Skip to main content
Inkprint

Guide

How to use Inkprint

Learn how to add text, check supported languages and quotes, read the measurements, and export your report. The guide also explains what a result can and cannot tell you.

Step 1

Add or upload your text

  1. Paste at least 100 words into the text box, or click Upload file to load a .txt, .md, or .docx file. Articles, essays, replies, and comments all work, and uploaded files are read in your browser.
  2. Check the word and character counts above the box. If a paste or import exceeds 50,000 characters, Inkprint trims it without an alert. Review the remaining text before analyzing.
  3. Click Analyze text. After a successful check, the result replaces the panel below and the input clears so you can run another check.
  4. Open More options to choose an example or select Clear everything to reset the text and report. Examples open saved reports immediately. Editing the text or original document requires a new analysis.

The run takes a few seconds. The button shows “Analyzing…” until the report is ready.

Check English and mostly English text

When you select Analyze text, Inkprint checks the language before scoring. This applies to pasted and imported text. English and mostly English writing can proceed, including English-dominant Taglish. Accepted mixed-language text carries a review note because its results need extra care.

Mostly non-English text, a balanced language mix, or an unclear language result stops the check before scoring. Inkprint explains the issue and keeps your text so you can review it. No new analysis report is produced.

The check considers the writing overall; a foreign name or occasional non-English phrase does not by itself make a text unsupported. Long documents are checked using excerpts spread across the text, so local language changes can be missed. The original document used for quote checking does not determine the submitted text’s language.

Step 2

Check quotes against the original document

Adding the original document is optional. It does one job: check the quoted sentences in your text against the document they came from.

  1. Paste your text, open More options, then select Add original document.
  2. Paste the document the text quotes from into the box. Inkprint trims excess text to fit 60,000 characters. Check that the quoted passages remain after trimming.
  3. Click Analyze text. The quote check panel lists one verdict per quoted sentence.

A worked example. This is the Quote-check sample, shown in full:

Text to check

The Vellmore Conservatory was "founded in 1912 in the old quarter", according to its own history page. Its main concert hall "seats more than 5,000 people in the main hall", which the page describes as the largest venue in the region. The annual spring gala "raises funds for scholarships", and the page says it has become the most important arts fundraiser in the city.

Visitors can join guided tours of the archive on weekends, and the conservatory continues to expand its educational programs for local schools. As a testament to its enduring cultural role, the institution reminds us that music education matters for every generation, and it invites the community to support its work.

Original document

The Vellmore Conservatory was founded in 1912 in the old quarter. Its concert hall seats 1,200 people. The annual spring gala raises funds for scholarships that support local students.

What the quote check reports

verified“founded in 1912 in the old quarter” matches the document word for word.
quote not found“seats more than 5,000 people in the main hall” does not appear in the document. The document says the hall seats 1,200 people.
needs review“raises funds for scholarships” matches, but the claim adds that the gala is the city’s most important fundraiser. The document does not say that, so the verdict sits below the confidence bar.

Inkprint matches only the words inside quotation marks. The claim around the quote decides whether the document supports it, contradicts it, or says nothing about it.

Try the sample in the workbench

The link opens the saved report with all three verdicts and fills in the text and original document so you can inspect or edit them.

Every quoted sentence gets one of four verdicts:

  • Verified: the document supports the claim.
  • Contradicted: the document says the opposite of the claim.
  • Quote not found: the quote does not appear in the document.
  • Unsupported: the document is silent on what the claim asserts.

Quotes shorter than 20 characters are skipped, and claims without quotes are not checked. The pattern score ignores the document entirely.

Step 3

Read the result

Panels run from the strongest evidence to the supporting detail. Read them in this order:

  1. Signal index: the score out of 100 and its band. The marker on the curved gauge shows the score’s position between few, mixed, and strong AI-writing signals. The index measures pattern strength, not the probability of AI authorship.
  2. Headline and summary: what stood out and which patterns drove it.
  3. Stats row: confidence in contributing graded checks, writing type, passages analyzed, and word count. A tentative type needs review. Yes/no checks have no separate confidence, and n/a means no graded checks contributed.
  4. Passage map: pattern strength in each sampled passage, so you can see where the signals cluster.
  5. Your text, marked up: click any highlight to see what matched. The legend names each mark.
  6. Signals: open a pattern to read what it checks and see the passage where it scored highest. Checks return a yes/no probability or a grade. The displayed percentages include context adjustments and do not measure the probability of AI authorship.
  7. Vocabulary: counts for the chatbot-linked words this check watches. They appear in human writing too, so treat them as weak evidence.
  8. Writing measurements: expand word-use counts, sentence lengths, and estimates of passive phrasing, adverbs, and hedges. Read the measurement guide before interpreting these numbers.
  9. Tool tokens and placeholders: exact matches such as leftover template text. These override the pattern score.
  10. Quote check: a verdict for each quoted claim when you add the original document. Chips mark quotes that need review.

