How to Write a Research Abstract

Q: How long should a research abstract be?

There is no single universal limit — it’s set by the target journal or conference, commonly in a 150-350 word range, with 250 and 300 words being frequently used caps. Always check the specific venue’s current author instructions rather than assuming a round number.

Q: Can I cite references in a research abstract?

No, as a general rule — most journals prohibit citations in the abstract, since it’s indexed and read independently of the reference list. The recognized exception is a required trial or systematic-review registration number (e.g., a ClinicalTrials.gov or PROSPERO identifier), which functions as a locator rather than a citation.

Q: Do all journals require a structured abstract?

No. ICMJE recommends structured abstracts for original research, systematic reviews, and meta-analyses, and most clinical/biomedical journals follow that pattern, but plenty of journals — PLOS ONE among them — require an unstructured single-paragraph abstract instead. Always follow the specific target journal’s instructions for authors.

Q: What’s the difference between an abstract and an introduction?

The introduction sets up the full paper for a reader who is about to read the rest of it, and can take as much space as needed to build context and cite prior work. The abstract is a free-standing summary of the entire paper — problem, methods, results, and conclusion — written to be read on its own, under a strict word limit, without citations.

Q: Should I write the abstract before or after the rest of the paper?

Write it last, or at minimum revise it after the paper is finished. An abstract drafted before results are final routinely ends up describing a study slightly different from the one that was actually completed — a common, avoidable source of abstract/full-text inconsistency.

An abstract is not a preview of your paper — it is a self-contained summary that has to work whether or not anyone reads another word of the manuscript. A reader deciding whether to open the PDF, an editor doing a first screen, and an indexing database deciding how to categorize your work all use the abstract, and only the abstract, to make that decision. That single fact drives almost every rule below: what belongs in it, how it is organized, and how ruthlessly it has to be cut to fit a word limit that is rarely negotiable.

This guide covers how to write a research abstract for a journal article or conference submission — structured and unstructured formats, section-by-section content, word limits and how to cut to them, what must never appear in an abstract, and the reporting-guideline requirements (PRISMA for systematic reviews, CONSORT for trials) that apply on top of a journal’s own format. It closes with two complete worked examples.

Structured vs. unstructured abstracts: which journals require which

A structured abstract breaks the summary into labeled sections — typically some version of Background, Methods, Results, and Conclusions. An unstructured abstract is a single paragraph with no subheadings, relying on sentence order alone to move from context to findings.

You generally don’t get to choose. The target journal’s instructions for authors specify the format, and following it precisely is a submission requirement, not a style preference. Two concrete, verified examples of how differently journals set this:

ICMJE’s Recommendations (the standard followed across most biomedical and clinical journals) call for a structured abstract on original research, systematic reviews, and meta-analyses, stating the study’s context/background, purpose, basic procedures, main findings, and principal conclusions. ICMJE deliberately does not fix the exact subheadings or a word count — it leaves both to the individual journal, which is why the labels you see vary (Background/Methods/Results/Conclusions is the most common four-part pattern, but Objective/Design/Setting/Participants/Results/Conclusions and other longer templates are also common in clinical journals).
PLOS ONE, by contrast, requires an unstructured abstract — a single paragraph, maximum 300 words, describing the main objective, how the study was done (without heavy methodological detail), and the most important results and their significance, with no citations and abbreviations kept to a minimum.

The practical rule: open the target journal’s own author instructions before you write a word, confirm structured or unstructured, and if structured, copy its exact subheading labels and order — don’t improvise your own. Some fields lean one way by convention (clinical and biomedical journals toward structured; humanities, much of the social sciences, and many physical-science journals toward unstructured) but conventions have exceptions on both sides, so always verify against the specific journal rather than the field average.

Reporting guidelines can add a further layer on top of a journal’s baseline structured format — see the reporting-guideline section below, and CASRAI’s Structured Abstract Format guide for a full side-by-side comparison of ICMJE, CONSORT-for-Abstracts, and PRISMA-for-Abstracts subheadings.

