AAAI Writing Style
Use this to make a technically sound draft readable to a broad AI program committee. AAAI rewards
clear AI contribution, not only subfield-specific benchmark wins.
AAAI framing
- State the AI problem, the new capability or insight, and the evidence in the first page.
- Make clear whether the contribution is method, theory, system, benchmark, dataset, evaluation,
human-AI interaction, social impact, or alignment.
- Explain why the result matters outside a single dataset or implementation.
- Keep claims aligned with the reproducibility checklist and supplementary evidence.
- Discuss limitations and ethical considerations when the method affects people, safety, privacy,
fairness, security, or social impact.
Two-column readability
- Use figures and tables as decision aids, not decoration.
- Keep notation lightweight and define it near first use.
- Use compact related-work contrasts instead of long literature catalogues.
- Make the Phase 1 summary easy: problem, method, evidence, limitation, checklist compliance.
- Avoid unsupported "general intelligence", "human-level", or "safe" claims.
Reading the paper as a Phase-1 reviewer
A non-specialist on AAAI's broad committee gives the first page a few minutes and decides whether the
paper is worth deeper reading. Write so that a planning or KR reviewer can summarize your learning
contribution, and vice versa. Audit the opening against what that reader needs to extract fast.
| First-page question |
Reviewer extracts |
Failure symptom |
| What is the problem |
one AI task statement |
jargon with no anchor |
| What is new |
the single contribution |
a list, no headline |
| Why believe it |
evidence in one line |
"see Section 6" only |
| What are the limits |
scope and caveat |
silence or overclaim |
Phrasing fixes that survive AAAI review
- Replace "we achieve state of the art" with the specific delta and the setting it holds in.
- Replace bare "safe" or "human-level" with a measured, scoped statement the evidence supports.
- Replace a long related-work catalogue with two or three sharp contrasts a non-specialist can follow.
- Tie every strong claim back to a checklist answer so rigor and prose agree.
Worked vignette
A multi-agent paper opens with three paragraphs of game-theory notation before naming its
contribution. A vision reviewer cannot find the AI claim and is likely to stop. The fix rewrites
sentence one as the problem, sentence two as the new coordination mechanism, sentence three as the
one-line evidence, and a final clause scoping the result to the studied setting, all on page one.
Output format
[AAAI fit sentence] <one sentence>
[Contribution type] method / theory / system / benchmark / dataset / evaluation / social impact
[First-page fixes] <problem, method, evidence, limitation>
[Checklist alignment] pass / needs revision
[Overclaim risks] <claims to narrow or evidence to add>
1---2name: aaai-writing-style3description: Use when revising an AAAI manuscript for broad-AI-audience fit, a first-page contribution statement legible to non-specialist Phase-1 reviewers, two-column readability, concise novelty claims, reproducibility-checklist alignment, hedged limitations and ethics, and policy-aware framing of AI-system and capability claims.4metadata:5 source: https://skillmd.com/skills/brycewang-stanford/aaai-writing-style6---78# AAAI Writing Style910Use this to make a technically sound draft readable to a broad AI program committee. AAAI rewards11clear AI contribution, not only subfield-specific benchmark wins.1213## AAAI framing1415- State the AI problem, the new capability or insight, and the evidence in the first page.16- Make clear whether the contribution is method, theory, system, benchmark, dataset, evaluation,17 human-AI interaction, social impact, or alignment.18- Explain why the result matters outside a single dataset or implementation.19- Keep claims aligned with the reproducibility checklist and supplementary evidence.20- Discuss limitations and ethical considerations when the method affects people, safety, privacy,21 fairness, security, or social impact.2223## Two-column readability2425- Use figures and tables as decision aids, not decoration.26- Keep notation lightweight and define it near first use.27- Use compact related-work contrasts instead of long literature catalogues.28- Make the Phase 1 summary easy: problem, method, evidence, limitation, checklist compliance.29- Avoid unsupported "general intelligence", "human-level", or "safe" claims.3031## Reading the paper as a Phase-1 reviewer3233A non-specialist on AAAI's broad committee gives the first page a few minutes and decides whether the34paper is worth deeper reading. Write so that a planning or KR reviewer can summarize your learning35contribution, and vice versa. Audit the opening against what that reader needs to extract fast.3637| First-page question | Reviewer extracts | Failure symptom |38| --- | --- | --- |39| What is the problem | one AI task statement | jargon with no anchor |40| What is new | the single contribution | a list, no headline |41| Why believe it | evidence in one line | "see Section 6" only |42| What are the limits | scope and caveat | silence or overclaim |4344## Phrasing fixes that survive AAAI review4546- Replace "we achieve state of the art" with the specific delta and the setting it holds in.47- Replace bare "safe" or "human-level" with a measured, scoped statement the evidence supports.48- Replace a long related-work catalogue with two or three sharp contrasts a non-specialist can follow.49- Tie every strong claim back to a checklist answer so rigor and prose agree.5051## Worked vignette5253A multi-agent paper opens with three paragraphs of game-theory notation before naming its54contribution. A vision reviewer cannot find the AI claim and is likely to stop. The fix rewrites55sentence one as the problem, sentence two as the new coordination mechanism, sentence three as the56one-line evidence, and a final clause scoping the result to the studied setting, all on page one.5758## Output format5960```text61[AAAI fit sentence] <one sentence>62[Contribution type] method / theory / system / benchmark / dataset / evaluation / social impact63[First-page fixes] <problem, method, evidence, limitation>64[Checklist alignment] pass / needs revision65[Overclaim risks] <claims to narrow or evidence to add>66```67