Is citing a review instead of the original research bad practice?

The question comes up because citing reviews is efficient and everyone does it, while the rule as usually stated makes it sound like a shortcut. The rule as usually stated is too blunt. What matters is whether the review is the appropriate source for the specific claim you are making.

Think about what a reader would do with the citation. If they follow it, do they land on something that supports your sentence? For "the effect has been replicated in three national samples", a review that surveys those samples is exactly the right destination. For "mentoring raised retention by eleven points", the reader needs the study that measured eleven points, and sending them to a review that mentions it is unhelpful and mildly misleading.

One workable split is by section. Penders recommends prioritizing reviews in introductions, where broad context is what the reader needs, and empirical papers in discussions, where the detailed contribution of each study is doing the work (Penders, 2018, p. 4). The same rules ask that a citation identify the exact contribution of the source rather than only place it in the bibliography (Penders, 2018, p. 2).

There is a second consideration about credit. Attributing a specific empirical finding to a review means the people who did the work do not get cited for it. Over a literature this has real effects on how contributions are recorded, and it is one of the reasons the practice is discouraged even when the claim is accurate.

There is also a risk in what the secondary source leaves out. Greenberg mapped a citation network around beta-amyloid claims in inclusion body myositis and found the selection was extreme. Of 32 citations to primary data on whether animal and cell-culture experiments were valid models, 31, or 97%, went to four highly supportive papers, and one, or 3%, went to six papers that weakened or rejected those models (Greenberg, 2009, p. 3). Eight key papers, seven of them from a single research group, carried 97% of the network's citation traffic (Greenberg, 2009, p. 5). A source that repeats the dominant interpretation without reporting the disagreement will show you a settled picture the primary literature does not support.

The same analysis shows how the distortion happens sentence by sentence. One primary paper found no beta-amyloid precursor protein or beta-amyloid in three of five patients and found it in only a few fibres in the other two, yet three later papers described those data as confirming a broader accumulation claim, and the resulting diversion generated 7,848 supportive citation paths (Greenberg, 2009, p. 3). Greenberg also describes citation transmutation, where a hypothesis was presented as established fact through citations to papers that stated only the hypothesis or did not address it at all (Greenberg, 2009, p. 6). Reading the original is what catches this.

The middle option is to cite both, and it is often the right one. "(Fenton, 2019, reviewed in Alvarez, 2022)" credits the study, points the reader at the wider context, and tells the truth about how you found it.

When is citing a review the right call?

When your claim is about the state of the literature rather than about a particular result. "The relationship has been examined across a range of settings" is a claim a review supports better than any single study. Reviews are also correct when you are citing the reviewer's own synthesis or argument.

Systematic reviews and meta-analyses deserve a note of their own. A pooled effect estimate is a result the review produced, not a summary of someone else's, so citing it to the review is correct and citing it to the included studies would be wrong.

That does not make the pooled estimate self-sufficient. Carlson and colleagues advise checking whether the review searched at least three bibliographic databases, looked beyond databases through methods such as hand searching, and included relevant unpublished studies (Carlson et al., 2023, p. 2). They also note that readers may need to inspect key included studies individually to judge the limitations and bias the pooled result inherits, including problems with missing data (Carlson et al., 2023, p. 3). If you rely on a meta-analysis for a load-bearing claim, that inspection is part of the citation, not an optional extra.

Reviews are also the right citation when you are explicitly signposting further reading. "For a fuller treatment, see" is a legitimate use and reads as generous rather than lazy.

How many citations does a broad claim need?

One good review, or two to four representative primary studies chosen to show the range rather than to prove volume. Stacking eight citations on an uncontroversial claim does not strengthen it. If the claim is contested, cite both sides rather than more of one.

Choosing representatives is a judgment worth making deliberately. Pick studies that differ in design, population, or era, so the citation group demonstrates that the finding is not an artifact of one approach. Three well-chosen citations do more work than eight from the same laboratory.

Where you have room, one sentence saying why those studies were chosen turns a citation list into an argument. "Established across observational and experimental designs" tells the reader something the parentheses cannot.