The standard sections, and roughly how much space each earns

Whether or not your target format uses visible subheadings, a research abstract almost always moves through the same four functions, in the same order. As a drafting heuristic — not a fixed rule any standard mandates — each does a different amount of work:

Background/context (shortest) — one to two sentences. State the specific gap or problem your study addresses, not a literature review. If a reader needs three sentences of background to understand why your study matters, that’s a sign the framing is too narrow or too broad for the audience of the target journal.
Methods (brief, but concrete) — two to four sentences. Design, participants or data source, sample size where it matters, and the primary analytic approach. Enough for a reader to judge whether the evidence type fits the claim — not a full protocol.
Results (usually the longest section) — three to six sentences, or more in a longer structured format. This is what the abstract exists to deliver. Report the primary outcome(s) with actual numbers (effect sizes, confidence intervals, or p-values where applicable), not just a direction (“improved,” “was associated with”) with no magnitude.
Conclusions (short, and no broader than the data support) — one to two sentences. State what the findings mean, for whom, and any genuine limitation that changes how the reader should weigh the result. This is the section most prone to overreach — a conclusion that claims more than the results section actually showed is one of the most common reasons an abstract misrepresents its own paper.

CASRAI’s Research Paper Abstract: A Worked Example guide dissects a real published structured abstract sentence by sentence against this same four-part logic, if you want to see it applied to an actual paper rather than a template.

Word limits, and how to cut to them

There is no single universal abstract word limit — it is set per journal or per conference, commonly somewhere in a 150-350 word range, with 250 words a frequently used cap for structured biomedical abstracts and PLOS ONE’s 300-word cap a concrete example at the unstructured end. Most submission systems reject an over-length abstract outright rather than truncating it, so the limit is a hard constraint, not a soft target.

A practical cutting sequence, roughly in the order to try it:

Remove every citation. Abstracts should not cite references — the one common exception is a trial or systematic-review registration number, which some reporting guidelines explicitly require (see below).
Cut hedging and throat-clearing. “It has been previously suggested that,” “in recent years, increasing attention has been paid to,” “the purpose of this study was to” — all can usually be deleted or replaced with a direct statement, at no loss of meaning.
Combine sentences that share a subject. Two short sentences describing the same cohort or the same outcome can usually merge into one without losing information.
Move procedural detail to the full-text Methods section. The abstract needs enough method to judge the evidence, not enough to replicate the study.
Cut the Background to its single load-bearing sentence. Background is almost always the section with the most room to shrink without losing anything a reader needs.
Trim adjectives and intensifiers (“significantly,” “substantially,” “highly”) that aren’t carrying a statistical meaning — if “significant” doesn’t mean statistically significant in that sentence, it’s usually deletable.

What belongs in an abstract — and what must never appear in one

Belongs: the study’s purpose, essential design/methods, the primary results with actual numbers, and conclusions that are no broader than those results support. For a structured format, exactly the labeled content each subheading calls for and nothing else.

Must not appear:

Citations to other work — an abstract is read and indexed independently of the reference list; a citation there is functionally orphaned, and most journals explicitly prohibit it. The recognized exception is a required trial/review registration identifier (ClinicalTrials.gov, ISRCTN, PROSPERO), which is a locator, not a citation.
Undefined abbreviations. An abstract is frequently indexed and read as a standalone unit, separated from the full text where the abbreviation was first defined — every non-standard abbreviation needs to be spelled out at first use inside the abstract itself, or avoided entirely if it only saves a few words.
Results that don’t appear anywhere in the main text. The abstract has to be a faithful compression of the paper, not an independent summary written from memory after the fact. Reviewers and editors do check this, and abstract/full-text inconsistency is a recognized quality problem in meta-research literature on reporting practices.
Conclusions the data don’t support. Overstating effect size, causality (implying causation from an observational design), or generalizability beyond the studied population is one of the most common and most consequential abstract errors, because the abstract is often the only part of the paper many readers ever see.
New information not derivable from the rest of the paper. The abstract summarizes; it does not introduce.

