Does a detector count citations, the bibliography and headings?
Not all of it. Before scoring, a full check in DetekceGPT leaves out the parts the author does not write in their own words or that follow a prescribed form: the bibliography, in-text references to sources, direct quotations with a stated source, footnotes, the table of contents, figure and table captions, title page details, the declaration of authorship, acknowledgements, the list of abbreviations and an abstract in another language. Headings stay in the text, but the detector never marks them as machine-written sentences. The demo without signing in leaves out none of this and scores the pasted passage in full.
What a full check leaves out
An academic paper contains long passages the author does not write in their own words or that have a fixed form. Before scoring, a full check finds them and removes them from the input for all seven detectors: the bibliography, in-text references such as (Smith, 2021) or [3], direct quotations in quotation marks followed by a reference to the source, footnotes, the table of contents with page numbers, figure and table captions, "Source:" lines, table rows and hypothesis statements such as "H1:".
The formal parts of a paper go the same way: title page details such as the supervisor, the faculty or the degree programme, the declaration of authorship, acknowledgements, the list of abbreviations and an abstract in a language other than that of the main text. Text converted from a PDF also loses the headers and footers repeated on every page. In e-mails and letters, the greeting, the polite closing lines, the signature and the quoted previous message are left out as well.
Why it does this
A bibliography formatted to a citation style is regular and predictable — exactly what a detector looks for in machine text. Left in the input, it would push up the score of every paper with honestly kept sources. A direct quotation, in turn, is not the author's text, so the detector would be judging someone else's sentence and attributing it to the person who merely quoted it.
The formal parts are also much alike across schools and years. The declaration is copied from a template, acknowledgements use set phrases and the title page is a form. With them left out, only what the author actually wrote is scored, and the number does not move because of a template.
What leaving things out does not solve
What gets left out is decided by the form of the text, not its meaning. A quotation is recognised by its quotation marks and source reference, a bibliography by the section heading and typical details such as a year, a DOI or "available from". A paraphrase of someone else's text without quotation marks stays in the input as the author's text, and a source written down in an unusual way may not be recognised — the result is then shaped by a part that is not the author's own.
The demo on the website does not leave anything out: it checks the pasted passage of up to 3,000 characters in full, with one detector. If you try it on a chapter, paste continuous text without the bibliography and without long quotations. In a full check, headings and short lines without a full stop are always excluded from marking, so you will never see them flagged as machine-written in the result.
FAQ
Do I have to delete the bibliography before a check?
Not in a full check; it is left out automatically. In the demo without signing in, yes — everything you paste is scored there.
Why did the detector judge a quotation as my text?
A direct quotation is only left out when it is in quotation marks and immediately followed by a reference to the source. A quotation without quotation marks or without a reference is judged as ordinary text.
Does the check recognise a table of contents and page numbers from Word?
Yes. Table of contents lines that end in dots or a tab and a page number are left out, as are lines made up almost entirely of numbers.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection