Analytics & Benchmark
Live stats from our database. Head-to-head comparison against every major fact-check bot on X. No cherry-picking — these are real numbers.
Our Numbers (Live)
per check
of mentions answered
+ 5,000 bias-rated via CSV
with FEC donation data
political lean tracked
29% of all checks
7.3% of all checks
32.8% of all checks
22.4% of all checks
Benchmark vs. Competitors
Feature-by-feature comparison against every major fact-check bot on X. Competitor data from published benchmarks, Tow Center (Columbia), and public testing.
| Feature | True Source | Grok | GPTZero | ArAIstotle | Community Notes |
|---|---|---|---|---|---|
| Claim-level breakdown | ✓ | ✗ | ✗ | Partial | ✗ |
| Cites real sources | ✓ | Fabricates | Academic only | ✓ | ✓ |
| Response time | <60 sec | Instant | Rate-limited | ~30 sec | 14+ hours |
| Proactive scanning | ✓ | ✗ | ✗ | ✗ | ✗ |
| Source ownership + FEC donations | ✓ | ✗ | ✗ | ✗ | ✗ |
| Poster political profiling | ✓ | ✗ | ✗ | ✗ | ✗ |
| Image/chart analysis | ✓ | ✗ | ✗ | ✗ | ✗ |
| Poster accuracy history | ✓ | ✗ | ✗ | ✗ | ✗ |
| Self-learning (gets smarter) | ✓ | ✗ | ✗ | ✗ | ✗ |
| Works without being tagged | ✓ | ✗ | ✗ | ✗ | N/A |
Accuracy Metrics
Self-published benchmark (Facticity AI). TIME Best Inventions 2024. Mention-triggered, not proactive.
94% citation error rate (Columbia Tow Center). Disqualified from comparison tests — 0 valid responses. User-initiated only (button or @mention).
Source-finder approach — surfaces evidence, no verdicts. 220M+ scholarly articles.
AI search engine, not a dedicated fact-checker. 37% of queries answered incorrectly (The Quint).
Benchmark Sources
Citation error rates: Columbia Tow Center for Digital Journalism study — tested 200 queries across 8 AI platforms. ArAIstotle: 14%, Perplexity: 42%, Grok: 94%.
Benchmark scores: Facticity AI comparison — 45 real tweets, 3 LLM judges (Gemini 2.5 Pro, Claude 3.7 Sonnet, GPT-4o), 180 blinded evaluations. ArAIstotle: 8.55/10, Perplexity: 5.91/10. Note: published by ArAIstotle's own team.
TrueSourceBot metrics: Live data from our production database. Updated every 5 minutes. "TBD" metrics require an independent third-party audit — we will not self-certify.
What We Analyze (8 Categories)
Every fact-check runs through all 8 analysis dimensions. No other bot does this.
Monthly Volume
All Fact-Checks
Showing 0 of 1,874 claims
How We Verify
Every tweet or claim submitted to TrueSource goes through our dialectical verification pipeline:
- Image content extraction — OCR via Claude Haiku vision extracts text, statistics, and claims from screenshots, infographics, and article images attached to tweets
- Atomic claim extraction — break complex statements into individual, verifiable claims
- Multi-source evidence search — cross-reference each claim against news archives, government data, and fact-check databases
- Dialectical analysis — search for evidence FOR and AGAINST each claim, weight by source reliability (MBFC data)
- Transparent verdicts — every verdict includes the evidence and reasoning, linked from the tweet reply
We publish these stats openly because accountability is the foundation of trust. If we ask you to question your sources, you should be able to question ours. Metrics marked "TBD" require independent third-party testing — we refuse to self-certify accuracy numbers.