Content — moves your score
- Novelty3/5
- Methodology4/5
- Results4/5
- Baseline2/5
- Data collection3/5
Hygiene — can only cost you
- Ethicsok
- Coherenceok
- Citation2/5
- Figure/Tableok
- Grammarok
- Reproducibilityok
Illustrative — sample scores for a sample manuscript.
Reviewed the way a program committee actually reads it.
Most review tools score every category the same — a spotless grammar check inflates your number exactly as much as genuine methodological rigor would. That's backwards. Five categories a committee actually judges you on — novelty, methodology, results, baseline comparisons, data collection — can move your score up or down. The other six can only ever cost you points, never inflate them. Clean grammar and a well-labeled figure are necessary. They're not a differentiator.
We compare against three prior retrieval baselines from 2021–2022.
Add at least one 2024 baseline — dense hybrid retrieval has moved past sparse methods since these were published.
None of the three cited baselines postdate 2022; a program committee reviewing a 2026 submission will expect something more recent.
We recruited 42 participants via university mailing lists.
State the demographic breakdown and compensation — reviewers will ask if it isn't in the paper.
Sample composition and incentive structure are standard questions for a user study; leaving them out invites a clarification request during rebuttal.
Illustrative — sample comments for a sample manuscript.
Not a black-box number — a specific comment on your specific text.
That Baseline score of 2/5 above isn't generic. It's this: your cited baselines are three years old, and here's exactly what to add. Every comment quotes your actual paragraph, tells you precisely what's missing or wrong, and — where the fix doesn't change your text so much as add to it — leaves the original untouched rather than pretending it needs to be rewritten.
Checked against what your figure actually shows — not just its caption.
When you describe a figure in your text, we send that exact figure image alongside your claim about it, so “Figure 4 shows a clear separation between clusters” is checked against the real plot — not assumed true because the caption says so. We also catch what a caption can't: a table or figure sitting in your paper that the body text never actually references.
Figure 4 shows a clear separation between clusters, supporting our claim that the learned embeddings capture task-relevant structure.
Soften or revise this claim — the actual clusters in Figure 4 visibly overlap in the bottom-left region.
Checked the referenced image directly, not just its caption. A reviewer who looks at the same figure will notice the same overlap.
Illustrative — sample comment for a sample manuscript.
Every reference, checked before anyone reads your draft.
LLM-assisted writing has a documented failure mode: plausible-looking references that don't actually exist. Before your review even runs, every citation in your bibliography is checked against real academic records — and cross-checked for retractions. When something can't be confirmed, we say “couldn't verify,” never “fabricated” — deliberately conservative, so you're never accused of something that's just a gap in an index.
Reference check — 4 of 31
- Vaswani et al., “Attention Is All You Need” (2017)Verified · Crossref
- Okafor & Lindqvist, “Sparse Retrieval at Scale” (2022)Verified · OpenAlex
- Chen & Ibarra, “Joint Encoders for Reasoning” (2023)Couldn't verify
- Patel, “Benchmarking Transfer Methods” (2019)Retracted
Illustrative — sample bibliography check.
Dr. M. Alvarado
Universidad de Chile — Dept. of Computer Science
First or last author on 3 closely related papers
R. Okonkwo
Affiliation unavailable
First or last author on 1 closely related paper
Illustrative — sample suggestions for a sample manuscript.
Reviewers who work in your subfield — not invented names.
Every candidate is found by live search across Semantic Scholar, OpenAlex, Crossref, and CORE — never a name an LLM recalls from training data with no way to check it's real. We rank by first- and last-authorship on genuinely related work, the two positions an actual program committee would think to invite. When we can't find an affiliation or a profile, we say so instead of guessing.
See exactly how your draft stacks up at the venue you're targeting.
We pull real, recently accepted papers from your target venue and compare your manuscript against them on methodology, results, and contribution — weighted toward what that venue is publishing now, not its most-cited papers from a decade ago. Every candidate paper is checked against the venue's actual publisher first, so a same-acronym collision never sneaks in.
“Shared Editing Signals in Distributed Teams” — CSCW 2025
Similar methodology (mixed-methods field deployment); your sample size is smaller — worth addressing in limitations.
“Trust Repair After Automation Failures” — CSCW 2025
Comparable contribution framing; this paper's baseline set is more recent than yours.
Illustrative — sample comparison for a sample manuscript.
Baseline comment — draft_v3.pdf2 replies
Do we have room to add the 2024 baseline it's asking for, or should we push back on this one in the rebuttal instead?
@Priya Shahyes — I'll rerun it against our test split, should have numbers by Thursday.
Illustrative — sample thread on a sample manuscript.
Discuss the actual finding, not a screenshot of it in Slack.
Invite co-authors to a project and they see the same review, in the same place. Every comment is threaded directly under the specific comment it's about — reply, @mention a collaborator, or resolve it once it's addressed — so the conversation stays attached to the finding it's actually about, not scattered across email.
Running this a lot? Bring your own key.
Same deep, search-grounded review as Pro — but your own API key pays for the tokens, not us, so there's no ceiling on how often you regenerate. Priced lower than Pro for exactly that reason. See pricing.
Rejected? Don't start from a blank page — or from zero.
Triage prior reviewer feedback, pick a new target venue informed by real comparison data, and revise with guidance built for a stronger resubmission. Every version you upload is scored, so the chart moves with your actual revisions — not a rewrite from scratch, and not a guess at whether it's working.
Overall score by version
Tracks score across each uploaded manuscript revision.
Illustrative — sample score history for a sample manuscript.
How it works
§1
Upload your draft
PDF, DOCX, or a LaTeX project (.zip) — whatever stage your manuscript is at.
§2
Get section-by-section review
Weighted scoring, citation checks, and venue/reviewer matching grounded in real literature.
§3
Revise and submit
Upload a new version for fresh feedback, invite collaborators, and — if rejected — work the resubmission workflow.
Frequently asked questions
What file formats can I upload?
PDF, DOCX, and LaTeX projects (as a .zip).
Is the AI review a guarantee my paper will be accepted?
No — it's an automated aid (AI-generated feedback plus citation/venue checks against real academic databases) meant to catch issues early, not an editorial or publication guarantee. Always use your own judgment before submitting.
Can I work on a paper with collaborators?
Yes — Pro plans can invite collaborators to a project; see Pricing for the exact limits.
What happens to my uploaded manuscript?
It's stored for your account and any collaborators you add. See our Privacy Policy for what's collected and which third-party services (AI review, citation/venue search) it's sent to in order to power those features.
Do you email me things I didn't ask for?
Just a welcome email and whatever notifications you've opted into (like being tagged in a comment) — you can turn notification emails off anytime from your profile settings.
Start free. Upgrade when you need more.
More collaborators, deeper AI review, full citation verification — all on Pro.