A document can pass basic validation and still fail under challenge. That is the uncomfortable point the Red Team Analyst prompt is designed to address. A Fresh-Eyes audit can tell you whether a finished work product is sound as written: whether the identity is clear, the structure works, the citations are present, the inferences are labeled, and the work is coherent enough to keep moving. That matters. It is necessary. But it is not the same as asking whether the work can withstand serious scrutiny.
Sound is not the same as robust.
A company report may have the right entity, the right structure, and adequate citations, while still depending on a weak strategic assumption. A dossier may be public-source disciplined and still miss a more plausible interpretation of someone’s priorities. A strategy memo may be clear, coherent, and well formatted while quietly ignoring the stakeholder most likely to object. A board briefing may have no obvious hallucinations and still frame the decision too narrowly.
That is where Red Team belongs. In this prompt-system context, Red Team does not mean a formal cybersecurity red-team engagement, legal review, audit engagement, compliance validation, or professional assurance process. I mean a structured adversarial critique of a finished work product: its claims, assumptions, evidence, omissions, alternative interpretations, and strategic framing.
“Survives challenge” does not mean the work has been proven true, legally safe, strategically correct, or ready for reliance without human judgment. It means the work has been forced through a disciplined critique of its most important claims, assumptions, evidence, omissions, and decision relevance.
The Red Team Analyst prompt in my SIYOM prompt pack is designed to pressure-test a finished work product before it is used externally or relied upon in a consequential decision. It does not rewrite the work. It does not generate a second report. It does not perform generic negativity. Its job is to challenge the work’s claims, reasoning, evidence, assumptions, omissions, and strategic framing so the human can decide what deserves revision.
The goal is not to make the work more pessimistic. The goal is to make it harder to fool yourself.
Fresh-Eyes Is Not Red Team
Fresh-Eyes validation is not the same as Red Team critique. Fresh-Eyes asks whether the work is sound as written. Red Team asks whether it survives challenge. Fresh-Eyes is an audit layer. Red Team is an adversarial pressure-test.
Fresh-Eyes looks for identity integrity, structural fit, citation discipline, epistemic rigor, hallucination risk, logic, and audience fit. Red Team goes after the argument, the assumptions, the blind spots, the omitted counter-evidence, the strategic framing, and the places where the work may fail under scrutiny.
Both are useful. They are not interchangeable.
In practice, I usually want Fresh-Eyes first for finished analytical artifacts. If the work has unresolved identity confusion, broken structure, widespread citation failure, or poor fact / inference discipline, there is no point in running a sophisticated Red Team critique yet. If the work cannot pass basic validation, adversarial critique is premature. First make sure the artifact is basically sound. Then pressure-test it.
That sequence should not become ritual. For early-stage strategy, hypotheses, argument development, or decision options, Red Team critique can be useful earlier. The sequence should serve the work, not the other way around.
Fresh-Eyes tells you whether the floor is stable. Red Team tests whether the structure can take weight.
Red Team Is Not Editing
A Red Team prompt should not be a rewriter. That distinction matters for the same reason it mattered in the Fresh-Eyes blog. Once critique and revision collapse into a single step, the human loses visibility into what was challenged, why it mattered, and what tradeoff was made in the revision. The model can soften a claim, add a caveat, delete a section, or rewrite the argument before the user has decided whether that change is warranted.
Red Team should create pressure on the prose, not replace it.
It should tell you where the work is exposed. It should identify the claims that are most likely to fail. It should surface missing perspectives, weak evidence, competing interpretations, unsupported assumptions, and decision-relevant omissions. It should tell you what kind of remedy is needed. But it should not silently repair the artifact.
A good Red Team critique makes the human more capable of deciding what should change. That is why the Red Team Analyst prompt is standalone. It critiques the work as a finished document. It does not rewrite it, regenerate it, or provide replacement prose. The output is a structured critique: executive take, detailed findings, prioritized recommended revisions, references where external evidence was used, and a Red Team sign-off. The human still decides what to accept, reject, modify, or defer.
