Each feature in the abstract moved the grand-award rate by six to eight points. Across 8,207 abstracts from 2021 to 2026, a novelty claim (novel, first to) ran 45.2% vs 22.2% without; a cost or affordability claim 39.2% vs 24.7%; a comparison with an existing method 39.2% vs 25.7%; a named metric such as accuracy, AUC or RMSE 33.9% vs 24.9%; a reported p-value 34.5% vs 26.6%. Abstracts with no quantified result won at 22.8%, with one or two at 29.3%, with three to five at 34.5%. Length stops mattering above 200 words: the median is 246 words for every award tier.
Feature by feature
Features are detected by pattern matching, so the flags are loose, but the direction is clear. Abstracts that mention limitations or future work (12.0%) ran slightly lower at 25.1%; reporting a sample size made no difference. Abstracts written in the first person singular (I developed) won at 34.6%, first person plural (we developed) at 28.9%, impersonal voice at 26.5%.
Quantified results in the abstract
| Feature | Share | Grand (with) | Grand (without) | Lift (pts) | p |
|---|---|---|---|---|---|
| Stated gap (however, current methods lack) | 31.5% | 30.3% | 25.6% | +4.7 | <0.001 |
| Comparison with existing method or baseline | 9.9% | 39.2% | 25.7% | +13.5 | <0.001 |
| Named metric (accuracy, AUC, F1, RMSE) | 24.1% | 33.9% | 24.9% | +9.0 | <0.001 |
| p-value reported | 6.2% | 34.5% | 26.6% | +8.0 | <0.001 |
| Sample size reported | 5.6% | 29.8% | 26.9% | +3.0 | 0.165 |
| Replication mentioned | 2.2% | 31.1% | 27.0% | +4.2 | 0.215 |
| Limitations or future work | 12.0% | 25.1% | 27.3% | -2.2 | 0.147 |
| Cost or affordability claim | 16.0% | 39.2% | 24.7% | +14.4 | <0.001 |
| Application or scalability claim | 19.8% | 30.9% | 26.1% | +4.8 | <0.001 |
| Novelty claim (novel, first to) | 21.2% | 45.2% | 22.2% | +23.0 | <0.001 |
| First person plural (we developed) | 6.6% | 28.9% | 26.9% | +2.0 | 0.320 |
| First person singular (I developed) | 7.0% | 34.6% | 26.5% | +8.1 | <0.001 |
The ladder: more features, higher rate
Eight features counted: stated gap, baseline comparison, named metric, sample size, cost claim, application claim, novelty claim, three or more quantified results. Zero features 17.2%, one 22.7%, two 30.2%, three 38.0%, four 44.7%, five 58.1%. Judges do not award from the abstract alone, but these features signal that the research itself was designed that way.
Number of paper-style features in the abstract (0 to 6)
Thin lines mark the 95% confidence interval; smaller samples give longer lines.
| Features | Projects | Grand rate (95% CI) |
|---|---|---|
| 0 | 2025 | 17.2% (95% CI 16 to 19) |
| 1 | 2653 | 22.7% (95% CI 21 to 24) |
| 2 | 1843 | 30.2% (95% CI 28 to 32) |
| 3 | 1043 | 38.0% (95% CI 35 to 41) |
| 4 | 436 | 44.7% (95% CI 40 to 49) |
| 5 | 167 | 58.1% (95% CI 51 to 65) |
| 6 | 36 | 55.6% (95% CI 40 to 70) |
Title patterns
Titles that name a system (SmartBlink: ...) are 11.2% of entries and won at 41.0%; titles containing novel won at 41.6%; titles with a colon at 33.8%; titles stating the method with using, via or with at 31.3%. Titles shaped as The Effect of X on Y or starting with Does or Can (6.8%) won at 12.7%, and titles with a question mark (2.4%) at 11.6%. By length: 1 to 6 words 10.3%, 7 to 10 words 23.1%, 11 to 14 words 28.7%, 15 or more 32.2%.
