Academic writing · Abstracts
Abstract examples across eight academic disciplines
An abstract is not a miniature paper. It is a screening device, and each discipline has decided differently about what a reader must know before opening the full text.
In this article
Every abstract answers the same editorial question — what must the reader know now, and what can wait for the full text — and the eight forms below answer it differently. The differences are not stylistic. They follow from who is screening, what counts as a finding, and whether the venue indexes the abstract as metadata or reads it as an argument.
Each example is annotated by the job each sentence does, because that is the part that transfers between fields. A structured biomedical abstract and a humanities abstract look nothing alike, but both fail in the same way: a final sentence that claims more than the work behind it supports.
1. Biomedical, structured
Biomedical journals ask for labelled sections — Background, Methods, Results, Conclusions — because the abstract is an indexing layer as much as a preview. UCI's guide to writing an abstract calls this the informative abstract: self-contained, adding no new information, presenting purpose, results, and conclusions in that order.
Word limits vary more than most writers expect, and the disagreement between sources is itself worth knowing. That same UCI guide puts the informative abstract at 250 words or less and reproduces a checklist from Mack's How to Write a Good Scientific Paper that specifies 200. Hamilton's APA guidance allows no more than 120. There is no portable number here — read the venue's instructions rather than a remembered limit.
The conclusion sentence is where these abstracts usually break. A reviewer reading “pharmacist review reduces post-discharge errors” from a retrospective cohort is reading a causal claim the design cannot carry. Naming the unmeasured outcome costs one clause and removes the objection.
2. Engineering and computer science, informative
Engineering abstracts read like compressed specifications: problem, approach, validation, result. The sentence that most often goes missing is the comparison point. An abstract that reports an improvement without naming the baseline, dataset, or deployment setting has hidden the contribution rather than stated it.
Problem. Tail latency in geo-distributed key-value stores is dominated by cross-region coordination, and existing quorum-tuning approaches assume stable inter-region round-trip times.
Approach. We present Driftlock, a quorum selection layer that re-ranks replicas from a rolling estimate of inter-region latency rather than a static topology map.
Validation. We evaluate Driftlock against unmodified Cassandra 4.1 and a static-quorum baseline on the YCSB workload suite across five simulated regions.
Result. Driftlock reduced 99th-percentile read latency by 31% relative to the static baseline in this test configuration, with no measurable change in throughput.
“Improved performance” is soft. “Reduced 99th-percentile read latency by 31% relative to the static baseline” is checkable, which is the only kind of claim worth putting in an abstract. If a sentence cannot be tied to a method, a dataset, or a result in the paper, it belongs in the introduction.
3. Humanities and social sciences, indicative
A humanities abstract often has no result in the scientific sense. It has a thesis, a source base, and a position in a critical conversation. The common failure is plot summary: reporting what a text contains without naming the intervention the paper makes.
Claim. This article argues that the serialised novel's cliffhanger was shaped less by narrative convention than by the billing cycles of nineteenth-century circulating libraries.
Source base. It draws on subscription ledgers from three provincial circulating libraries, 1838–1861, alongside publishers' correspondence held in the same collections.
Position. Against accounts that treat serial suspense as a purely formal development, it situates the device within the material economics of lending, extending recent work in book history on distribution as a shaping force.
Stake. Reading the cliffhanger as a retention mechanism rather than an aesthetic choice reframes a familiar feature of Victorian fiction as evidence of how readers actually paid for what they read.
Three things should be identifiable: the object of analysis, the interpretive lens, and the claim. A draft missing any of them reads as descriptive rather than analytical, which is what “the author discusses several themes” signals to an editor screening submissions.
4. Chemistry and materials science, graphical
A graphical abstract moves the work from sentences to composition, and its constraint is reproduction size. Dense labels, small arrows, and full mechanisms fail because the reader decodes the image at thumbnail scale, often in a list of search results. The figure should answer one question: what changed.
What the image carries. A single left-to-right transformation: starting material, one arrow labelled with the catalyst, product. No intermediate steps, no inset mechanism.
What the image omits. The catalytic cycle, the substrate scope table, and all yields except the headline figure. Those belong in the paper.
Caption. “A cobalt–phosphine catalyst converts the aryl halide to the coupled product at room temperature (82% isolated yield); the full substrate scope and mechanism are reported in the text.”
Reproduction check. Legible at 300 px wide, and in grayscale — no distinction in the figure depends on hue alone.
The trade-off is honest compression. A graphical abstract can oversimplify when the real contribution is conditional or comparative. In that case the figure should carry the headline result and the caption should carry the condition, rather than the image implying a generality the paper does not claim.
5. Public health and epidemiology, structured with implications
Public health abstracts add a layer: who can act on this. An implications sentence with no named actor — clinicians, a school system, a public agency — leaves the policy relevance for the reader to infer. This genre also carries a harder honesty requirement, because the language sits close to advocacy.
