An abstract is a short, self-contained summary of a paper, thesis, talk or grant, usually one paragraph of 100 to 350 words depending on where it will appear. An informative abstract reports the purpose, method, results and conclusion, so it can stand in for the work; a descriptive abstract says only what the work covers. The 12 examples below are sorted by type and field, and the table near the end lists the limits that journals, a university and a funder set.
For the step-by-step method and six more examples, see the guide to how to write an abstract. Every study, dataset and number below is invented to show the form. The venue rules are real and come from each publisher's or institution's own instructions.
Abstract types at a glance
The UNC Writing Center's handout on abstracts separates two main types. A descriptive abstract tells the reader what kind of information the work contains and stops there: no results, no conclusions. An informative abstract adds the results, conclusions and recommendations, so it can serve as a stand-in for the work. The handout mentions a third type, the critical abstract, as rarely used. Structured and unstructured describe the layout, not the content: an informative abstract can be either.
| Type | What it includes | Typical length | Examples |
|---|---|---|---|
| TypeDescriptive | What it includesPurpose, method and scope; no results | Typical lengthUsually 100 words or fewer | Examples1 |
| TypeInformative | What it includesPurpose, method, results and conclusions | Typical lengthThe venue's limit; rarely more than 10% of the work | Examples2, 4–7 |
| TypeStructured | What it includesAn informative abstract under labeled headings | Typical lengthSet by the journal | Examples3, 11 |
| TypeThesis or dissertation | What it includesQuestion, method, findings, often chapter by chapter | Typical lengthSet by the graduate school | Examples8, 9 |
| TypeConference | What it includesThe work, its result or claim, and what the session offers | Typical lengthSet by the call | Examples10 |
| TypeReview or grant | What it includesThe question and search, or the aims and approach | Typical lengthSet by the journal or funder | Examples11, 12 |
A word limit alone doesn't tell you which type to write. The UNC handout offers a rough guess for assignments that give only a length, 100 to 120 words for descriptive and 250 or more for informative, but says the better course is to ask your instructor or check the publisher's instructions. Example 4 below is the counterexample: an informative abstract, with its result and its number, in under 100 words.
Descriptive vs informative abstract
The clearest way to see the difference is one report written both ways. The report evaluates what happened when a city added a smartphone app to its 311 service-request line.
This report examines how adding a smartphone app to the 311 service line of Linden Falls changed who files service requests with the city. It compares request records from the two years before and after the app's launch in 2023, describes the request types and neighborhoods covered, and reviews the city's outreach to residents without smartphones. It closes with options for monitoring equal access to the service.
When Linden Falls added a smartphone app to its 311 service line in 2023, officials expected more requests and faster repairs. This report asks whether the app also changed who files requests. We compared 212,400 requests from the two years before the launch with 251,900 from the two years after, matched each to its census tract, and interviewed 14 call-center and outreach staff. Total requests rose 19%, and the median time to close a pothole request fell from nine days to six. The increase came almost entirely from tracts in the top half of household income, where requests rose 34%, while requests from the lowest-income quarter of tracts fell 8%. Calls to the phone line, still the main channel in those tracts, went unanswered longer after staff were moved to app messages. Because the city schedules repairs largely in the order requests arrive, the shift sends crews to wealthier streets first. We recommend restoring phone staffing, scheduling repairs from inspections rather than request volume, and publishing requests by tract every quarter.
What changes. The descriptive version lists what the report contains, much like a table of contents in prose, and a reader still has to open the report to learn anything. The informative version gives the design, the sample, the main results with their numbers and the recommendations. Use the descriptive form for a long report, a book chapter or a review essay without one central finding; use the informative form when there is a result to report, unless your instructor or venue asks for the other.
Structured vs unstructured abstract
A structured abstract puts each part under a heading. The ICMJE recommendations, which many medical journals follow, say original research, systematic reviews and meta-analyses require structured abstracts, and that the format differs from journal to journal. Here is one hospital study in The BMJ's headings, then as one paragraph.
Objectives: To assess whether a nighttime quiet protocol that changes staff routines improves patients' self-reported sleep on a general medicine ward.
Design: Controlled before-and-after study.
Setting: Two adult general medicine wards in one teaching hospital.
Participants: 412 adults who stayed at least two nights on the intervention ward (208 before and 204 after the change) and 398 on the control ward over the same periods.
Interventions: From 11 pm to 5 am, staff dimmed corridor lights, moved routine observations of stable patients outside those hours, silenced non-urgent bedside alarms and closed room doors unless patients preferred them open. The control ward kept its usual routine.
