Journal Word Limits: What Counts, and What Going Over Costs

Whatever your target journal's author instructions say counts, and the definitions differ a great deal. The same 6,000-word manuscript can be within one journal's limit and well over another's, because they leave out different parts. These are the rules on seven journals' own pages, read on 25 September 2026:

Nature Communications' figure is not a gate at submission: the same page says it "is flexible with regard to the format of initial submissions" and that "Within reason, style and length will not directly influence consideration of a manuscript." Detailed formatting instructions come from the editor at revision.

If you are 800 words over by Word's count, find your journal's row, or its equivalent on its own page, before you cut a sentence. You may be under already.

Read your journal's line for each of these; the pages above do not treat them the same way, and some do not mention them at all:

• Abstract. JAMA, Nature Communications and Frontiers leave it out of the main count, and Nature Communications caps it at 200 words; PLOS ONE, which has no overall limit, caps the abstract at 300. Nature goes the other way: its typical length is "2500 words (summary paragraph plus body text)", so its summary paragraph is inside.
• Methods. Nature Communications leaves the Methods out of its 5,000 words and adds that "Methods are typically less than 3000 words." A journal that counts the Methods changes the whole arithmetic.
• References. Left out at JAMA, Nature Communications and Frontiers, and at Nature too: "Title, author list, acknowledgements and references are not included in total word counts." Inside the seven pages at IEEE Transactions on Electron Devices.
• Tables, figures and legends. JAMA leaves tables and figures out of the word count but limits their number; Nature Communications leaves figure legends out but limits each to 350 words; Frontiers leaves out captions and the contents of tables.
• Footnotes and in-text citations. Frontiers counts them, "The manuscript length includes only the main body of the text, footnotes, and all citations within it", and its guidance for the Results section asks for none there: "Footnotes should not be used and must be transferred to the main text."
• Appendices. Frontiers: "Appendices should be no more than one page long and count towards the overall manuscript length and word count."

Final manuscript check: AI disclosure, word count, citations and DOCX integrity has a section on settling the word-count definition before editing, and it applies here unchanged.

Nature and the two IEEE journals measure space as well as words, so a figure competes with your text for the same budget. Nature makes the exchange rate explicit: "A ‘modest’ display item is one that, with its legend, occupies about a quarter of a page (equivalent to ~270 words)", and "An uninterrupted page of text contains about 1250 words."

That changes where cutting pays. On a page-limited journal, merging two figures or moving a table to supplementary material can recover more room than an afternoon of trimming sentences, without rewriting any of them. Check first that the journal allows it and what it counts; IEEE Transactions on Evolutionary Computation, for one, states its maximum length "(excluding supplementary materials)". Moving a table that readers need in order to follow your results costs more than it saves.

At some journals being over is not only a reason for an editor to send the paper back. It is a bill.

IEEE Transactions on Evolutionary Computation charges $200 for every printed page beyond 10 in a regular paper, 15 in a survey and 6 in a letter, and caps the total, with charges, at 15, 21 and 9 pages. In its own words, authors "will be assessed a mandatory page charge of $200 (per page) for every printed page over these limits."

IEEE Transactions on Electron Devices allows an eighth page "when absolutely necessary", on these terms: "For regular papers exceeding seven pages in length, a mandatory page charge of $200 per page (for each page beyond seven) will be assessed. This overlength page charge is obligatory."

These are two IEEE journals, not IEEE as a whole; page limits and charges are set journal by journal.

We also tried to read the author instructions of The BMJ, The Lancet, NEJM, Science and Cell on 25 September 2026. Our automated reader was served a bot-check page each time, and two of the addresses we tried turned out to have moved, so none of them is in the table above. We have not described their rules from memory or from third-party summaries. If you are submitting to one of them, its own author instructions page is the only source that counts.

Once you know what counts, cut in this order:

1. Move before you cut. Methods detail, extra tables and secondary analyses can sometimes go to supplementary material, but only if your journal permits it, does not count it, and readers can still follow the work without it.
2. Cut repetition across sections. Results restated in the Discussion, and a Discussion that re-summarises the Introduction, are the usual places a paper says the same thing twice.
3. Then tighten sentences. Numbers, units, citations and qualifiers are your findings; they should come through the cut unchanged.

If you use an AI tool for any of this, check your publisher's policy first. Publishers draw different lines between correcting your text and generating it, and several require a disclosure; Publisher AI disclosure policies, compared goes through them one by one.

That is the step HumanPen's condense option is for. Reduce the word count to a target while facts, numbers and citations stay, returned as the same editable file. It works on English documents. Set the target from your journal's definition, not from Word's total, and compare what comes back with your original, number by number and citation by citation.

Does the abstract count toward a journal's word limit? It depends even among these seven journals. JAMA, Nature Communications and Frontiers leave it out of the main count; Nature puts its summary paragraph inside the typical word length.

Do references count? It depends on how the limit is measured. JAMA, Nature, Nature Communications and Frontiers leave the reference list out of the word count, while IEEE Transactions on Electron Devices includes references inside its seven pages.

Can I use an AI tool to cut words from my manuscript? Check your publisher's policy before you do. Publishers do not agree on where shortening falls: some exempt only basic grammar and spelling checks, others treat heavier language editing as permitted with a disclosure, and a pass that rewrites your sentences may fall outside a narrow exemption even though it starts from your text. Publisher AI disclosure policies, compared shows where each one draws the line.

What if my journal's page does not say what counts? Ask the editorial office before you cut, and report the count you used on your title page or cover letter. Frontiers, for one, asks authors to "indicate the number of words and the number of figures and tables included in your manuscript on the first page."

• JAMA, Instructions for Authors, read 25 September 2026.
• Nature Communications, Submission guidelines: Article, read 25 September 2026.
• Nature, Formatting guide, read 25 September 2026.
• PLOS ONE, Submission guidelines, read 25 September 2026.
• Frontiers, Author guidelines and Frontiers in Medicine, Article types, read 25 September 2026.
• IEEE Electron Devices Society, Information for authors: IEEE Transactions on Electron Devices, read 25 September 2026.
• IEEE Computational Intelligence Society, Information for authors: IEEE Transactions on Evolutionary Computation, read 25 September 2026.