Screening methodology
Every manuscript on ASNAP is screened by the same method, whether the report is read by its author, an editorial office or an institution. This page describes what the report measures, how its figures are produced, where the method is known to fail, and what it deliberately does not do.
Overlap with sources
The text is split into overlapping word sequences, which are compared with scholarly records (Crossref and OpenAlex metadata, open-access full texts, arXiv) and with web pages retrieved for the candidate passages. Each matched passage is marked and linked to its source, with the share of the text that source covers.
- Overlap is the share of words inside matched passages, after sources covering less than 0.5 % of the text are set aside; the number of such minor sources is stated.
- Quotations in quotation marks or with a citation marker are reported separately when they are recognised; recognition is heuristic and misses some citation styles.
- The reference list is excluded from overlap when it is detected, so matching entries do not inflate the figure.
Overlap is not a measure of misconduct. Standard phrases, methods descriptions, definitions and correctly cited quotations all produce it. The report exists so that each passage can be reviewed against its source.
Indicators of machine-generated text
Each unit of running prose receives a score between 0 and 1 from a classifier trained to distinguish text written by people from text produced by large language models. Units above the threshold stated in the report are highlighted; the document figure is the share of evaluated text above that threshold, not a probability that the document was generated.
Headings, tables, formulae, references and short fragments are not evaluated. Texts with less than about 3,000 characters of running prose receive no indicators. Figures are shown only for languages for which the detector has been validated, and the report states the detector version.
The method has known limits: writing by authors whose first language is not English, formulaic prose and heavily edited text receive more false signals; generated text paraphrased by a person receives fewer. For this reason the report never states that a text is machine-generated. It lists passages so that an editor can ask about them rather than draw a conclusion from a number.
Reference check
Entries of the reference list are parsed for DOIs, titles and identifiers and looked up in Crossref. Each entry receives one of the statuses below; a retraction notice carried by Crossref is shown with a link.
| Found | A Crossref record matches the DOI, or the title matches a record closely. The matched title is shown for comparison. |
|---|---|
| Not found | The DOI is not registered, or no record matched. The entry may contain an error or describe a work that does not exist. |
| DOI resolves to another work | The DOI exists, but the record it points to does not match the title and authors in the entry. |
| Registered outside Crossref | The DOI is valid but registered with another agency (DataCite, mEDRA and others), so the title was not compared. |
| Not checked | The entry could not be parsed, or the time allowed for the check ran out. |
“Not found” is a prompt to verify the entry by hand, not a statement that the work is fabricated.
Compliance with author guidelines
For a manuscript submitted to a journal on ASNAP, the report also lists how the text meets the journal’s stated requirements: structure, length, abstract and keywords, reference style and required statements, including the disclosure of AI-assisted tools where the journal asks for one. Requirements are taken from the journal’s own author guidelines.
What the screening does not do
- It does not issue verdicts such as “AI-generated: yes or no”, accuracy percentages or guarantees.
- It does not rewrite text or help a text evade detection.
- It does not add manuscripts to any comparison database and does not use them to train models.
- It does not search publishers’ closed full-text collections; coverage is limited to open records and the web.
- It does not replace peer review or the judgement of the editorial office.
Contesting a result
If a correctly quoted passage is marked as overlap, an existing reference is marked as not found, or indicators appear on text written without any tool, write to [email protected] with the report and the passage in question. We review such cases and use them to adjust the method. Editors are asked to treat a report as a starting point for a conversation with the author, in line with COPE guidance on handling suspected misconduct.
Last updated 18 October 2026.