Main outcome measures: Sleep quality on the second night, scored 0 to 100 on a five-item questionnaire. Secondary outcomes were self-reported awakenings and falls.
Results: Mean sleep scores on the intervention ward rose from 48 to 61, against 50 to 52 on the control ward (difference in differences 11 points, 95% confidence interval 6 to 16). Patients reported 1.2 fewer awakenings a night (0.7 to 1.7). Falls were similar before and after the change (4 and 3).
Conclusions: Reported sleep improved on the ward that changed its night routines, without new equipment, while sleep on the control ward barely changed. Fall counts were similar in the two periods but too small to show whether risk changed. A cluster randomized trial across several hospitals would show whether the change causes the improvement and is safe.
Hospital patients often sleep poorly, and much of the disturbance comes from staff routines rather than illness. We tested whether a nighttime quiet protocol improved patients' self-reported sleep in a controlled before-and-after study on two general medicine wards of one teaching hospital. From 11 pm to 5 am, staff on the intervention ward dimmed corridor lights, moved routine observations of stable patients outside those hours, silenced non-urgent alarms and closed doors, while the control ward kept its usual routine. Among 412 adults on the intervention ward and 398 on the control ward who stayed at least two nights, mean sleep scores on a 0–100 questionnaire rose from 48 to 61 on the intervention ward and from 50 to 52 on the control ward, a difference of 11 points (95% CI 6 to 16). Patients reported 1.2 fewer awakenings a night, and fall counts were similar in the two periods. Reported sleep improved after staff changed their night routines, without new equipment; a cluster randomized trial across hospitals would show whether the change causes the improvement.
What changes. The headings replace the transitions and make it impossible to skip a part, which is why clinical journals like them. The paragraph keeps the main finding and its interval but compresses the supporting detail: it drops the before-and-after split of participants (208 and 204), the interval for awakenings and the fall counts themselves, and it adds one sentence of context to read naturally. Both versions word the conclusion with care. A controlled before-and-after study can show that reported sleep improved after the change, not that the change caused it, and four falls against three are too few to show whether risk went up or down. The study is invented; the caution is about the wording, not about real quiet protocols. If a journal names its headings, copy them exactly, in its order. These are the sets two major medical journals publish:
| Journal and article type | Headings, in order |
|---|---|
| Journal and article typeJAMA, original research (up to 350 words) | Headings, in orderImportance; Objective; Design; Setting; Participants; Intervention(s) or Exposure(s); Main Outcome(s) and Measure(s); Results; Conclusions and Relevance; Trial Registration (trials only) |
| Journal and article typeThe BMJ, research (250–300 words) | Headings, in orderObjectives; Design; Setting; Participants; Interventions; Main outcome measures; Results; Conclusions; Trial registration |
| Journal and article typeJAMA, systematic review (up to 350 words) | Headings, in orderImportance; Objective; Evidence Review; Findings; Conclusions and Relevance |
| Journal and article typeJAMA, narrative review (up to 400 words) | Headings, in orderImportance; Observations; Conclusions and Relevance |
Abstract examples by field
Fields differ in what counts as a result. Economics wants the estimate, sociology the pattern and the argument, the natural sciences the finding in terms an outsider can follow, philosophy the claim. The guide to writing an abstract already covers psychology, public health, literature, computer science, history and environmental science, so these four come from elsewhere.
Economics: an abstract under 100 words
We estimate how the shift to remote work after 2020 affected house prices in small towns near large cities. Using 1.9 million sales in 2,300 towns, we compare towns that had fast broadband before 2020 with otherwise similar towns that did not. Prices in connected towns rose 7.4 percent more by 2024, and the gap was largest within 90 minutes of a metropolitan center. Rents rose less than prices, consistent with buyers who expect to work from home for years rather than months.
Why it works. Under a 100-word cap there is no room for a literature sentence, so the first line is the question. The design (a comparison of connected and unconnected towns) takes one sentence, the estimate gets a number, and the last sentence uses a second result to interpret the first.
Sociology: a qualitative study
Food-delivery platforms rate couriers on speed and customer feedback, but couriers cannot see how the ratings are calculated. Drawing on 46 interviews with couriers in two cities and eight months of observation at restaurant pickup points, this article examines how couriers make sense of a rating system they cannot inspect. Couriers built shared theories of the algorithm, passed on in group chats and at pickup counters, and changed their work to fit them: refusing short orders at certain hours, avoiding restaurants believed to lower ratings and accepting unpaid waiting time to protect a score. Many of these theories contradicted the platform's published rules, yet they organized the work as firmly as the rules did. I argue that opacity does not leave workers passive. It shifts the work of managing the system onto them and onto each other. The article adds to research on algorithmic management by showing how informal knowledge, rather than the code itself, sets the terms of daily work.
