Conference Abstract Review

ReviewerZero's Conference Abstract Review integrates with conference submission systems to provide automated integrity screening for abstracts, extended abstracts, and full submissions. Designed for platforms like CTI Meeting Technology's cOASIS, this feature helps program committees identify potential issues at scale.

Integration Overview

Conference Abstract Review adds AI-powered screening to your submission workflow:

Automated screening - Process submissions as they arrive
Scalable analysis - Handle thousands of abstracts efficiently
Structured output - Machine-readable results for your platform
Flexible deployment - Works with text submissions and PDF uploads

Review Types

Three review modes support different submission formats and processing requirements.

Abstract Review

For standard conference abstracts (150-500 words). Evaluates:

Research objective clarity
Methodology appropriateness
Significance of findings
Writing quality
Venue fit (when conference context is provided)

Provides thorough analysis with detailed feedback.

Figures and tables that belong to the abstract can be sent with it (graphics), and the review reads them as part of the submission. When the text refers to a figure or table that was not sent, the review treats the results the text attributes to it as reported.

Extended Abstract Review

For multi-page submissions (2-6 pages) submitted as PDFs. Evaluates:

Research question clarity and significance
Appropriateness of the proposed approach
Potential contribution to the field
Feasibility of the research plan
Presentation quality within page constraints

Calibrated for preliminary research—does not penalize for incomplete methodology or limited results typical of extended abstracts.

Flash Abstract Review

For real-time feedback during the submission process. Provides:

Immediate response for interactive author feedback
Focus on critical issues only
Concise, actionable output

Ideal for giving authors instant feedback as they complete their submission, improving submission quality before final submission.

Comparison

Feature	Abstract Review	Extended Abstract Review	Flash Abstract Review
Input format	Text and optional graphics (JSON)	PDF file	Text (JSON)
Typical length	150-500 words	2-6 pages	150-500 words
AI text detection	Yes	Yes	No
Plagiarism detection	Yes	Yes	No
Figure/image integrity checks	No	Yes	No
Literature search	Yes (full)	Yes (full)	Yes (lite)
Caller-supplied guidelines (`venue_guidelines`)	Yes	Yes	Yes
Response type	Synchronous	Asynchronous	Synchronous
Response time	~30 seconds	Background processing	~15 seconds
Use case	Final screening	Document review	Real-time author feedback

Response Structure

All review types return structured JSON with consistent fields for easy integration:

Summary

Brief summary of the submission in the reviewer's interpretation—useful for quick triage.

Issues

Problems categorized by priority:

Priority	Description
High	Critical issues that may warrant rejection
Medium	Significant concerns to flag for reviewers
Low	Minor improvements

Each issue includes:

title - Short issue identifier
description - Detailed explanation
impact - Why this matters
solution - Suggested fix
priority - High, medium, or low

Strengths

Array of key positive aspects identified in the submission.

Additional Comments

Supplementary suggestions beyond issue identification.

Venue Fit

When conference context is provided, includes assessment of how well the submission matches the venue scope and audience:

is_good_fit - Boolean indicating overall fit
fit_score - Score from 0-100
venue_topics - Topics covered by the venue
matching_topics - Topics from the abstract that match
missing_elements - Elements the abstract is missing for this venue
fit_explanation - Detailed explanation

Returns null if no venue is specified.

Novelty Assessment

Judged against the closely related works the literature search retrieved for the abstract (see Related Literature). The search issues several targeted queries built from the abstract's problem, method, and population, keeps only papers a relevance pass considers closely related, and a model then compares the abstract's stated contribution against those specific papers.

novelty_score - 0 to 100, on an anchored scale:

Band	Meaning
0-20	A retrieved work already reports the same finding with the same method and population
21-40	An incremental variation of retrieved work; the core claim is already established
41-60	A new angle (population, comparison, outcome, or mechanism) on a problem the retrieved works cover
61-80	A core claim the retrieved works do not cover: a new method, mechanism, or first controlled evidence
81-100	Nothing retrieved overlaps the core claim and the abstract states specifically what is new

When no related work is found, the score comes from the abstract's own statement of its contribution and stays within 50-85 unless the abstract says specifically what is new. Absence of retrieved work is weaker evidence than presence: it can mean a novel topic or a niche one that is poorly indexed, and the explanation says so. The body is judged, not the title: text that describes no study of its own (background or definitions only) scores 0-20 with impact_potential low.

impact_potential - high when the result, if it holds, would change practice or understanding for a broad group; medium when it advances a specific line of work or a defined population; low when it confirms known results, is descriptive without a comparison, or is narrow in scope.
novelty_explanation - Two to four sentences for a program committee reader, naming the works compared against by first author and year, what overlaps, what the abstract adds, and the basis for the score.

