We Tested Two Common SEO Tips on 11,945 PubMed Papers. Only One Held.

Publisher SEO guides have converged on two recurring pieces of advice for researchers writing paper titles:

1. Use common, frequently-used terms — don't bury your paper in rare jargon, because papers with "more common and frequently used terms also tend to have increased citation rates." (This claim appears prominently in the Royal Society / Proc B 2024 guide, and echoes through Elsevier, Wiley, and Enago guidance.)
2. Write declarative titles — state the finding, don't describe the topic. "X promotes Y via Z" rather than "A study of X in Y."

Both tips are repeated everywhere. Neither had been tested together at population scale. So we tested them — on 11,945 PubMed papers across four biomedical fields, published in 2023, with citations pulled from the NIH iCite database in April 2026 (2–3 years of citation accumulation).

One held. The other didn't.

The Sample

We pulled 3,000 PubMed IDs for each of four MeSH-defined fields, restricted to 2023 original research articles in English with abstracts. After filtering for papers with usable metadata and citation records from iCite, we had 11,945 papers.

Test 1 (the null): "Use common terms in your title"

To test this, we needed a principled measure of "how common are the words in this title?"—relative to what's normal in the field.

Our method:

1. For each field, we built a word-frequency rank from every title and abstract in the sample. "Cancer" ranks 1 in cancer biology; "tolvaptan" ranks somewhere in the thousands.
2. For each paper, we tokenised its title (dropping stopwords and trivially common filler), looked up each word's rank, and computed the mean log-rank. Lower score = uses more common terms.
3. We binned papers into quintiles by this score and looked at median citations per quintile.
4. As a robustness check, we also computed the share of title words falling in the top 500 field terms.

If the claim is right, Q1 (most common terms) should have clearly higher citations than Q5 (rarest terms).

What we found

The correlations are trivially small and inconsistent in sign. In cancer biology and neuroscience, there's a tiny edge toward common terms. In infectious disease, it's actually reversed — papers with rarer terms get more citations. Cardiology shows nothing at all.

The top-500 share check gave the same result: median citations did not trend monotonically with common-term share in any field.

The one honest caveat

When we stratified by title length, one pattern stood out. In the short-title bin (≤8 words, n=1,475 pooled), titles with common terms had a higher mean (7.87 vs 5.68, Mann-Whitney U, p=0.012). But the medians were identical (3.0 vs 3.0). Common-term short titles produced more high-citation outliers, not more typical wins.

This is a weak positive signal at best, and only applies to very short titles — which themselves are rare in modern biomedical publishing. It's not enough to rescue the general tip.

Test 2 (confirmed): "Write declarative titles"

We classified every title in the sample into one of four structural types using a simple heuristic:

• Declarative: contains a tensed finding-verb (promotes, reduces, drives, predicts, is, identifies, regulates, inhibits, etc.), and no colon or question mark.
• Descriptive: default — noun-phrase titles, no finding-verb.
• Compound: contains a colon.
• Question: ends with a question mark.

This is conservative. The classifier only marks a title as declarative when it clearly states a claim. Titles like "Role of X in Y" or "Characterisation of X" are descriptive; titles like "X drives Y" are declarative.

What we found

Declarative titles beat descriptive ones on every metric: +33% median citations (4 vs 3), +29% mean (6.85 vs 5.30), +46% chance of reaching 10 citations (20.5% vs 14.0%), and roughly half the rate of zero-citation outcomes (10.3% vs 18.3%).

Question titles remain the worst-performing structure — consistent with Jamali & Nikzad (2011), who found question titles attract downloads but not citations.

Does the effect survive field-level testing?

Yes — in every single field.

Does the effect survive length-matching?

Also yes. If declarative titles just happened to be a more popular length, we'd expect the effect to collapse when we control for title length. It doesn't:

The declarative advantage holds at every length bin. In fact, for longer titles (13+ words) the effect gets larger — meaning if you're going to write a long title, making it declarative matters even more.

Why One Tip Holds and the Other Doesn't

Our best explanation: declarative titles carry semantic content about the paper's finding; common-term density does not.

