names one to look for.

Documentation teams apply it with an audience constraint to test consistency against an existing set. Academic writers apply it to locate the weakest section before review. Second-language writers apply it where a description of how a passage lands is worth more than any index, because the question is fit rather than complexity.

The misapplication is treating the output as a metric, whether by citing it, tracking it across drafts, or re-running it for stability. None of those is supported by how the result is produced, and the page states so rather than allowing the inference.