If the assessment itself cannot run, novelty_explanation says so and novelty_score and impact_potential carry the defaults 50 and medium; treat those as "not assessed", not as a judgment.

Array of related works found via literature search. It is always in the response: one entry per work that novelty_explanation counts, and an empty array when no closely related work was found. Each entry includes:

title - Paper title
authors - Author names (omitted when the source has none)
year - Publication year
venue - Journal or conference name (if available)
url - DOI link to the paper (https://doi.org/...), or null when none is known
relevance - Why this work is relevant (Flash Abstract Review only)

AI Text Detection (Full review only)

Detection of AI-generated content:

ai_probability - Probability (0-1) that text is AI-generated
classification - "human", "ai", or "mixed"
headline - Summary of the detection result

Returns null when the check could not run. Treat null as "not available", not as "human-written": a negative result is an object with classification set to "human". Not included in Flash Abstract Review.

This check is the only source of AI-generation verdicts in the response. The reviewer is given its result and does not form its own; an issues entry about AI-generated text appears only when the check classified the text as ai or mixed and the venue's guidelines address AI use.

Plagiarism Detection (Full review only)

Detection of potential plagiarism:

plagiarism_detected - Boolean indicating if plagiarism was found
percent_plagiarized - Percentage of text that appears plagiarized
matched_sources - Array of matched source URLs and similarity scores

Returns null when the check could not run. Treat null as "not available", not as "nothing found": a negative result is an object with plagiarism_detected set to false. Not included in Flash Abstract Review.

This check is the only source of plagiarism verdicts in the response, so plagiarism_detection and issues always agree: when it finds matching sources, issues contains one high-priority entry reporting the match with the check's figures and sources; when it finds none, no issue asserts plagiarism. If the reviewer nevertheless remarks that the text reads as copied, that remark is labeled as the reviewer's impression, not a detected match, and carries at most medium priority.

Input Parameters

Abstract Review & Flash Abstract Review

Parameter	Required	Description
`abstract`	Yes	The submission text (min 50 characters)
`title`	No	Submission title
`authors`	No	Author list with names and affiliations
`venue`	No	Target conference or journal name for venue fit assessment. Looked up on the web. Mutually exclusive with `venue_guidelines`
`venue_guidelines`	No	The venue's guidelines as Markdown (up to 50,000 characters). The review checks the abstract against exactly these rules and reports each violation as an issue titled with the rule it breaks; no web lookup. Mutually exclusive with `venue`
`graphics`	No	Abstract Review only; Flash Abstract Review rejects it with `400`. Up to 5 figures or tables that belong to the abstract, read by the review as part of the submission. Each item has `name` (required, the name the abstract uses, e.g. "Figure 1") and `image` (required, the file's content as base64, not a URL; PNG, JPEG or WebP, up to 3 MB and 40 megapixels)

{
  "abstract": "... The arm means are shown in Figure 1. ...",
  "graphics": [
    { "name": "Figure 1", "image": "iVBORw0KGgoAAAANSUhEUgAA..." }
  ]
}

Extended Abstract Review

Parameter	Required	Description
`file`	Yes	PDF file of the extended abstract (max 6 pages)
`venue`	No	Target conference or journal name for venue fit assessment. Mutually exclusive with `venue_guidelines`
`venue_guidelines`	No	The conference's guidelines as Markdown (up to 50,000 characters). The guidelines check runs against exactly these rules instead of searching the web. Mutually exclusive with `venue`
`public_view`	No	If true, make the analysis publicly accessible and return URL
`notify_email`	No	If true, send email notification when analysis completes

Integration Patterns

Submission-Time Screening

Run Flash Abstract Review as authors submit to provide immediate feedback, allowing them to improve their abstract before final submission.

Batch Processing

Process accepted submissions through Abstract Review or Extended Abstract Review to flag potential issues for program committee review.

Reviewer Support

Include ReviewerZero analysis in reviewer materials to help reviewers focus on content quality rather than formatting or basic issues.

Availability Monitoring

Add GET https://www.reviewerzero.ai/api/v1/status to your monitoring system to detect service degradation proactively. It requires no authentication and returns 200 when the service is operational or 503 when it is degraded. See Service Status for the response format.

Venue Context

Providing venue information improves review quality:

With venue context - Tailored feedback on scope fit and audience alignment
Without venue context - General academic quality assessment

Venue context is either a venue name (venue), which is looked up on the web, or the venue's own guidelines (venue_guidelines), sent as Markdown with each request. There is no stored guidelines registry: an integration that reviews many submissions for the same event sends the same guidelines text every time, which yields the same rule set on every run.

AI Review - Full manuscript review for journal submissions
Guideline Compliance - Check venue-specific formatting requirements
Integrations - Connect with cOASIS and other submission systems

Conference Abstract Review

On this page