A few habits make the report easier to trust:

  • Start with the exact matches and the strongest signals, then read the evidence. One signal is weak on its own. Several patterns can overlap, so review their evidence together.
  • Human writing reaches the mixed and strong bands too. Read the flagged passages before forming a view.
  • Check the passage map and the per-passage bars when the document is split. A single dense passage can raise the whole score.
  • A review note flags a score near a band boundary, an unclear writing type, or uncertainty in the patterns. Have a person review the evidence before acting on the result.

Read the writing measurements

Open View measurements in the report to inspect word use and sentence structure. These numbers describe style. A high or low value does not identify AI writing, and there is no target range to aim for.

Measurements cover prose across the full text retained after any input trimming, including passages outside the sampled analysis. They exclude code, Markdown headings, list lines, tables, and URLs. The prose word count can therefore differ from the report’s total word count. Adding an original document does not change these measurements.

Lexical diversity
The average share of unique words in overlapping 40-word windows. For example, 30 unique words in a window of 40 gives 75%. Higher values mean less repetition within those windows. Fewer than 40 prose words gives “Not enough prose”. Topic-specific vocabulary can lower this value without indicating AI use.
Sentence length and variation
Mean length is the average number of words per sentence. Variation divides the spread of lengths, or standard deviation, by that average. Lower values mean more uniform lengths; fewer than two sentences gives “Not enough sentences”. The length distribution groups sentences into five ranges and shows each range’s count and share of sentences.
Short sentence runs
A run contains at least two consecutive sentences of 1–4 words in the same paragraph. The report counts runs and gives the longest run and total short sentences. A short sentence can be grammatically complete; this is a length estimate, not a grammar check.
Function words
Common connecting words such as “the”, “and”, “of”, and “is”. Each entry shows its count and percentage of prose words. Only words on the checked list with at least one match appear; contracted forms are counted separately. These frequencies reflect habits and writing type, not a complete vocabulary profile.
Contractions and personal pronouns
Counts and percentages of prose words. Pronouns are grouped into first person (“I”, “we”), second person (“you”), and third person (“he”, “she”, “it”, “they”). Contracted pronouns can count in both groups, so these percentages do not add up to 100%. Distinguishable possessives do not count as contractions.
Possible adverbs
A count and percentage of prose words estimated from word endings and a small word list. The estimate can miss adverbs or count words used differently in context. It does not measure writing quality.
Possible passive phrasing
The count and percentage of sentences with passive-like verb patterns, such as “was written”. Some adjective phrases also match, and some passive constructions are missed. Read the sentence before treating a match as passive voice.
Stacked hedges
The count and percentage of sentences containing at least two possible hedge words or phrases, such as “might” and “perhaps”. The report also totals possible hedges. Cautious wording can be appropriate; these counts do not judge whether it is evasive.

These measurements add context to the findings. Sentence-length variation supports the existing rhythm check, and hedge counts supplement the existing hedge check. They do not add duplicate scored patterns. Compare texts of similar length and writing type, and read the passages before interpreting a difference. English-oriented estimates need extra care in mixed-language text.

Print or download a compact report

Select Print report after a check to print or save a PDF, or Download markdown for a .md file. Both formats include the score, up to eight strongest patterns, check summaries, writing measurements, and all review notes. They omit the submitted text, excerpts, and quoted claims to keep the report concise. Additional patterns remain available on screen.

Measurements include their units and limitations, sentence-length ranges, and counts for matching function words. Both exports use the measurements already in your report; exporting does not run another analysis.

Use A4 or Letter paper at the default scale, and check the print preview before saving. Reports can span several pages depending on their findings, measurements, and review notes. Browser print settings can change the page count. The Markdown file uses headings and tables for reading or sharing.

Limits

Limits to keep in mind

Inkprint reports on patterns in a text, and it has clear boundaries:

  • Minimum length is 100 words. Reports between 100 and 149 words carry a short-sample caveat.
  • The text box holds up to 50,000 characters, and the original document box holds up to 60,000.
  • Trimming prefers a paragraph break, then a sentence ending. If no suitable boundary is available, the cut can fall inside a sentence.
  • When you paste into an existing draft, only the new text is trimmed to fit. Text outside your selection stays in place.
  • Uploads accept .txt, .md, and .docx files up to 2 MB. Word formatting, images, and footnotes are not extracted, and .doc and .pdf files are not supported. Files larger than 2 MB are rejected.
  • For long text within the input limit, Inkprint scores up to four passages spread across the retained text. The report shows when coverage is partial.
  • Quote checks need quotes. Only quoted sentences are matched against your document.
  • Scores are signs, not proof. A result cannot tell you who wrote a text or why it was written.
  • Repeat runs vary. The same text can move a few points between checks.
  • Checks are rate limited per connection, and a repeat check of the same text returns the cached report for up to an hour.
  • Only English and mostly English text can proceed to scoring. Accepted mixed-language text needs extra care; see supported languages.

Fair use

Share results with care

Read the evidence, keep the context, and ask about process. Do not accuse anyone based on a score.

About Inkprint explains responsible use in full, including guidance for students who work with AI tools.

Read the teacher guide to fair writing review or the student guide to responsible AI use for ways to use the report in education. AI assistance can support learning when it follows the assignment’s rules.