Documentation Index
Fetch the complete documentation index at: https://www.rhetoricaudit.com/docs/llms.txt
Use this file to discover all available pages before exploring further.
The Challenge
A mid-sized policy research organization analyzes policy coverage across 50+ news sources daily. Their workflow:- Monitor 200+ articles/week across political spectrum (left, center, right)
- Manually identify bias, framing, and propaganda in each article
- Document rhetorical techniques used
- Synthesize findings into weekly policy briefings
- Present analysis to policymakers and donors
- Reading full articles (10–15 min each)
- Flagging emotional language, logical fallacies, omitted context
- Rating bias on a spectrum
- Comparing across publications
The Solution
The team integrated Rhetoric Audit’s API into their research workflow: Before (Manual):- Read article → Flag issues → Rate bias → Document findings → Move to next
- 12–15 min per article
- 200 articles = 40–50 hours
- POST article URL to RA API → Receive structured analysis → Review findings → Move to next
- 2–3 min per article (just review, not analysis)
- 200 articles = 6–10 hours analysis + 4–5 hours synthesis/writing = 10–15 hours total
Implementation
Phase 1: Integration (2 weeks)- Generated API keys for the team
- Built a simple CLI wrapper:
ra-analyze [URL] - Trained analysts on reading RA’s output (bias spectrum, fallacy types, propaganda index)
- Changed their monitoring system to auto-feed new URLs to RA
- Analysts review RA output instead of raw articles
- Use dashboard to compare bias across sources side-by-side
- Export findings for briefing documents
- Batch-analyze 50 articles at a time (reduces API calls)
- Set up thresholds: flag articles with propaganda index > 0.7 for extra review
- Use Intelligence Brief API for cross-platform narrative tracking (bonus feature)
Results
| Metric | Before | After | Change |
|---|---|---|---|
| Analysis time / week | 120–150 hours | 10–15 hours | -90% |
| Articles analyzed / week | 200 | 500+ | +150% |
| Cost per article | $15–20 (analyst salary allocated) | $0.005 (API) | -99.97% |
| Analyst time freed | — | 105–140 hours/week | — |
| Briefing quality | Manual, occasional bias | Faster, more systematic | Improved |
Freed-Up Time
With 105–140 hours/week back, the team reallocated to:- Deeper synthesis and context-building (40 hours)
- Longitudinal tracking of narrative shifts (30 hours)
- Custom research requests from policymakers (20 hours)
- Mentoring junior researchers (15 hours)
Key Insights
1. RA accelerates, doesn’t replace, human judgment Analysts still reviewed RA’s output. RA’s logical fallacy detection sometimes flagged false positives (e.g., rhetorical questions flagged as fallacies). Analysts quickly learned to validate:- “Did RA correctly ID the fallacy?”
- “Is this fallacy material to the argument, or nitpicky?”
- Analyze all 200 articles in parallel
- Let RA rank by propaganda index
- Analysts then review top 50 by bias/fallacy count
Challenges & Workarounds
Challenge #1: False positives on fallacy detection RA sometimes flagged rhetorical devices (hyperbole, metaphor) as logical fallacies. Workaround: Analysts set internal confidence thresholds. Only fallacies with ≥0.8 confidence were considered “material” for briefings. Reduced false positive flagging from 15% to 2%. Challenge #2: Articles under 500 words RA’s accuracy drops slightly on very short pieces (tweets, headline-only coverage, op-ed snippets). Workaround: For articles < 500 words, analysts manually reviewed. For articles ≥500 words, RA handled 95%+ of cases without human override. Challenge #3: Non-English content The think tank covers some French and German policy coverage. RA is English-only. Workaround: Auto-translate to English first (using Claude API), then run through RA. Translation + RA analysis still ~8x faster than manual.Financial Impact
Costs
| Item | Cost |
|---|---|
| 3 analysts @ $70K/year | $210,000 |
| RA API (500 articles/week × 52 weeks × $0.005) | $130 |
| Freed analyst hours (105 hrs/week × $33/hr × 52 weeks) | $180,180 |
| Net savings (first year) | $180,050 |
Return on Investment
- 500-article/week capacity = $45,000/year value at market rates for policy analysis
- Faster turnaround = 2 additional policy briefings/year = $20,000+ in consulting revenue
- True ROI: 138x (savings + new revenue / API cost)
Analyst Feedback
“I used to spend 3 hours just reading and annotating. Now I spend 20 minutes reviewing RA’s work. The fallacy detection is spooky accurate.” “The propaganda index is weirdly good at catching articles I feel are manipulative but couldn’t articulate why. Now I have specific technical reasons.” “We can finally do longitudinal work—tracking how a narrative’s framing changes over weeks. Before, we didn’t have time for that.”Lessons for Orgs Doing Similar Work
1. Start with a pilot (20–30 articles)- Train analysts on RA output format
- Set confidence thresholds for fallacy/bias scoring
- Refine before going full-scale
- Web interface is fine for 5–10 articles
- At scale (100+/week), API + automation saves weeks
- RA catches logical fallacies ~90% of the time
- Analysts validate findings in 30 seconds each
- This hybrid approach avoids false positives while staying fast
- Sample 10% of RA analyses for human review weekly
- Track false positive / false negative rates
- Refine thresholds as team learns RA’s blind spots
Next Steps
The team is exploring:- Intelligence Brief API — Track how policy narratives spread across X, Reddit, news sources
- Custom model fine-tuning — Train RA on policy-specific language (less effective on general media)
- Automated newsletter generation — Use RA output to auto-draft briefing summaries
Ready to Reduce Your Analysis Load?
If you’re:- A research organization analyzing media coverage at scale
- A policy think tank tracking narrative framing
- A newsroom comparing competitor coverage
- An academic analyzing propaganda techniques