ClaimBuster

FreeProfessional

Publisher: idir.uta.edu | Type: Non-detector | Cost: Free

TL;DR

Long-running academic claim-extraction baseline from UT Arlington's IDIR lab that scores English-language sentences for fact-check-worthiness, with a free API, browser plugin, and Slack integration. The toolkit's English-language reference point in 2B.2, useful for regional English-language press, diaspora social media, and transnational claims; for SEA-language content the operational paths sit elsewhere in the cell.

What it does

ClaimBuster ingests text and returns a check-worthiness score per sentence: a continuous score that ranks how likely the sentence is to contain a verifiable factual claim that warrants fact-checker attention. The model is an academic baseline that has been maintained over multiple years through the Duke Reporters' Lab Tech & Check Cooperative, with documented use in political-debate monitoring (US presidential debates and similar high-volume English-language transcript processing).

The tool's value to a regional fact-check operation is not SEA-language coverage (explicitly outside its scope) but the role it plays as an English-baseline reference point. Where a fact-checker has English-language input that does not warrant the heavier integration of a Check-style coalition deployment (a regional English-language press article, a diaspora-focused social-media stream, a transnational claim appearing in English regional news), ClaimBuster is the maintained academic option. UT Arlington runs the infrastructure; the Duke Reporters' Lab pairs the tool with broader fact-check operationalisation.

When to use it

• An English-language regional press article needs systematic check-worthiness scoring before reporter time is allocated to verification of specific claims.
• A regional research project needs a free academic baseline for English-language claim ranking on transnational content appearing in regional English-language outlets.
• A diaspora-social-media monitoring workflow needs lightweight English-language claim triage; the Chrome plugin or Slack integration is operationally lighter than standing up an Alegre or Check deployment.
• A trainer running an English-language fact-check workshop wants a free, documented academic tool to demonstrate claim-extraction concepts before introducing the multilingual paths through Alegre and country-specific entries.

Limitations

Limitations

• Multilingual claim detection remains uneven. The tool is primarily English; non-English content should not be expected to receive the same quality of check-worthiness scoring.
• English-only operational language. Bahasa Indonesia, Malay, Filipino / Tagalog, Thai, Sinhala, Tamil, and Lao are not covered; for SEA-language content the operational paths are Dissect (Sinhala), Yudistira (Bahasa), Alegre (multilingual XLM-R), and X-CLAIM (Tamil pan-Indian).
• Returns check-worthiness scores, not verified fact-checks; a high score is a flag for human attention, not evidence the claim is false.
• Documented use is primarily political-debate monitoring in the United States and similar English-speaking contexts; SEA-region documented use has not been surfaced in the source set. Validation on regional English-language content is the operator's responsibility.

Privacy and threat model

ClaimBuster runs as a free API and web service; queries travel to UT Arlington's infrastructure for processing. The threat model at the user-facing layer is low for typical use; the inputs are typically already-published English-language transcripts or articles, not source-identifying material. The research-academic operator identity (UT Arlington IDIR, Duke Tech & Check) is institutionally stable and outside any of the six focus countries' regulatory environments.

For institutional partners considering Slack or browser-plugin integration, the same general discipline applies as to other cloud-hosted tools: classify what is being submitted, do not route source-identifying material through public APIs, and prefer the bring-your-own-environment paths (other entries in 2B.2 that are self-hostable) when the threat model warrants.

Country and platform applicability

Decision 7 framing applies: ClaimBuster is an NLP-class tool, and content-language coverage is the operational selection criterion. ClaimBuster is documented as English-only at operational scale; the b2-disclaimer-pass-English-only flag is the explicit acknowledgement that this tool enters the shortlist with a content-language disclaimer rather than through a SEA-language pass.

• Indonesia: not applicable for Bahasa Indonesia content. English-language regional press or diaspora English-language streams about Indonesian topics may be processed; the bilingual context is the validation responsibility of the operator.
• Laos: not applicable. Lao is not covered; the 2B.2 Lao gap is structural across all NLP-class entries in this cell. The Lao path remains Google Cloud Translation routing (2A.1 cross-link) plus partner-mediated review.
• Malaysia: not applicable for Malay. English- language Malaysian regional press may be processed; the recommended primary paths for Malaysian content are Alegre and the 2B.3 LLM layer (MaLLaM).
• Philippines: not applicable for Filipino / Tagalog. English-language Philippine press is processable — Philippines has substantial English-language press output — with the validation caveat above.
• Sri Lanka: not applicable for Sinhala or Tamil. English-language Sri Lankan press (FactCheck.lk's English-facing archive, English-language Sri Lankan news outlets) is processable with the validation caveat.
• Thailand: not applicable for Thai. English-language Thai press is processable with the validation caveat.

