Reviewed as of 27 September 2026.
A citation may appear impeccable, complete with author, title, year and a phrase in quotation marks, yet not originate from the cited book. This is a frequent error in academic work and cultural journalism, and search engines and artificial intelligence assistants make it easier when they return phrases without saying where they come from. This guide proposes the method I apply to work out what kind of source I am looking at before I build an argument on it.
Four levels
Not all sources serve to prove the same thing. I classify them into four levels, from closest to furthest from the content you want to attribute.
Level 1: the consulted text. You have read the passage in the book, article, or document and can locate it. If you attribute a literal phrase to an author, this is the necessary level. Whenever the edition permits, note the page. In texts without stable pagination, retain the chapter, section, paragraph, locator, or version.
Level 2: the institutional, regulatory, or archival source. This may be an archive inventory, a museum catalogue, an exhibition record, an official resolution, or research published by an institution. It does not replace an author's text, but it may be a primary source for the fact it documents: that a work entered a collection, was exhibited on a specific date, or was the subject of a decision.
Level 3: the bibliographic record. This is the record from a library, repository, or publisher containing edition details: author, title, publisher, date, ISBN, extent, or version. It serves to identify a publication. By itself, it does not allow one to affirm what the text maintains.
Level 4: the promotional or commercial text. This includes the back cover, the jacket flap, the publisher’s synopsis, the bookseller's description, and other materials conceived to present or sell the work. It may guide a search, but it should not be attributed to the author unless it is confirmed that they wrote it.
The hierarchy is not absolute. An official resolution is level 2 regarding what it resolved, but it may be the primary source for the administrative act. A back cover may be the object of study and thus become the document being analysed. The level depends on the assertion one wishes to support.
The most frequent trap
The typical mistake is to cite a level 4 as if it were a level 1. An effective phrase from a back cover circulates associated with the author's name; someone cites it and others repeat the attribution. Over time, it appears to form part of the book, even though no one has located the passage.
It happened to me with a phrase about net art that several sources attributed to a specific book. It fitted exactly what I wanted to argue. I consulted the full text and the phrase did not appear. It originated from the editorial description.
The alarm of the perfect fit
From that case I drew a rule: if a quotation fits suspiciously well with what you want to prove, check it more carefully. Rounded formulations that condense a thesis into a single line are frequent in promotional materials. The actual reasoning of a book is usually more extensive, conditioned and difficult to isolate without losing nuances.
This does not make a clear sentence suspicious in itself. It only means raising the level of checking when the finding appears tailored to the argument.
Verifying the link and the chain of attribution
A source may claim to be supported by an institution and link to another page. It may also cite a news item that refers to a press release, or a review that attributes a phrase to a book without indicating a page. Open every link and verify:
- who actually publishes the content;
- which document contains the data;
- whether the page supports the specific assertion;
- whether an earlier source exists that is closer to the fact;
- which date, edition or version you are consulting.
Do not cite the authority mentioned by a page as if you had consulted its document directly. If you cannot access the original source, make the intermediation clear.
Reading in a repository is acceptable
Reading a book or article in a repository and citing the located passage is standard research, provided the file corresponds to the edition or version you identify. What is relevant is having consulted the text. What is not acceptable is treating the record or the synopsis of the repository as if they were the content of the work.
In publications with several versions, note which one you have read. In digital documents that are subject to updates, save the date of consultation and, whenever possible, a copy or persistent link.
Not having found does not equate to non-existence
"I have not located any document that proves it" does not mean "it did not occur". A search may be incomplete, archives may not be digitised and terminology may have changed. If you do not find the data, qualify the assertion: "not located in the sources consulted" or "I have not found documentary evidence".
Absence can only function as evidence when there is an independent reason to expect the data to appear in a sufficiently complete record. Even then, the scope of the archive and the search must be explained.
When artificial intelligence is involved
An assistant may conflate the four levels, fabricate a page, or reproduce an erroneous attribution circulating online. It may assist in locating candidates, search terms, or documentary paths, but it does not replace verification.
If you use it:
- request the exact source and the passage that supposedly supports each claim;
- demand that it distinguishes between original text, metadata, institutional source, and promotional material;
- open the document and locate the passage yourself;
- verify the author, edition, version, date, and page;
- record what could not be verified and do not present it as fact.
Before submission
- Has every literal quotation been verified against the text and can it be located?
- Does the edition or version cited match the one consulted?
- Does every factual datum originate from a source adequate to prove it?
- Does any attribution actually originate from a back cover, synopsis, or review?
- Do the links lead to the mentioned document and contain the data?
- Have you distinguished between consulting a source and knowing it through an intermediary?
- Have you written 'not located' when the search does not allow for the assertion that something does not exist?
On the open conversation
This guide outlines the source classification method I apply in my research and in this notebook. It is a practical tool, not a universal taxonomy. If you teach methodology, work in a library or an archive, and can improve it, the guide remains open.
Sources
Proprietary method. No external sources were used to formulate the classification.