A declarative verb ("promotes," "reduces," "identifies") tells a reader something concrete about what the paper concludes. That information travels through abstract previews, search snippets, and LLM-generated summaries. A common-term title ("Cancer and cells in 2023") tells the reader almost nothing. High-frequency words are generic by definition.

The Royal Society's claim probably overgeneralises from information-retrieval intuition ("common words match more queries") into citation behaviour ("therefore, more citations"). But citation behaviour isn't driven by retrieval alone — it's driven by researchers deciding your paper is worth reading and worth citing. Generic words don't help with that decision. Specific claims do.

What This Means for Your Next Paper

1. Stop worrying about whether your title uses "rare" terminology. If a precise term is the right term — a gene name, a compound, a mechanism — use it. The common-term advice is not empirically supported.
2. Do worry about whether your title states a finding. Before submission, look at your title and ask: does it contain a verb that carries a claim? If it's a noun phrase ("Role of X in Y", "Characterisation of Z"), rewrite it.
3. Avoid question titles. They are the lowest-performing structure in every dataset we've seen.
4. The length effect on declarative titles is worth knowing. If your paper needs a longer title (mechanism, context, multiple conditions), making it declarative matters more, not less.

Before → After examples from the dataset

Before (descriptive): "Mitochondrial dysfunction in neurodegenerative disorders."

After (declarative): "Mitochondrial dysfunction drives neuronal death in Parkinson's disease."

Before (descriptive): "Single-cell transcriptomics of tumour-infiltrating lymphocytes."

After (declarative): "Exhausted CD8+ T cells reactivate upon PD-1 blockade in melanoma."

Limitations

• Correlation, not causation. Papers with declarative titles may come from more confident research teams with stronger findings, which would independently attract citations. We cannot rule this out with observational data. But the effect replicates across four unrelated fields and is stable under length-matching, which partially addresses the strongest confounders.
• The declarative classifier is a heuristic. It relies on a fixed list of finding-verbs and would miss some valid declarative titles and might over-count some edge cases. The rate of declarative classification (9.7% of titles) is consistent with prior hand-coded studies.
• Within-field frequency rank may not capture what Google Scholar's algorithm actually weights. Our test measures how researchers in the field actually use words; the Royal Society claim is about reader behaviour. If "common terms" means something different — e.g., terms common in the broader English-language corpus rather than the field — the result could differ. We don't believe that's the correct interpretation of the original claim, but it's a defensible alternative worth naming.
• Citation window is 2–3 years. Citation patterns shift over longer horizons. The effect sizes here reflect early-career citation accumulation.
• Four biomedical fields only. We can't say the declarative effect generalises to physics, mathematics, social sciences, or humanities. The null result on common terms likely generalises more broadly (because the signal was absent even in the best case), but that's our inference, not our data.
• iCite citation counts differ from Scopus or Google Scholar. iCite uses NIH's own citation graph, which is comprehensive for biomedical literature but differs in coverage from commercial databases.

Replication

The raw data (12k PubMed IDs, titles, abstracts, iCite citation counts as of 04 April 2026) and the analysis script are available on request. We encourage replication with different fields, longer citation windows, or alternative declarative classifiers.

References

1. Hansford HJ, et al. Title, abstract and keywords: a practical guide to maximize the visibility and impact of academic papers. Proc R Soc B. 2024;291(2027):20241222. doi:10.1098/rspb.2024.1222
2. Paiva CE, Lima JPSN, Paiva BSR. Articles with short titles describing the results are cited more often. Clinics. 2012;67(5):509–513. doi:10.6061/clinics/2012(05)17
3. Jamali HR, Nikzad M. Article title type and its relation with the number of downloads and citations. Scientometrics. 2011;88:653–661. doi:10.1007/s11192-011-0412-z
4. Letchford A, Moat HS, Preis T. The advantage of short paper titles. R Soc Open Sci. 2015;2(8):150266. doi:10.1098/rsos.150266
5. Hutchins BI, Yuan X, Anderson JM, Santangelo GM. Relative Citation Ratio (RCR): A new metric that uses citation rates to measure influence at the article level. PLoS Biol. 2016;14(9):e1002541. (iCite methodology) doi:10.1371/journal.pbio.1002541