Why it works. Qualitative abstracts report a pattern instead of a number, so the result sentence lists concrete behaviors. The argument is stated in the first person, as single-author qualitative work often is, and the last sentence names the conversation the article joins.
Ecology: a Nature-style summary paragraph
Birds that sing at dawn compete with the sounds of a waking city. Songbirds in noisy urban areas can sing at a higher pitch than birds of the same species in quieter places¹⁻³, and the difference is linked to low-frequency background noise⁴. Whether the shift is a flexible response or a lasting difference between populations has been hard to test, because cities differ from the countryside in light, temperature and food as well as sound⁵. Here we show, using a natural experiment created when a major road closed for months of repairs, that songbirds beside the road lowered their pitch within weeks of the closure and raised it again within days of the reopening, while birds at comparison sites did not change. The speed of both shifts points to flexibility within individuals rather than adaptation across generations, and suggests that quieter roads could restore natural song within a single season.
The superscripts 1 to 5 are illustrative: they show where a real summary paragraph would cite five studies, and there are no papers behind them to look up.
Why it works. It follows the order in Nature's formatting guide: a basic introduction for readers outside the field, the rationale, the main result introduced by "Here we show" and wider context at the end (the guide asks for two or three sentences there; this short sample uses one). It stays inside the guide's ideal of 200 words. Numbers and measurements are left out unless essential, which runs against the usual advice to give the main result as a number.
Philosophy: an argument abstract
Can you break a promise to yourself? A familiar argument says you cannot: a promise binds because the person promised can release you from it, and no one can be both the promiser and the person who releases. This paper argues that the argument mistakes a feature of most promises for a condition of all of them. Drawing on cases of resolutions written down and witnessed by no one, I argue that what makes a promise binding is not a second party's power of release but the promiser's invitation to be held to account, an invitation one can extend to one's future self. If this is right, promises to oneself are genuine promises, and breaking one is a moral failure rather than a change of mind.
Why it works. The "result" of a philosophy paper is its claim, and it is stated outright, with the view it opposes and the reason. Present tense throughout, no method section, and a last sentence that says what follows if the reader accepts the argument.
Thesis and dissertation abstracts
Thesis abstracts run longer and often walk through the chapters, so examiners and later readers can find the part they need. The limit comes from your graduate school. MIT's thesis specifications, for example, ask for an abstract of generally no more than 500 words, single-spaced, immediately after the title page, and describe it as a brief summary, not a lengthy introduction to the thesis.
New rail stations raise land values nearby, and planners have long debated whether that gain displaces the small businesses already there. This dissertation asks how the opening of light-rail stations between 2009 and 2019 affected independent businesses within half a mile of 41 stations in three US metropolitan areas. Chapter 2 builds a panel of 18,600 business establishments from commercial listings and state licensing records, linked to station opening dates. Chapter 3 compares establishments near new stations with those near stations planned but not yet open, and finds that the closure rate of independent businesses rose by 3.1 percentage points in the five years after opening, while the closure rate of chain stores did not change. The effect was concentrated among businesses that rented their premises and in neighborhoods where commercial rents had been below the regional median. Chapter 4 draws on interviews with 37 owners, 22 of whom had closed or moved, and shows that rent increases at lease renewal, rather than construction disruption, were the reason owners gave most often. Chapter 5 evaluates the two policies the regions tried, construction-period grants and commercial rent stabilization near stations, and finds that only the second was associated with lower closure rates. The dissertation argues that station-area planning has treated commercial displacement as a temporary construction problem rather than a lasting effect of rising rents, and it proposes lease protections that cities can attach to station-area zoning.
Why it works. The chapter-by-chapter walk-through reads as one argument because each chapter adds a step: the data, the main effect, the mechanism, the policy test. Every chapter reports a finding, not a topic, and the final sentence states the dissertation's claim and its practical use.
Universities increasingly replace lawn with ornamental flower beds, but it is unclear whether these plantings support pollinators as well as native plants do. This thesis compared insect visits to 12 native and 12 non-native ornamental flowering species planted side by side in a campus garden. From May to August 2025, I recorded visitors to each species in 288 timed observations of 15 minutes and identified the insects to family. Native species received 2.3 times as many bee visits per flower as the ornamentals and drew 14 bee families against 8. Two ornamentals, however, attracted more hoverflies than any native species. The results suggest that campus plantings meant for pollinators should favor native species, with a few ornamentals kept for other insect groups. Because the study covered one garden over one season, a survey across several campuses would show whether the pattern holds.
