Report findings, do not describe topics
The commonest fault is the abstract that describes the paper rather than reporting it: this study explores, this paper examines, implications are discussed. A reader finishes knowing the subject and nothing else, which is the opposite of the job.
Replace every one of those with the actual result. Not that differences were examined, but what the difference was, in the units it was measured in, with whatever interval or significance level your field expects to see alongside it.
The word limit is a structure
Divide it roughly by section and the abstract writes itself: a sentence or two of background, one of aim, two or three of method, the largest share to results, one on what follows. Where structured abstracts are required the headings do this for you.
Where they are not, the same proportions still apply, and the section that always overruns is background. Cutting it is usually the fastest way to bring an over-length abstract into range without losing anything a reader needs.
It has to stand alone
No undefined abbreviations, no citations unless your field expects them, no reference to a figure the reader cannot see. It will be read in a database listing, detached from everything around it, by somebody deciding whether to open the paper.
It is also where keywords do their work, so the terms somebody would actually search for should appear in it. Write it last, from the finished results, then read it once cold — anything that leaves you asking what happened is the thing to fix.
One test catches nearly every weak abstract. Hand it to somebody who has not read the paper and ask what was found. If they can only tell you what it was about, the abstract is describing rather than reporting, and editing individual sentences will not repair a structural problem.
Three steps, then the record keeps itself.
Send the finished paper
The abstract is written from results rather than intentions, so it comes last and needs the completed work.
Get it built to the limit
Question, method, results with numbers, conclusion, proportioned to whatever word count applies.
Read it cold
Detached from the paper, as a database listing. If it does not say what was found, it is not finished.
Queries on record.
How long should an abstract be?
Whatever the journal, program or conference specifies, and the limit is usually enforced rather than advisory. Where nothing is stated, 250 words is a safe convention for a research paper. The limit matters less than the proportions inside it, and the section that reliably overruns is the background.
Should there be numbers in it?
In any empirical work, yes, and their absence is the clearest single sign of a weak abstract. A reader should finish knowing the size of what you found, not merely its direction. Where your field expects confidence intervals or significance values in an abstract, include them; where it does not, the effect size still belongs.
Can I adapt my introduction?
It is where most weak abstracts come from. An introduction argues that a question is worth asking; an abstract reports what happened when it was answered. Those are different documents with different jobs, and writing from the results section produces a much better abstract in noticeably less time.
How should I choose keywords?
Use the terms somebody looking for your work would actually type, checked against the controlled vocabulary your field's databases use rather than invented phrasing. Where a subject heading exists for your topic, using it materially affects whether your paper is found at all, which is the entire purpose of the field.
Is a dissertation abstract different?
It is usually longer and it matters more, because it may be the only part anybody outside your committee ever reads and it becomes the permanent description of your work in a repository. The structure is identical. The care taken over it should be greater and is almost always less.
Elsewhere in the hall.
Send the assignment and the deadline.
A specialist credentialed in your own field reads what you actually have and replies within two hours with a plan and a price. The reading costs nothing, and if the honest answer is that we are not the right hall for it, you will get that instead.