Platform applicability: not platform-specific. ClaimBuster operates on text input across web, API, Chrome plugin, and Slack integrations; the typical use is on transcripts, articles, or text streams the operator already has in machine-readable form.

How to access

Free at idir.uta.edu/claimbuster and via the documented API endpoints. The Chrome plugin is available from the Chrome Web Store; the Slack integration is documented on the IDIR / Duke Reporters' Lab pages. Account creation is required for some API tiers but not for the basic web-form scoring; institutional partners considering high-volume programmatic use should review the IDIR documentation for current rate limits.

Cost (current as of 2026-05)

Free under the academic-open-source model. The infrastructure is run by UT Arlington's IDIR lab with Duke Reporters' Lab Tech & Check Cooperative support; sustainability is on the maintainers, not on the user.

Quickstart

1. Open idir.uta.edu/claimbuster or install the Chrome plugin from the Chrome Web Store.
2. Submit English-language text (a transcript, an article, a social-media post) to the scoring endpoint.
3. Review the per-sentence check-worthiness scores; high-score sentences are flagged for fact-checker attention rather than auto-classified as false.
4. Carry the priority list into the verification workflow: source triangulation, cross-check against existing fact-checks via Google Fact Check Explorer (1B.5), image / video forensics through 1B tools where applicable.
5. For SEA-language content, do not attempt to use ClaimBuster as a primary tool; route to the appropriate 2B.2 entry (Dissect, Yudistira, Alegre, X-CLAIM) by content language.
6. For institutional partners considering integration into a newsroom workflow, evaluate the Chrome plugin and Slack integration against the heavier Alegre / Check pipeline; ClaimBuster is operationally lighter but English-only.

In the toolkit's workflow

Source-history pillar non-detector tool per Architectural Anchor 1. Sits in 2B.2 AI claim extraction as the English- baseline reference point alongside the SEA-language operational primaries: Sri-Lanka-specific Dissect, multilingual Alegre, Bahasa-specific Yudistira, Tamil pan-Indian X-CLAIM. ClaimBuster's role in the cell is the maintained academic baseline: "if the input is English, ClaimBuster is the documented academic option; for SEA-language content, route through the language-appropriate entry."

Standard combinations:

• With Dissect at 2B.2 – Sri-Lanka-specific Sinhala primary versus ClaimBuster's English baseline; the two cards illustrate the polarity of the 2B.2 shortlist: one locally tuned and language-specific, one academic and English-only, with multilingual Alegre between them.
• With Meedan Alegre at 2B.2 – multilingual XLM-R covers SEA languages; ClaimBuster is the English reference point. A coalition handling both English regional press and multilingual tipline content runs both rather than choosing.
• With Google Fact Check Explorer at 1B.5 – for cross-checking high-score ClaimBuster claims against existing fact-checks; English-language ClaimReview-indexed output is well covered by Google's Fact Check Explorer.
• With InVID-WeVerify at 1B.1 when the English-language content includes images or video that need forensic verification before the verification proceeds.

Decision-tree references: T5 escalation names ClaimBuster at T5.9 (text professional) as the academic baseline option for English-language text-input claim triage; SEA- language inputs route to the language-appropriate 2B.2 entry instead of ClaimBuster.

This card carries no detector signal class: ClaimBuster produces a non-detector source-history signal (the check-worthiness score per sentence). Per Anchor 2 the score is one signal class supporting human triage; the verification work that follows is the next signal class in the editorial workflow.

Override notes

Override notes

• English-only disclaimer pass (b2-disclaimer-pass-English-only): ClaimBuster enters the 2B.2 shortlist on the explicit understanding that it is English-only at operational scale. The card surfaces this in the TL;DR (regional English-language press, diaspora social media, transnational claims), in the Limitations admonition, and in the Country and platform applicability section's Decision 7 framing. SEA-language content routes to the appropriate language-specific entry in 2B.2 rather than to ClaimBuster.

Sources

• Hassan, N. et al. ClaimBuster: The First-Ever End-to-End Fact-Checking System. Proceedings of the VLDB Endowment, 2017. idir.uta.edu/claimbuster.
• UT Arlington IDIR Lab. ClaimBuster — automated claim detection. University of Texas at Arlington. idir.uta.edu/claimbuster.