Graphical and video abstracts

Many journals now request a visual summary alongside — not instead of — the text abstract. Requirements vary by publisher: Elsevier calls it a graphical abstract (single image, publisher-set minimum resolution, required vs. optional set per journal); Cell Press requires one for most of its titles with a few named exceptions; the Royal Society of Chemistry calls the same concept a “table of contents entry”; and some medical and social-media-facing journals additionally use a “visual abstract,” a more heavily templated multi-panel graphic aimed at a lay or clinical audience rather than the formal table-of-contents thumbnail. A growing number of journals also accept or request a short video abstract, typically a 1-3 minute author-narrated summary hosted alongside the article. For the full specs by publisher and step-by-step guidance, see CASRAI’s Graphical Abstract: Examples and Guidelines and How to Create a Video Abstract for a Research Paper.

How the abstract drives indexing and discovery

Before full-text search existed, and still today wherever a reader encounters a citation before the article itself, the abstract is what a database or a search interface actually indexes and matches against a query — PubMed/MEDLINE, Google Scholar, Scopus, and Web of Science all weight title and abstract text heavily in relevance ranking. The American national standard governing this is ANSI/NISO Z39.14, Guidelines for Abstracts — maintained by NISO, oriented specifically toward abstracting and indexing use, and the source of the long-standing library-science distinction between an indicative abstract (describes what the document covers, without stating findings) and an informative abstract (summarizes the actual methods and results, able to substitute for reading the full text on many practical questions). Most journal-article abstracts in the sciences are informative abstracts, because that is what an indexing database and a time-pressed reader both need.

The practical implication for discoverability: use the actual terms a searcher or a database would match on — the specific population, intervention, method, or variable names — rather than only generic phrasing, since the abstract (together with the title and author-supplied keywords) is frequently the only full-text-searchable surface a paper has in a given index.

Reporting-guideline abstract requirements: PRISMA and CONSORT

Two reporting guidelines specifically extend abstract-writing requirements beyond a journal’s own house format — this is the layer general abstract-writing advice usually skips, and it is a real submission requirement, not a stylistic suggestion, for the study types it covers.

CONSORT for Abstracts (Hopewell, Clarke, Moher, Schulz and colleagues for the CONSORT Group, published in The Lancet and PLoS Medicine, 2008) applies to randomized controlled trials. It does not mandate a fixed set of subheadings the way a journal’s own template does — it requires that specific information be present somewhere in the abstract, structured or not: the trial design, how participants were selected and how many were randomized, the interventions and comparator, the primary objective, how randomization and blinding were done, the numbers analyzed for the primary outcome and the result with a measure of precision, important harms, and — critically, and easy to omit if you’re not specifically checking for it — the trial registration number and primary funding source. Many journals implement this inside their own structured template by adding fields like Trial Design, Randomization, and Trial Registration.
PRISMA 2020 for Abstracts applies to systematic reviews and meta-analyses. It is a 12-item checklist covering: identifying the report as a systematic review/meta-analysis in the title; objectives (including the PICOS elements — participants, interventions, comparators, outcomes, study designs); eligibility criteria; information sources and search currency; risk-of-bias assessment method; number and type of included studies; synthesis of results; a plain-language description of the effect; limitations; interpretation; funding; and registration (for example, a PROSPERO number).

For a full item-by-item comparison of how ICMJE, CONSORT, and PRISMA subheadings map onto each other, see Structured Abstract Format. For a complete, annotated PRISMA-compliant systematic-review abstract, see Systematic Review Abstract: A PRISMA-Aligned Worked Example. Case reports have their own reporting-guideline abstract requirement too — see How to Write a Case Report Abstract: The CARE Guidelines Structure.

