alivirgo Ask AI to Flag Speculative Statements (Epistemic Tagging) (AI Skill)
Overview
Because AI models are trained to write in a consistent, authoritative tone, they deliver a verified historical date ("The Apollo 11 moon landing was in 1969") and an unverified financial estimate ("Competitor X generates $45M in ARR") using the exact same confident cadence.
The Epistemic Tagging Protocol forces the AI to prepend explicit visual markers ([VERIFIED FACT], [HEURISTIC / ESTIMATE], [SPECULATION]) to its claims, making uncertainty instantly recognizable.
The 3-Tier Epistemic Tagging Schema
┌─────────────────────────────────────────────────────────────┐
│ The Epistemic Tagging System │
│ │
│ [VERIFIED FACT] ──► Established data, code syntax, or │
│ explicitly quoted text in source │
│ [HEURISTIC] ──► Industry rule-of-thumb or standard │
│ statistical pattern │
│ [SPECULATION] ──► Educated guess, prediction, or │
│ unverified market projection │
└─────────────────────────────────────────────────────────────┘
Master Epistemic Tagging Prompt Templates
Pattern 1: The Tagged Market & Intelligence Brief
Use for market research, competitor teardowns, and trend forecasting:
Analyze [COMPETITOR / MARKET TREND / TOPIC].
Formatting Directive:
Every paragraph or major bullet point MUST begin with one of these 3 epistemic tags:
- `[VERIFIED FACT]`: If supported by public SEC filings, official documentation, or verified press releases.
- `[HEURISTIC / ESTIMATE]`: If derived from industry benchmarks or standard sizing models.
- `[SPECULATION]`: If based on rumors, future projections, or AI deduction.
Do not smooth over uncertainties with confident language.
Pattern 2: The "Admit Unknowns" Guardrail
Use for scientific, technical, or regulatory inquiries:
Answer [QUESTION].
Constraint:
If you are uncertain about any sub-component, write: `⚠️ [UNVERIFIED: Reason for uncertainty]`.
Do not attempt to generate plausible-sounding guesses.
Real-World Case Study
Scenario: Estimating a Competitor's Tech Stack & Cloud Spend
Un-Tagged Output (Confusing Fiction with Fact)
"Company ABC runs on AWS EKS with Snowflake for their data warehouse. They spend approximately $250,000 per month on cloud infrastructure and have 45 backend engineers in Eastern Europe."
❌ (The user has no idea which parts are verified truths vs wild guesses).
Epistemically Tagged Output (Crystal Clear Calibration)
[VERIFIED FACT]: Company ABC lists open job requisitions requiring AWS EKS, Terraform, and PostgreSQL (Source: Official Careers Page, Aug 2024).[HEURISTIC / ESTIMATE]: Based on standard B2B SaaS gross margins and their last reported 150-person headcount, estimated annual AWS cloud spend is between $1.8M and $2.5M (~$150k-$200k/mo).[SPECULATION]: They are likely migrating away from Redshift to Snowflake based on recent LinkedIn hiring of two Snowflake Data Architects, but this has not been publicly confirmed.
Summary Best Practices
- Use for High-Stakes Presentations: Epistemic tags allow executives to know immediately what data can be defended in front of a board vs what is an estimate.
- Enforce visual brackets: Tagging in brackets (
[FACT]) makes scanning 10x faster than reading long qualifying phrases ("It is worth noting that we estimate...").