AI-Powered Extraction
Parse references from text or PDFs into structured CSL-JSON using AnyStyle or your preferred LLM — including OpenAI, Anthropic, Google, and DeepSeek. Supports batch extraction of entire bibliographies.
The browser extension that catches AI-hallucinated references
Extract, search, and validate academic references in seconds. The Source Taster reliably distinguishes real sources from AI-fabricated ones — 93% APA match rate, 100% hallucination detection.
93%
Verification rate
100%
Hallucination detection
<3s
per reference
5+
Academic databases
AI hallucinates fake references
ChatGPT and other LLMs generate references that look correct but don't exist. Studies show up to 40% of AI-generated citations are fabricated — and traditional tools can't catch them.
- 62.8% of researchers struggle with time-consuming manual checks
- 87.8% want automatic warnings about non-existent sources
- 31% skip reference checks entirely due to lack of time
LLMs are known to generate fictional academic citations that appear genuine but do not exist in any academic database, posing significant risks to research integrity.
Everything you need to verify sources
From extraction to evidence — a complete verification pipeline in your browser.
5-Database Search
Simultaneously query OpenAlex, Crossref, Semantic Scholar, Europe PMC, and arXiv. Smart early termination finds high-confidence matches faster.
Transparent Scoring
Weighted field-by-field comparison with Levenshtein-Damerau distance. Every decision is transparent — you see which fields matched and by how much.
PDF Import
Drag-and-drop PDF files directly into the extension. Full text is parsed and all embedded references are extracted automatically.
Batch Verification
Check entire reference lists at once. Average processing time is under 3 seconds per source — including extraction, search, and scoring.
Privacy-First
Stateless extension with AES-256-GCM encrypted API keys stored in the backend. No telemetry, no analytics. Your data stays yours.
How it works
Four phases power every verification — fully automated, fully transparent.
1
Extract & normalize
Paste a reference or upload a PDF. AnyStyle or an LLM parses it into standardized CSL-JSON, normalizing names, dates, and formatting.
2
Find candidates
The system searches by DOI/ISBN first, then by title+author+year across five academic databases with intelligent early termination.
3
Match & score
Every field is compared using fuzzy string metrics. A weighted score aggregates the results — title and authors carry the most weight.
4
Classify & show evidence
References are classified as Exact Match, Strong Match, Possible Match, or No Match. Each result includes a detailed breakdown of why.
Built on evidence
Validated in a controlled study with curated datasets as part of a Master's thesis at the University of Siegen.
93%
Exact match rate (APA)
Average score of 99.11% across genuine APA references
100%
Hallucination detection
Every synthetic non-existent source was correctly identified
96.22%
Robustness score
Average score on perturbed references — resilient to formatting errors
<3s
Average time per reference
Full pipeline: extraction, candidate search, matching, and classification
APAMLAChicagoIEEEVancouverHarvardACSAMAOxfordSpringer
Technical details
Architecture, data sources, and compatibility at a glance.
Supported databases
- OpenAlex
- Crossref
- Semantic Scholar
- Europe PMC
- arXiv
Extraction methods
- AnyStyle (CRF-based, local)
- LLM-powered (OpenAI, Anthropic, Google, DeepSeek)
- PDF parsing (unpdf)
Compatibility
- Chrome, Edge, Brave, Opera
- Firefox
- Node.js 20+ (API backend)
Security
- AES-256-GCM encrypted API keys
- Stateless extension
- No telemetry
- Open source (MIT)
Simple, transparent pricing
The extension is free and open source forever. Upgrade to a cloud tier for convenience.
Free (Self-Host)
€0 /month
Full open-source extension + self-hosted API.
- MIT license — free forever
- Unlimited DOI-based verification
- Bring your own AI key (BYOK)
- 5 academic databases
- Self-host with Docker
- Community support (GitHub)
Most popular
Coming Soon
Pro
€9 /month
Hosted cloud API — no setup needed.
- Hosted API — no server required
- 1,000 AI extractions / month
- Priority processing
- All 5 academic databases
- Email support
- Automatic updates
Coming Soon
Team
€29 /month
For labs, research groups, and universities.
- Everything in Pro
- 5,000 AI extractions / month
- Team workspace & shared keys
- Usage dashboard
- Priority email support
- Early access to new features
Enterprise
Custom
For institutions with advanced requirements.
- Everything in Team
- Unlimited AI extractions
- SSO / SAML integration
- Audit logging
- On-premise deployment
- Dedicated support & SLA
Frequently asked questions
Everything you need to know before installing.
Is The Source Taster free?
Yes, the extension is free and open source under the MIT license — always. Self-host the API and bring your own AI key to use it completely free with no limits. A managed cloud version with included AI credits is available at a fair monthly price for those who prefer convenience.
Do I need an API key to use it?
Basic DOI-based verification works without any API key. To use LLM-powered extraction, you need a key from OpenAI, Anthropic, Google, or DeepSeek. The key is stored encrypted in our backend, never exposed to your browser.
Which browsers are supported?
The extension works on Chromium-based browsers (Chrome, Edge, Brave, Opera) and Firefox. Side panel mode requires Chrome 114+ or Edge 114+.
Can it verify any citation style?
Yes. The system was tested against 10 major citation styles (APA, MLA, Chicago, IEEE, Vancouver, Harvard, ACS, AMA, Oxford, Springer) with 93% APA exact match rate across all of them.
What happens if a database is unavailable?
The system handles timeouts and rate limits gracefully. It continues processing remaining references, shows clear error messages, and preserves already-completed results. You can retry failed lookups individually.
Can I export my verification results?
Yes, you can copy individual results or export entire batch verifications. Results include the verified metadata, match score, and evidence breakdown — ready to paste into your reference manager or document.
How is my privacy protected?
The extension is stateless — no reference data is stored locally. API keys are encrypted (AES-256-GCM) before being sent to our backend keystore. We do not collect usage data, telemetry, or analytics.
Get started in under a minute
Free, open source, and privacy-first.
Chrome Web Store
For Chromium browsers — Chrome, Edge, Brave, Opera
Firefox Add-on
For Firefox Desktop — download the XPI file
Optional: Bring your own API key for AI-powered extraction & rate limit increases.
Research methodology & dataset
Controlled study with 425 curated references across 10 citation styles.