Why it works. An honors thesis abstract is closer to a journal abstract than to a dissertation's. It uses the first person for the work the student did, keeps the numbers that carry the result, reports the exception and ends on a limit paired with the study that would remove it. For planning the larger project, see the guide to writing a dissertation.
Conference abstract example
A conference abstract is judged by a committee, often before the work is finished, and then read in a program. The conference abstract examples cover the limits recent calls set, some in characters rather than words, and five more samples, including one written before the results were in. This one adds a talk in linguistics.
Switching at the seams: language alternation at topic shifts in bilingual podcasts
Research on code-switching has mostly examined spontaneous conversation, where speakers switch languages for a word, a quotation or a joke. Podcasts offer a different setting: talk that is unscripted but planned, for listeners who share both languages. This paper analyzes 40 hours of 12 Spanish–English podcasts made in the United States, transcribed and coded for every switch between languages. Hosts switched most often at topic boundaries, where 41% of all switches occurred, and least often inside a single anecdote. Switches into Spanish clustered around family and food, switches into English around work and technology. I argue that in planned bilingual talk, switching works as punctuation, marking the structure of an episode for listeners, and that this function is easy to miss in studies of conversation alone. The talk will play short clips of each pattern.
Why it works. The title carries the finding. The abstract names the usual setting of the research and why this one differs, gives the corpus in one sentence and the pattern in two, and ends with what the audience will hear, which is the first thing a program reader wants to know.
Review and grant abstracts
A review abstract reports the search and what the evidence adds up to; JAMA's format for systematic reviews gives it five headings. A grant abstract describes work not yet done, so its results are aims. For how the review itself is organized, see the literature review examples.
Importance: Office workers spend most of the workday sitting, and employers increasingly buy sit-stand desks to reduce it, but evidence on how long the effect lasts is mixed.
Objective: To estimate the effect of sit-stand desks on workplace sitting time at short-term and longer-term follow-up.
Evidence Review: MEDLINE, Embase and the Cochrane Central Register of Controlled Trials were searched from inception to January 2026 for randomized and cluster randomized trials comparing sit-stand desks with usual desks among office workers. Two reviewers independently screened studies, extracted data and assessed risk of bias. Results were pooled with random-effects meta-analysis.
Findings: Nineteen trials with 2,684 participants were included. Sit-stand desks reduced workplace sitting by 84 minutes per 8-hour workday at follow-up of 3 months or less (95% CI, 61-107 minutes; 11 trials) and by 52 minutes at 6 to 12 months (95% CI, 30-74 minutes; 6 trials). Trials that added coaching or prompts showed larger reductions at both time points. Eight trials were judged at high risk of bias, mainly because sitting time was self-reported.
Conclusions and Relevance: Sit-stand desks reduced sitting at work, but the effect fell by about 40% within a year. Employers who install them may need prompts or coaching to sustain the change, and trials with device-measured sitting and longer follow-up would settle how much of the effect lasts.
Why it works. The Evidence Review heading does the job a methods section does in a primary study: databases, dates, eligible designs and who screened. The findings give the number of trials and participants first, then each pooled estimate with its interval and the number of trials behind it, and the risk of bias in one sentence.
Project Summary/Abstract: Falls are a common and serious cause of injury in adults aged 65 and older, and several widely prescribed medications raise the risk. Primary care practices rarely see every medication a patient takes, because prescriptions come from several clinicians and pharmacies. The long-term goal of this project is to make medication-related fall risk visible in routine primary care. Its objective is to develop and test a tool that reads pharmacy refill records and flags patients whose combination of medications places them at high risk. Aim 1 will build the tool's risk score from five years of linked refill and emergency department records for 84,000 adults aged 65 and older in a regional health system. Aim 2 will test, in a cluster randomized trial in 20 primary care practices, whether a monthly list of flagged patients with deprescribing suggestions reduces fall-related emergency visits over 18 months. Aim 3 will interview clinicians and patients to learn which suggestions were acted on and why. The project will produce a tool that health systems can run on records they already hold.
Project Narrative: Some combinations of common medicines make falls more likely in older adults, but their doctors often cannot see every medicine they take. This project will build and test a tool that uses pharmacy records to spot patients at risk, so doctors can review those medicines before a fall happens.
Why it works. NIH limits the Project Summary/Abstract to 30 lines of text and asks that it be understandable to a scientifically literate reader, so the summary moves from problem to long-term goal, objective and one sentence per aim, with an approach in each. The Project Narrative, which becomes public, gets at most three sentences on the project's relevance to public health; this one uses two and no technical terms.