Background. Seasonal influenza vaccination coverage among school-age children in the region has remained below the national target since 2019.
Methods. We analysed coverage records for 38 schools across two districts for the 2023–24 and 2024–25 seasons, comparing schools with on-site vaccination clinics to those referring families to community providers.
Results. Coverage was 61.4% in schools with on-site clinics and 47.9% in referral-only schools. The difference persisted after adjustment for school-level deprivation index.
Conclusions. On-site delivery was associated with higher coverage in this sample.
Implications. District health officers considering clinic placement may find on-site delivery worth piloting. These are observational data from two districts; schools selecting on-site clinics may differ in ways not captured by the deprivation adjustment.
The test for the final sentence is whether it survives without exaggeration. If the implication reaches past the data, the abstract is doing editorial work the study has not earned — the failure mode that makes public-facing summaries of research untrustworthy. For recording those constraints systematically while drafting, see extracting study limitations from research papers.
6. Physics and mathematics, conference
Conference abstracts are read for scheduling as much as for content, and the audience already knows the field. The contribution belongs in the first sentence, not after a paragraph of orientation. Notation consistency matters more than prose quality here: a misread symbol obscures a claim faster than an awkward sentence does.
Contribution. We prove that every locally finite graph with bounded degree admits a spanning subgraph of girth at least g retaining a constant fraction of the original edge density, for every fixed g ≥ 5.
Method. The proof combines an entropy-compression argument with a local recolouring step, avoiding the probabilistic deletion method used in earlier partial results.
Position. This settles the bounded-degree case of a question raised in that earlier work; the unbounded-degree case remains open.
These abstracts also carry a publication risk. A conference abstract may remain the only public version of a result for some time, sitting beside a later full paper. It should stand alone without presenting a preliminary result as settled. Citing one has its own conventions, covered in how to cite a conference paper.
7. Computer science, extended abstract
Extended abstracts run two to four pages and may carry pseudocode, a system diagram, or preliminary results. The extra space should change the evidence, not the structure — the failure mode is using the additional pages to widen the preamble until the text is neither concise nor complete.
Page 1. Contribution in the first paragraph, then the problem setting and the design in enough detail to judge novelty. No literature survey.
Page 2. One system diagram and a compact result block — two or three measurements against a named baseline, with the experimental setup stated.
Pages 3–4. Trade-offs, the cases where the approach does not apply, and what the full evaluation will add. References in the venue's required style.
Status line. “These are preliminary results from a partial implementation; the full evaluation is in progress.”
Layout carries real weight at this length. Dense type with no visual anchors is hard to read on screen, and figures that depend on colour alone fail in print. Reference formatting should match the venue from the start rather than being fixed at submission; for ACM venues, see the ACM reference format.
8. Plain-language summary
Plain-language summaries answer a different reader — a policymaker, a patient, a journalist, a funding panel. Agencies and open-access publishers increasingly require them. The discipline here is translating without flattening: everyday phrasing is fine, but uncertainty has to survive the translation.
The problem. People recovering from heart surgery often go home with a long list of medicines, and mistakes with those medicines are common in the first few weeks.
What we looked at. We reviewed the records of 412 people discharged from one hospital to see whether having a pharmacist check the list before they left made a difference.
What we found. Mistakes were less common when a pharmacist had checked the list. Because we looked back at existing records rather than running a trial, we cannot say the check itself caused the difference.
Why it matters. The result is a reason to test pharmacist checks properly in a trial. It is not yet a reason to change how hospitals discharge patients.
The useful test is external. If someone outside the field cannot explain the summary back in their own words, it is still too technical. If they can explain it back but overstate the finding, the uncertainty did not survive the rewrite — which is the more common and more serious failure.
The verification pass
Across all eight forms the transferable discipline is control: identify the reader, name the problem early, state the method or source base, report the finding at the precision the evidence supports, and stop. An abstract fails when it tries to do the work of the introduction, the discussion, and the conclusion at once.
| Form | What the first sentence must do | Most common failure |
|---|---|---|
| Biomedical, structured | Name the clinical problem and the uncertainty | Causal conclusion from a non-causal design |
| Engineering, informative | Name the technical gap | Improvement claimed with no baseline |
| Humanities, indicative | State an arguable thesis | Plot summary instead of intervention |
| Graphical | Show the single transformation | Illegible at reproduction size |
| Public health | Name the population problem | Implication with no named actor |
| Conference, physics/maths | State the result or theorem | Contribution buried after context |
| Extended abstract | State the contribution in paragraph one | Preliminary results presented as settled |
| Plain-language | Name the problem in everyday terms | Uncertainty lost in translation |
A final pass should check the word limit, the required headings, terminology, scope, stated limitations, and the venue's own rules. That pass earns its time because the abstract is the only section most readers, editors, and indexing systems see. The same claim-by-claim method used on a full manuscript applies here at smaller scale — the method is in the submission audit guide, and the broader workflow it belongs to is in writing a research paper you can defend.