How the method works

Before you upload your work you deserve to know what happens to it and how the numbers on your report are produced. This page explains the method in plain language. Nothing here is a secret — a defensible standard should be understandable.

Step 1 — You submit and pay

You create an account, enter your details, upload your document (Word, PDF or text) and pay the fee shown on the page (individual fee as displayed in the app; institutions with an agreement have their own fee, and students on an institutional bundle pay nothing). Your work is stored encrypted and is visible only to you and your reviewers — see Security.

Step 2 — What your work is compared against

• Published scholarship and books — a reference corpus of academic papers and books.
• Your institution's private repository — works previously approved at your own institution (so a “recycled” project is caught).
• The national comparison corpus — works approved at other institutions using the Service (cross-institution copying).
• The open web — distinctive passages from your text are searched live on the web so public sources are found. Only short excerpts are searched, never the whole document.

Two things are deliberately excluded: your own earlier submissions (a corrected resubmission is never flagged as a copy of itself) and any rejected or errored submission by anyone (rejected work is not repository material).

The reference list / bibliography at the end of your document is left out of the comparison, so citing your sources properly does not raise your score.

Step 3 — How the scores are calculated

Similarity score

Your text is broken into sentences and overlapping 4-word sequences (“shingles”). Each sentence is compared with every source:

• Verbatim match — the sentence shares long runs of identical word sequences with a source. Counted as copied words.
• Paraphrase match — the sentence shares most of its meaningful vocabulary with a source sentence (a term-frequency cosine comparison) even though the words are reordered or swapped. Counted as paraphrased words.

The overall similarity score is the percentage of the words in your document that fall in matched passages. Your report shows the verbatim and paraphrase parts separately, and highlights every matched passage side-by-side with the source it matched, so you and your reviewer can see exactly why.

AI-writing score

Separately, the engine estimates the likelihood that passages were machine-generated, using signals such as unusually uniform sentence length and rhythm, and a high density of stock phrases characteristic of AI prose. It is reported as a percentage with the reasons listed. By default the AI score is advisory: it changes the verdict only where your institution has chosen to set an AI threshold.

Step 4 — The bands

If your institution has set an AI threshold, the same three bands apply to the AI score, and the overall band is the worse of the two. Institutions can set their own thresholds; the one that applies to you is the one on your report.

Step 5 — A human decides

The engine never issues a certificate on its own. Where your institution has deployed the Service, its named reviewers (for example a College of Postgraduate Studies) see the report and make the decision under their regulations. For direct individual submissions the Intelliscan Standards Institute reviews. AMBER results are always read by a person; a reviewer can approve an AMBER work whose matches are legitimate quotations, or ask for revision.

Step 6 — Certificate and verification

Approved work receives a Certificate of Originality with a unique ID and a QR code. Anyone can confirm it is genuine at verify.intelliscan.africa/verify/<ID>, which shows the public record (name, matriculation number, institution, title, result, date) — never the document. Every step from payment to certificate is written to a tamper-evident, hash-chained audit trail.

If your work is flagged

Your report highlights each flagged passage next to the source. The optional AI Rewrite Coach can suggest an original re-wording of a passage you select (that passage is sent to Anthropic's AI only when you click). You then revise and submit again; your earlier attempt is never used against you.

Honest limits

• No system can see every source in the world. A GREEN result means no significant match was found in the sources available on that date, not that none exists anywhere.
• Similarity is not the same as plagiarism — properly quoted and cited material can match. That is why a person reviews borderline cases.
• AI-writing detection is probabilistic. It is presented as a signal for a reviewer, with reasons, and is advisory unless your institution decides otherwise.

Start a check →   Try the free 200-word demo

See also: Privacy · Terms · Data Retention & Deletion · Security