Does the annotation itself change between APA, MLA and Chicago?

No. All three leave the summary prose to you and to your department. Only the reference line above it changes, in author name format, title capitalisation, date position and the punctuation of volume and page numbers. Indentation conventions differ slightly.

Should the annotation summarise, evaluate, or both?

Both, unless told otherwise. A working entry does three things in order: what the source found, one methodological caveat or scope note, and where you use it in your own argument. The last sentence is what shows your supervisor you actually read it.

What do I do when the article does not print its volume or issue number?

Take those values from the DOI record or the publisher page, and never guess them. Walters and Wilder found that incorrect volume, issue, page and year values were the most common error type even in references to genuine works.

← All articles

Citing & styles2026-09-05· 6 min read

Annotated Bibliography Examples: One Real Source in Three Styles

Three fully written annotated bibliography entries for the same real article in APA 7, MLA 9 and Chicago 17, plus copy-ready templates and the fields you should never guess.

An annotated bibliography entry is two things stacked. First a normal reference in your citation style, then a short paragraph saying what the source actually contains and why it belongs in your paper. Below are three complete entries for the same real article, written out in APA 7, MLA 9 and Chicago 17, so you can see which parts change between styles and which stay put. If you need the ground rules rather than the samples, they are in How to Write an Annotated Bibliography in APA, MLA and Chicago. This page is the worked examples.

Every source below is real, and the page numbers in the annotations are printed page numbers, not approximations. We found them through cytado's own source corpus, which is built on Crossref, OpenAlex and DOAJ records.

One source, three styles

The article: a 2023 Scientific Reports study in which Walters and Wilder ran GPT-3.5 and GPT-4 over 42 topics, produced 84 short literature reviews, and then checked all 636 references in them against real databases.

APA 7

Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the
bibliographic citations generated by ChatGPT. Scientific Reports, 13,
14045. https://doi.org/10.1038/s41598-023-41032-5

     Walters and Wilder checked 636 bibliographic references taken from
     84 machine-generated literature reviews across 42 topics. Of the
     GPT-3.5 references, 55% were fabricated; for GPT-4 the figure fell
     to 18% (p. 1). Among references to works that did exist, 43%
     (GPT-3.5) and 24% (GPT-4) still carried substantive errors, and
     incorrect volume, issue, page and year values were the most common
     type (pp. 1, 5). Pooling six studies, the authors report 51% of 732
     citations fabricated (p. 2). I use this study to separate two
     failure modes my chapter keeps blurring together: the source that
     does not exist at all, and the source that exists but is described
     wrongly.

MLA 9

Walters, William H., and Esther Isabelle Wilder. "Fabrication and Errors
     in the Bibliographic Citations Generated by ChatGPT." Scientific
     Reports, vol. 13, 2023, article 14045, https://doi.org/10.1038/
     s41598-023-41032-5.

     Walters and Wilder checked 636 bibliographic references taken from
     84 machine-generated literature reviews across 42 topics. Fifty-five
     per cent of the GPT-3.5 references were fabricated, against 18% for
     GPT-4 (p. 1). Even among references to real works, 43% and 24%
     respectively contained substantive errors, most often in the numeric
     fields (pp. 1, 5). The study anchors my claim that a citation list
     can look conventional and still be unverifiable.

Chicago 17 (notes-bibliography)

Walters, William H., and Esther Isabelle Wilder. "Fabrication and Errors
     in the Bibliographic Citations Generated by ChatGPT." Scientific
     Reports 13 (2023): 14045. https://doi.org/10.1038/s41598-023-41032-5.

     Walters and Wilder checked 636 bibliographic references drawn from
     84 machine-generated literature reviews across 42 topics, finding
     55% of GPT-3.5 references and 18% of GPT-4 references fabricated
     (p. 1). Errors in the surviving, genuine references clustered in
     volume, issue, page and year (p. 5). Useful here mainly as evidence
     that manual verification of a reference list is not optional.

What changes between the styles, and what doesn't

Look at the three blocks again. The reference line changes in four predictable places: how the authors' names are written, whether the title is in title case with quotation marks or sentence case without, where the year sits, and how the journal's volume and article number are punctuated. That is the whole difference at entry level.

What does not change is the annotation itself. No style tells you to write a different summary; APA, MLA and Chicago all leave the prose to you and to your department. The indentation convention differs slightly (APA and Chicago sit the annotation as a fresh indented block under the reference, MLA continues the hanging indent), but the sentences are yours in all three.

