Writing an Abstract

The abstract is what almost everyone reads, and often all they read. Around one in ten of the people who see it will go on to read the full paper, and many will cite the work from the abstract alone. Reviewers form early opinions here that are hard to move later. So the abstract has to do the job of every section of the paper in 150 to 350 words.

The shape of an abstract#

A working abstract answers four questions in order.

Context and question (1 to 2 sentences) — what is the problem and why does it matter, ending at the specific gap you address.
Approach (1 sentence) — what you did, in plain English, without method jargon or software names.
Key result, with numbers (1 to 3 sentences) — the headline finding, stated with direction and magnitude.
Conclusion and significance (1 to 2 sentences) — what changes if the finding holds, stated concretely.

Read in sequence these answer why the work was needed, how it was done, what it showed, and why it matters. Most papers expand this to five to ten sentences, with the context block usually taking two or three.

The abstract is a miniature of the paper’s argument, not a teaser for it. It should stand on its own and give the reader the whole story in compressed form. Do not hold back the result to lure people into the paper. State the finding.

Write it last#

Write the abstract after the rest of the paper is finished, when you know exactly what you are summarizing. Last does not mean rushed. By the time you reach it you may want to be done, but this is the most-read paragraph you will write, so spend the effort.

Do not paste sentences from the introduction or discussion. In particular, use a different opening sentence from the one that opens the introduction. Write two strong openers and give one to each.

Give the result numbers#

The single most common failure is a result stated without numbers. “Herbivore exclusion increased tree establishment” tells the reader almost nothing. “Herbivore exclusion led to a 20-fold increase in trees reaching one meter over five years” tells them what happened. Report the direction and the size of the effect in plain language. Statistical detail that fit the results section will not fit the word budget here, so pick the one or two findings the reader most needs.

A worked skeleton#

Context and question. [One or two sentences: the topic, then the gap. End on the specific question. Do not open with a caveat or with “Scientists have long known”.]

Approach. [One sentence: system plus design, in plain English. “We tested X by doing Y.”]

Result. [One to three sentences: the headline finding with direction and magnitude. Add a second finding only if it materially adds.]

Significance. [One or two sentences: one specific implication, not a vague gesture at “importance for many fields”.]

A worked example#

An infectious-disease abstract built on the skeleton.

Symptom-based isolation is a mainstay of outbreak control, but it fails if much transmission happens before symptoms appear, and the size of that pre-symptomatic share is unknown for pathogen X. We estimated the serial-interval and incubation-period distributions from 312 contact-tracing pairs and used them to infer the fraction of transmission occurring before symptom onset. We found that 48% (95% CI 39 to 57%) of transmission occurred pre-symptomatically, and a transmission model showed that isolating only symptomatic cases left the reproduction number above one. Controlling pathogen X therefore requires testing- or contact-based interventions rather than symptom-based isolation alone.

Four blocks, each answering one question, with the key result carried by a number.

Common failures#

Result with no numbers. Vague statements like “was associated with changes” hide the finding. Give direction and magnitude.
Method dump. Three sentences on technique is an introduction to the methods section, not an abstract. Cut to one.
Overclaiming. “Groundbreaking” and “first ever” without unique data annoy readers and train them to skip your next paper. Make a concrete claim instead.
Vague significance. “Important for many fields” says nothing. Name the one thing that changes.
New acronyms. Never introduce an acronym in an abstract. Spell the term out.
Downbeat opener. Limitations belong in the discussion. Open with motivation or with the result.

A note on the title#

The title is the first filter, and most people who see it never reach the abstract. Carry the finding, not the method. “Neural networks double the accuracy of protein structure prediction” beats “A neural network method for protein structure prediction”. Be specific about direction: “Lion populations decline faster where human density is higher” beats “Human density is linked to lion population change”. Put the message at the front, aim for five to fifteen words, and use plain English. Avoid acronyms except the universally familiar ones such as DNA and COVID-19. Shorter titles get cited more, so cut every word that does not carry weight, starting with “A study of” and “An investigation into”.

Why it matters#

The abstract and title decide who reads the paper and what most people remember of it. A sharp title brings the right readers in, and an abstract that states the question, the approach, the result with numbers, and the significance gives them the argument even if they read nothing else. Writing them well is not decoration on top of the science. It is how the science reaches the people who will use it.