Abstract word limits by venue
No limit is universal, and many journals set a different one for each article type. These are the rules at a range of venues as of November 2026, from a 100-word economics abstract to a 500-word thesis abstract.
| Venue and article type | Field | Limit | Other rules |
|---|---|---|---|
| Venue and article typeAmerican Economic Review, submissions | FieldEconomics | Limit100 words | Other rulesRequired with every submission |
| Venue and article typeNature, Articles | FieldSciences | LimitIdeally no more than 200 words | Other rulesA fully referenced summary paragraph for readers outside the discipline |
| Venue and article typeAPA Style (7th ed.), papers | FieldPsychology, education, social sciences | Limit250 words | Other rulesNot usually required in student papers; keywords go one line below |
| Venue and article typeIEEE journals, articles | FieldEngineering, computing | LimitUp to 250 words | Other rulesOne paragraph; no abbreviations, footnotes, references or equations; 3–5 keywords |
| Venue and article typeThe BMJ, research articles | FieldMedicine | Limit250–300 words | Other rulesStructured; up to 400 words for CONSORT- or PRISMA-style abstracts |
| Venue and article typePLOS ONE, research articles | FieldAll sciences | Limit300 words | Other rulesNo citations; avoid abbreviations; clinical trials give the registry and number |
| Venue and article typeJAMA, original research | FieldMedicine | Limit350 words | Other rulesStructured; separate Key Points of 75–100 words |
| Venue and article typeJAMA, narrative reviews | FieldMedicine | Limit400 words | Other rulesThree headings |
| Venue and article typeMIT thesis | FieldAny field | LimitGenerally no more than 500 words | Other rulesSingle-spaced, right after the title page |
| Venue and article typeNIH grant application (Forms I) | FieldHealth research | Limit30 lines of text | Other rulesPlus a public Project Narrative of at most three sentences |
Conference calls set their own limits, from a 250-word paragraph to a character count that includes the title, and the conference abstract guide lists several. Your school's thesis manual or your instructor's style guide overrides everything in the table.
Adapting an example to your venue
Borrow the shape of an example, never its sentences, and then fit it to the rules of the place you are submitting. The venues above disagree on more than length:
- Citations. PLOS ONE allows none in the abstract and IEEE asks for no references, while Nature wants its summary paragraph fully referenced. Check before you add or remove them.
- Abbreviations and numbers. IEEE wants no abbreviations, PLOS ONE asks you to avoid them, and Nature discourages numbers and measurements unless they are essential. Elsewhere, the ICMJE recommendations and APA's reporting standards both ask for the main findings with their effect sizes.
- How length is counted. Words at most journals, lines at NIH, characters at some conferences. Paste your draft into the submission form early and trust its counter over yours.
- Extra summaries. JAMA asks for Key Points (a question, the findings and their meaning) in addition to the abstract. The BMJ asks for a box saying what was already known on the topic and what the study adds. Write them from the abstract, not the other way round.
- Registration and data. The ICMJE recommends that journals publish a clinical trial's registration number at the end of the abstract, and asks authors whose data sit in a public repository to give the data set's identifier there too.
Before you submit, you can attach the paper to a chat with the Strokes Assistant and ask it to list every number or claim in each abstract that differs from the paper or doesn't appear in it.
Frequently asked questions
How long is an abstract?
Most journal abstracts run 100 to 350 words, and the venue sets the limit, often by article type: JAMA, for example, allows 350 words for original research and 400 for a narrative review. Thesis abstracts can run longer. A descriptive abstract is usually 100 words or fewer. Check the current author instructions for your article type, since limits change.
What is a structured abstract?
An abstract divided under labeled headings, such as Objectives, Design, Results and Conclusions. The ICMJE recommendations say original research, systematic reviews and meta-analyses require one, and that the format varies by journal. JAMA's headings for original research run from Importance to Conclusions and Relevance; The BMJ uses its own set.
Can an abstract include citations?
It depends on the venue. PLOS ONE does not allow citations in the abstract, and IEEE asks for no references. Nature takes the opposite line: its summary paragraph for Articles is meant to be fully referenced. If the instructions are silent, leave citations out and name a key author in the sentence instead.
Is a thesis abstract different from a journal abstract?
Usually it is longer and covers more ground. Many thesis abstracts walk through the chapters, so examiners and later readers can find the part they need, and the limit comes from the graduate school rather than a journal. MIT, for example, asks for generally no more than 500 words, single-spaced, right after the title page.
Does an abstract need keywords?
Often, yes. IEEE asks for three to five keywords or phrases, and APA Style places a keywords line below the abstract when one is required. Choose the terms readers and indexing services would use for your subject, including the main topic phrase even if it is in your title, then add the method, the population or a common synonym if space allows. Follow the venue's rules on number and format.