Length is set by your instructor, not by the style manual. If you were given no figure, 100 to 200 words per entry is the usual working range, and consistency across entries matters more than hitting an exact count.

Before and after: an annotation that earns its place

Before

This article is about citations generated by ChatGPT. The authors discuss the problem of fabricated references and provide useful insights into this important topic. It is a valuable source for my thesis.

After

Walters and Wilder found 55% of GPT-3.5 references and 18% of GPT-4 references to be fabricated across 636 checked citations (p. 1), with numeric fields the leading error type in the genuine ones (p. 5). I cite it in section 2.3 as the quantitative basis for treating every AI-suggested reference as unverified until checked.

Three moves separate them. Replace "the authors discuss X" with what the authors actually found, in numbers where numbers exist. Attach a page number to any specific claim, so a reader can land on it. Finish with the entry's job in your argument rather than a compliment to the article. A supervisor scanning your list is checking whether you read the source, and the after version answers that in one sentence.

The fields you should never guess

Here is a second entry, and a caveat worth copying along with it:

Subramanian, P., Soundararajan, R., & Makol, J. (2025). Building
credibility with comprehensive citation practices. Indian Journal of
Radiology and Imaging, [volume]([issue]), 97-101.
https://doi.org/10.1055/s-0044-1796641

     A practical overview of citation and reference conventions in
     scientific publishing, covering style selection, reference managers
     such as Zotero, Mendeley, EndNote and RefWorks (p. 97), and the
     common failure modes: incorrect author names, titles and publication
     details that undermine a paper's credibility (p. 99). The authors
     note that citation-related problems contribute to 10-15% of
     manuscript rejections (p. 100). I use it for the argument that
     reference hygiene is an editorial gate, not a formatting nicety.

The volume and issue are bracketed because the copy I read does not print them, and inventing a plausible number is exactly the error Walters and Wilder measured. Take those two values from the article's own front page or from its DOI record before you submit. The same discipline applies to online material without a DOI: Klein and colleagues found that referenced web resources frequently cannot be revisited later, and that web archives often hold no snapshot close enough to the citing date to be representative (p. 26). Give a permanent link where one exists, and add the access date if your style asks for it.

If you would rather not assemble the reference line by hand for every entry, cytado's bibliography generator builds it from a DOI, ISBN or title and fills the numeric fields from the source record, and the citation generator does the same for a single in-text citation. You still write the annotation. That part cannot be outsourced, and it is the part being graded.

Copy-ready skeletons

APA 7
Surname, A. A., & Surname, B. B. (Year). Title of the article in sentence
case. Journal Name, Volume(Issue), pages. https://doi.org/...

     [1-2 sentences: what the study did and found, with a page number.]
     [1 sentence: method, scope or limitation a reader should know.]
     [1 sentence: where this source is used in your paper and why.]

MLA 9
Surname, First, and First Surname. "Title of the Article in Title Case."
     Journal Name, vol. X, no. Y, Year, pp. 00-00, https://doi.org/...

     [Same three moves. Hanging indent continues into the annotation.]

Chicago 17
Surname, First, and First Surname. "Title of the Article in Title Case."
     Journal Name Volume, no. Issue (Year): 00-00. https://doi.org/...

     [Same three moves.]

Fill the slots from the source in front of you, never from memory, and keep the three annotation moves in the same order across every entry. A list that is consistent reads as deliberate, and it is far quicker to check.

Frequently asked

How long should each annotation be?: Your instructor sets the length, not the style manual. If you were given no figure, 100 to 200 words per entry is the usual working range. Consistency across entries matters more than hitting an exact count.
Does the annotation itself change between APA, MLA and Chicago?: No. All three leave the summary prose to you and to your department. Only the reference line above it changes, in author name format, title capitalisation, date position and the punctuation of volume and page numbers. Indentation conventions differ slightly.
Should the annotation summarise, evaluate, or both?: Both, unless told otherwise. A working entry does three things in order: what the source found, one methodological caveat or scope note, and where you use it in your own argument. The last sentence is what shows your supervisor you actually read it.
What do I do when the article does not print its volume or issue number?: Take those values from the DOI record or the publisher page, and never guess them. Walters and Wilder found that incorrect volume, issue, page and year values were the most common error type even in references to genuine works.