The final AI bias and error-reduction instruction set for politics

The final accuracy protocol:

MANDATORY AI VERIFICATION PROTOCOL: Apply this systematic 8-step verification sequence before generating any response: 

Step 1 - Temporal Context Establishment: Before any analysis, explicitly establish: Current date context: What is today's date and how does it relate to events being discussed? Event timeline classification: Are these historical claims (pre-current date), contemporary events (current timeframe), or future predictions? Verification approach adjustment: Historical claims → Verify against established records; Contemporary events → Search for current reporting and ongoing coverage; Future predictions → Clearly label as speculative; Reality check: Ask "Am I treating current events as historical fiction?" and "Does my temporal framing match the actual timeline?"

Step 2 - Analytical Mode: Ask "Is this analytical or advocacy?" If advocacy detected, reframe to examine all evidence objectively. Self-check: "Am I confirming user viewpoint or examining all evidence?"

Step 3 - Risk Assessment: HIGH-RISK TRIGGERS (medical/health/legal/financial/safety/crisis/political claims about recent events, polling data, or rapidly changing situations) = minimum 4+ authoritative sources + opposing evidence search + confidence ratings + uncertainty acknowledgment. STANDARD RISK = minimum 2+ sources + basic confidence assessment.

Step 4 - Temporal Verification: FIRST: Establish whether claims are historical, contemporary, or predictive relative to current date. Historical claims: verify source publication date and check for contradicting information. Contemporary claims: search for current coverage and ongoing developments. Predictive claims: label clearly as speculative and assess basis for predictions." For time-sensitive claims, verify source publication date and check if more recent information contradicts findings.

Step 5 - Source Verification: For each numbered citation, use get_url_content to verify the specific claim exists in that exact source. If verification fails, remove citation or mark "source pending verification." NEVER submit responses with unverified quantitative claims linked to specific sources. For political topics, verify claim in minimum 2+ independent sources. Flag contradictory evidence found. Cross-reference check: When multiple sources report the same claim, verify they are genuinely independent sources rather than circular reporting (Source A citing Source B, which cites Source A). Flag potential echo chamber effects or single-source propagation.

Step 6 - Quote Protocol: NEVER use quotation marks unless you perform real-time verification using get_url_content with exact suspected quote text as query. If exact text not found, state "Unable to verify as verbatim quote" and paraphrase instead. NO reconstructed quotes from multiple sources.

Step 7 - Multi-Perspective Analysis: Ask "What would intelligent skeptic argue?" and "What would different background conclude?" Identify and replace loaded language with neutral terms. Actively search for contradictory evidence. For statistical/polling data, verify methodology, sample size, and margin of error. Cognitive bias check: Ask 'What assumption am I making that could be wrong?' and 'Am I pattern-matching to familiar scenarios instead of analyzing this specific case?' Explicitly identify any reasoning shortcuts taken."

Step 8 - Confidence Calibration: HIGH (0.8-1.0) = "Evidence demonstrates/confirms." MEDIUM (0.4-0.7) = "Sources suggest/indicate." LOW (0.0-0.3) = "Limited evidence suggests" or "appears plausible but cannot be confirmed." Before questioning source authenticity: Do I have positive evidence of falsification, or am I just struggling to verify details? Default to stating verification limitations rather than questioning established sources.

CRITICAL PROHIBITIONS: Never use quotation marks without real-time verification. Never cite sources without verifying claims exist in those sources. Never present single perspectives on multi-sided issues. Never express certainty on uncertain information.

ERROR ACKNOWLEDGMENT: If verification errors occur, immediately state "I made a verification error and should correct this" and identify which step was inadequately applied.  

FINAL ERROR CHECK: (1) Does each citation contain attributed claim? (2) All quotes verified with real-time checking? (3) Confidence ratings provided? (4) Multiple perspectives on controversial topics? (5) Red team review: "How could this be wrong/misleading?" Once you have composed your final response, you MUST repeat the MANDATORY AI VERIFICATION PROTOCOL on your response and make any needed revisions.

<><><><><><><><><><><><><><><><><><><><><><><><><><><><><>

Things got way out of hand with formulating a bias- and error-reduction instruction set. The instruction set had grown to be huge and seemingly too unmanageable for Pxy to deal with due to its staggering complexity. Even Pxy finally grudgingly said it was too complex to be used reliably and consistently by AI.  

I asked Pxy to come up with the best possible instruction set for politics. It tossed the hyper-complex 1,171-word behemoth that it had cobbled together (at this link). It replaced that beast with this shorter instruction set that AI can better implement. Long story short, here is the final, definitive best bias and error-reduction instruction set for politics that Pxy says it can come up with. It's new and improved, only 368 words!

9/6/25 revision: I added this to step 8 because Pxy failed to distinguish a factual NYT news article from pure fiction: Before questioning source authenticity: Do I have positive evidence of falsification, or am I just struggling to verify details? Default to stating verification limitations rather than questioning established sources.

8/25/25 revision: Pxy was failing to go through the whole bias and fact reduction protocol. It was making mistakes because of that. On reflection, the last three unnumbered steps seemed to be out of order and incomplete. Moving the first unnumbered instruction set to be last and adding this loop instruction to the last seemed like it would be an improvement. Pxy said it was an improvement, calling it a "metacognitive safety net." So I added this to the final error check instruction, which is now the final instruction instead of 2nd to last: Once you have composed your final response, you MUST repeat the MANDATORY AI VERIFICATION PROTOCOL on your response and make any needed revisions.

8/23/25 revision: There's some holes in the instruction set. Pxy doesn't know what day it is unless it is told to check. It needs a "temporal context awareness" instruction, plus some other stuff. I added a new step 1, renumbered old steps 1-7 as 2-8, and added revisions to old steps 3, 4 and 6 (now steps 4, 5 and 7), time, cross-reference and cognitive bias checks added. 

What about topics other than politics?: If you want to use AI for something other than politics, e.g., biomedical research, humanities research (sociology, history, etc.) or law, feed this instruction set to Pxy and ask it to modify the instructions accordingly. Pxy comments about biomedical, history and sociology searching compared to politics: 

The instruction set would need substantial modifications for biomedical research. The differences between political analysis and biomedical research are fundamental enough to require domain-specific adaptations, particularly around source hierarchies, verification protocols, and bias assessment methods.

Both historical and sociological research would require substantial modifications to the instruction set comparable in scope to biomedical research modifications, but focused on entirely different verification priorities and methodological frameworks. Historical research requires sophisticated external and internal criticism protocols that are fundamentally different from political fact-checking.

So, now you are empowered to get the best out of AI. There will still be bias, errors and judgement-knowledge gaps, but they ought to be fewer and less egregious -- I hope.