Worked example 1: a structured (IMRaD/clinical) abstract

The example below is an illustrative composite, written for this guide to demonstrate structured-abstract mechanics — it is not a real study, and no data, institution, or finding described in it should be treated as real. For a dissection of an actual published structured abstract, see the Research Paper Abstract worked-example guide instead.

Background: Adherence to oral antihypertensive medication remains poor in outpatient settings, and text-message reminders have shown mixed results across prior small trials. We evaluated whether a twice-daily automated text-message reminder improved medication adherence in adults newly prescribed an antihypertensive.

Methods: We conducted a parallel-group randomized controlled trial at three outpatient clinics. Adults (n=412) starting a first antihypertensive prescription were randomized 1:1 to receive twice-daily text-message reminders plus usual care, or usual care alone, for 12 weeks. The primary outcome was the proportion of participants with ≥80% adherence at 12 weeks, measured by electronic pill-bottle monitoring. Analysis was intention-to-treat.

Results: 391 participants (94.9%) completed follow-up. In the intervention group, 71.2% achieved ≥80% adherence versus 54.8% in the usual-care group (absolute difference 16.4 percentage points, 95% CI 7.1-25.7, p<0.001). No significant between-group difference was found in systolic blood pressure change at 12 weeks (mean difference -2.1 mmHg, 95% CI -5.4 to 1.2, p=0.21).

Conclusions: A twice-daily text-message reminder improved short-term medication adherence but was not associated with a measurable blood-pressure difference over 12 weeks. Longer follow-up is needed to determine whether the adherence gain translates into a clinical outcome difference. Trial registration: [illustrative registration number, not a real identifier].

Why this works as a structured abstract: each labeled section does exactly one job and nothing else. Background states the gap in two sentences without a literature review. Methods gives design, sample size, and the primary outcome definition — enough to judge the evidence, not a full protocol. Results reports the primary outcome first, with an actual effect size and confidence interval, then honestly reports a null secondary finding rather than omitting it. Conclusions states what the finding means and names the real limitation (short follow-up) instead of overreaching into a claim about blood-pressure benefit the data don’t support.

Worked example 2: an unstructured (humanities/conference) abstract

The example below is also an illustrative composite, written for this guide — not a real study, dataset, or archive finding.

Scholarship on nineteenth-century transatlantic abolitionist rhetoric has largely treated published pamphlets and newspaper columns as the primary evidentiary base, leaving the private correspondence among abolitionist organizers comparatively unexamined as a rhetorical source in its own right. This paper draws on a corpus of unpublished letters exchanged among three regional abolitionist committees between 1834 and 1848 to trace how the movement’s public rhetorical strategies were debated, tested, and revised before they reached print. Close reading of the correspondence shows organizers explicitly weighing the persuasive risks of moral-suasion appeals against more confrontational political-action framing well before that tension became visible in the published record, and shows individual writers adapting their private arguments differently depending on the recipient’s regional political context. The paper argues that treating private correspondence as a rhetorical workshop, rather than only a logistical record, reveals an internal strategic debate that published abolitionist writing later smoothed over into a more unified public voice. This has implications for how movement rhetoric more broadly should be read: public texts may represent a settled outcome of debate rather than the debate’s actual range.

Why this works as an unstructured abstract: there are no subheadings, but the same four functions still appear in order — the opening two sentences establish the gap (private correspondence is understudied as a rhetorical source), the middle states what was examined and how (a specific, bounded letter corpus, close reading), the next sentences report the actual finding (organizers debated rhetorical strategy privately before it appeared in print), and the closing sentences state the argument’s significance without overclaiming beyond what a close-reading method of a bounded corpus can support. Note also what it doesn’t do: no citations, no undefined jargon, and a claim scoped to “movement rhetoric more broadly should be read” rather than an unsupportable claim about the entire abolitionist movement.

Frequently asked questions

How long should a research abstract be?