IJCAI Writing Style
Use this when revising the main paper. IJCAI rewards clear, significant, original AI work
that a broad program committee can evaluate quickly.
Revision rules
- Put the AI contribution in the first page: problem, gap, core idea, evidence, and why it
matters beyond a narrow benchmark.
- Keep the submission self-contained; supplementary material should support, not rescue, the
main argument.
- Spend the 7-page body on claims reviewers can verify: formal setup, method, theorem or
algorithm, baselines, ablations, and limitations.
- Use third-person citation language for prior work by the authors and remove
acknowledgements, contribution statements, and identity-revealing phrasing from the review
version.
- Include an ethics or broader-impact statement when sensitive data, high-stakes deployment,
safety, discrimination, privacy, consent, copyright, or societal harm is plausible.
- Treat LLM assistance conservatively. Style polishing may be acceptable in some cycles, but
LLM-written manuscript text, fake citations, or unverified claims can trigger rejection.
Writing for a cross-subfield committee
IJCAI's PC mixes symbolic AI, search, planning, KR, constraint reasoning, multi-agent, game
theory, ML, NLP, and vision. Write so a reviewer outside your niche reaches the contribution
fast, because the out-of-area reviewer often drives summary-reject risk.
| Reader risk |
Symptom in the draft |
Page-discipline fix |
| Out-of-area reviewer lost early |
Contribution buried after a long formal preamble |
State problem, gap, and claim on page one |
| Notation overload |
New symbols before motivation |
Introduce notation only when the claim needs it |
| Niche jargon |
Subfield acronyms undefined |
Define on first use; assume a broad AI reader |
| Body overflow |
Method spills toward the reference pages |
Move proofs and grids to the appendix, keep claims in the body |
Worked vignette: framing a KR paper for breadth
A knowledge-representation paper opens with three paragraphs of description-logic preliminaries
before naming its contribution. A planning or learning reviewer may summary-reject for unclear
significance. The fix: open with the reasoning task, why current encodings fall short, and the
new tractable fragment, then defer the heavy formalism. The technical depth stays; only the
ordering changes so a broad IJCAI reader sees the payoff within two reads.
Reviewer pushback and the venue-specific fix
- "Could not find the contribution." Rewrite the first page as problem, gap, idea, evidence,
significance; this is the single highest-leverage IJCAI edit.
- "Reads like a journal extension." Compress to the body limit and cite the longer work in
third person rather than reproducing it.
- "Significance only within one benchmark." Add a sentence on why the result matters across AI,
not just on the chosen suite.
Output format
[Writing diagnosis] clear / overloaded / underspecified / policy-risky
[First-page fix] <new framing>
[Compression cuts] <what to move or delete>
[Anonymity edits] <phrases to rewrite>
[Ethics/reproducibility edits] <specific additions>
1---2name: ijcai-writing-style3description: Use when revising an IJCAI or IJCAI-ECAI paper for broad AI significance, 7-page-body compression, self-contained exposition, double-blind wording, optional ethics discussion, reproducibility clarity, and policy-safe use of LLM editing.4---56# IJCAI Writing Style78Use this when revising the main paper. IJCAI rewards clear, significant, original AI work9that a broad program committee can evaluate quickly.1011## Revision rules1213- Put the AI contribution in the first page: problem, gap, core idea, evidence, and why it14 matters beyond a narrow benchmark.15- Keep the submission self-contained; supplementary material should support, not rescue, the16 main argument.17- Spend the 7-page body on claims reviewers can verify: formal setup, method, theorem or18 algorithm, baselines, ablations, and limitations.19- Use third-person citation language for prior work by the authors and remove20 acknowledgements, contribution statements, and identity-revealing phrasing from the review21 version.22- Include an ethics or broader-impact statement when sensitive data, high-stakes deployment,23 safety, discrimination, privacy, consent, copyright, or societal harm is plausible.24- Treat LLM assistance conservatively. Style polishing may be acceptable in some cycles, but25 LLM-written manuscript text, fake citations, or unverified claims can trigger rejection.2627## Writing for a cross-subfield committee2829IJCAI's PC mixes symbolic AI, search, planning, KR, constraint reasoning, multi-agent, game30theory, ML, NLP, and vision. Write so a reviewer outside your niche reaches the contribution31fast, because the out-of-area reviewer often drives summary-reject risk.3233| Reader risk | Symptom in the draft | Page-discipline fix |34| --- | --- | --- |35| Out-of-area reviewer lost early | Contribution buried after a long formal preamble | State problem, gap, and claim on page one |36| Notation overload | New symbols before motivation | Introduce notation only when the claim needs it |37| Niche jargon | Subfield acronyms undefined | Define on first use; assume a broad AI reader |38| Body overflow | Method spills toward the reference pages | Move proofs and grids to the appendix, keep claims in the body |3940## Worked vignette: framing a KR paper for breadth4142A knowledge-representation paper opens with three paragraphs of description-logic preliminaries43before naming its contribution. A planning or learning reviewer may summary-reject for unclear44significance. The fix: open with the reasoning task, why current encodings fall short, and the45new tractable fragment, then defer the heavy formalism. The technical depth stays; only the46ordering changes so a broad IJCAI reader sees the payoff within two reads.4748## Reviewer pushback and the venue-specific fix4950- "Could not find the contribution." Rewrite the first page as problem, gap, idea, evidence,51 significance; this is the single highest-leverage IJCAI edit.52- "Reads like a journal extension." Compress to the body limit and cite the longer work in53 third person rather than reproducing it.54- "Significance only within one benchmark." Add a sentence on why the result matters across AI,55 not just on the chosen suite.5657## Output format5859```text60[Writing diagnosis] clear / overloaded / underspecified / policy-risky61[First-page fix] <new framing>62[Compression cuts] <what to move or delete>63[Anonymity edits] <phrases to rewrite>64[Ethics/reproducibility edits] <specific additions>65```
brycewang-stanford