Words in the title
| Title feature | Share | Grand (with) | Grand (without) | p |
|---|---|---|---|---|
| Colon (system name or subtitle) | 32.0% | 33.8% | 23.9% | <0.001 |
| Question mark | 2.4% | 11.6% | 27.4% | <0.001 |
| Named system or acronym | 11.2% | 41.0% | 25.3% | <0.001 |
| Contains novel | 11.9% | 41.6% | 25.1% | <0.001 |
| using / via / with / through | 28.1% | 31.3% | 25.4% | <0.001 |
| Effect of X on Y, Does, Can, How | 6.8% | 12.7% | 28.1% | <0.001 |
| Development of, Design of | 2.6% | 30.2% | 27.0% | 0.287 |
Winner vocabulary, non-winner vocabulary
Words over-represented in First and Second Award abstracts (515) versus no-award abstracts (5,030): custom, billion, robust, pipeline, novel, presents, architecture, millions, algorithm, annually, enables, cost, scalable, low-cost, validated, existing, rapid, datasets, optimized, framework, achieved, worldwide. Over-represented in no-award abstracts: took, plates, agar, survey, hypothesis, divided, experiment, valuable, purpose, should, amount, what, concluded, correlation, school. One side presents a system and its validated performance; the other narrates a procedure.
The 2026 to 2027 ISEF rules prohibit using generative AI to write the research plan, abstract, poster or references. Read the features above as a checklist of what the research must contain when the student writes it up. Abstract length was 246 words at the median for every grand tier and 243 for non-winners.
Show the full table (30 rows)
| Word | In First/Second abstracts (515) | In no-award abstracts (5,030) | Log odds |
|---|---|---|---|
| custom | 40 | 92 | 1.435 |
| billion | 34 | 117 | 1.035 |
| robust | 31 | 107 | 1.031 |
| pipeline | 25 | 95 | 0.934 |
| novel | 228 | 935 | 0.867 |
| presents | 45 | 189 | 0.839 |
| architecture | 27 | 117 | 0.804 |
| millions | 26 | 113 | 0.801 |
| algorithm | 66 | 297 | 0.772 |
| algorithms | 43 | 196 | 0.757 |
| annually | 28 | 128 | 0.751 |
| enables | 29 | 134 | 0.741 |
| cost | 75 | 351 | 0.733 |
| scalable | 29 | 136 | 0.726 |
| low-cost | 43 | 206 | 0.707 |
| validated | 31 | 150 | 0.696 |
| targets | 25 | 121 | 0.694 |
| expensive | 39 | 190 | 0.69 |
| computational | 54 | 266 | 0.681 |
| capture | 26 | 128 | 0.677 |
| vitro | 27 | 133 | 0.677 |
| existing | 65 | 323 | 0.673 |
| rapid | 37 | 184 | 0.67 |
| regions | 38 | 190 | 0.664 |
| motion | 25 | 125 | 0.662 |
| million | 72 | 362 | 0.661 |
| datasets | 35 | 176 | 0.658 |
| highly | 55 | 279 | 0.652 |
| constructed | 31 | 157 | 0.65 |
| optimized | 36 | 184 | 0.642 |
Frequently asked questions
Data and method
Two public sources. From the Society for Science abstract database (abstracts.societyforscience.org) we compiled every ISEF finalist project from 2014 to 2026: 17,565 projects with title, category, fair country, awards and full abstract. From the isef.net project library we took the award-tier flags for the 8,134 ISEF projects of 2021 to 2026 and joined them to the abstracts by year and booth id. Compiled on 2026-10-11.
A grand award means a First to Fourth Award or Best of Category, roughly the top 27% of each category. A special award is given by a sponsoring organization with its own judges. ISEF 2020 was held without judging and is excluded from every rate. The abstract database carries no 2025 awards and misses most Materials Science grand awards after 2017, so award flags for 2021 to 2026 combine both sources. An AI mention means the title or abstract contains a machine-learning term; it does not mean the project is about AI. All numbers are correlations, and Korean per-category figures rest on 2 to 24 projects, so read them as direction rather than precision.
Sources
- https://abstracts.societyforscience.org/
- https://isef.net/search-results/projects
- https://www.societyforscience.org/isef/international-rules/
The figures on this page are category-level base rates compiled from public ISEF results. They do not predict or promise any student's award, selection or admission. CIT does not guarantee awards, selection or admission.
ISEF data series
Want to pressure-test a project idea against this data?
Tell us the student's grade, field of interest and target fair. We will map a realistic preparation order against these numbers.