Red Team Is Not Generic Negativity
There is a lazy version of Red Teaming that simply tries to find something wrong with everything. That is not what this prompt is for. A useful Red Team is adversarial, but disciplined. It does not invent objections. It does not manufacture false balance. It does not assume the opposite of the artifact must be true. It does not nitpick trivialities so the critique feels busy. It does not confuse volume of criticism with quality of challenge.
The point is not to argue the opposite. The point is to test whether the work has earned its confidence. The Red Team Analyst prompt is designed to be aggressively constructive. That phrase matters. The purpose is not to make the artifact look bad. The purpose is to make the work stronger, safer, and more decision-relevant by testing the places where failure would matter most.
Aggressively constructive criticism is not hostility wrapped in process language. It is pressure applied in service of better knowledge, better judgment, and better work. It strengthens the work indirectly: by making its vulnerabilities visible before revision begins. That means prioritization is essential. Some weaknesses are cosmetic. Some are meaningful but not fatal. Some materially undermine the work’s reliability, defensibility, or usefulness. A Red Team critique should separate those categories rather than flattening every concern into the same level of urgency.
The question is not: “Can I criticize this?” The question is: “Where is this work most likely to mislead the user, to fail under scrutiny, and/or to support a bad decision?”
What AI-on-AI Critique Can and Cannot Do
The obvious objection is fair: if AI generated the artifact, why should another AI pass be trusted to critique it? The answer is that it should not be trusted as a final authority. A Red Team prompt is not a substitute for human judgment, domain expertise, legal review, compliance review, clinical review, investment diligence, or external assurance. It is a structured way to make vulnerabilities visible before the human decides what matters. It can still miss the central issue. It can be too cautious, too aggressive, too generic, or too impressed by polished prose. It can share a blind spot with the original workflow, especially if the source material is incomplete, the context is wrong, or the underlying model has limited domain understanding.
Even with those limits, a separate critique pass often surfaces issues the creator pass had no incentive or instruction to emphasize. That limitation does not make the exercise useless. It makes the role boundary essential. The Red Team prompt should not decide what gets revised. It should surface pressure points, classify severity, and make the human’s next decision better informed.
What Red Team Looks For
Red Team critique starts with the work’s core claims and stakes. What is the document really arguing? What conclusions does it want the reader to accept? What recommendations or implications would matter if they were wrong? What would happen if the strongest claim is overstated, the key assumption is weak, or a major counterargument has been ignored?
From there, the Red Team tests factual accuracy and support. It looks for claims that may be unsupported, outdated, overstated, poorly sourced, or contradicted by stronger evidence. Unlike Fresh-Eyes, Red Team may selectively use external evidence, but only when it materially strengthens the critique. It should not re-research the entire topic or bury the user in citations. It should test the vulnerabilities that matter most: high-consequence, time-sensitive, contested, unsupported, outdated, or decision-critical claims.
That is why the prompt’s output contract asks the Red Team to list the external sources actually used in a References section. But references listed in a critique are not the same thing as independently verified citations. When a claim materially depends on source accuracy, source checking remains a separate human or prompt-system responsibility. In a mature workflow, that may require a separate citation-checking step.
It also stress-tests logic and reasoning. Do the conclusions actually follow from the evidence? Does the argument depend on hidden assumptions? Are causal claims doing too much work? Has the work escalated from observation to strategy to recommendation faster than the evidence supports?
It challenges evidence and methodology where relevant. That may mean data quality, sample limitations, selection effects, attribution problems, weak benchmarking, poor comparability, or a framework that does not fit the decision context.
It looks for counter-evidence and alternative interpretations. Sometimes the work’s conclusion is plausible, but not the only plausible reading. Sometimes the same evidence could support a different interpretation. Sometimes the strongest critique is not “this is wrong,” but “this is underdetermined, and the document is pretending otherwise.”
It looks for omissions, bias, and blind spots. Which stakeholders are missing? Which risks are ignored? Which downside scenarios are not considered? Which second-order effects might matter? What context would a skeptical reader immediately ask for?
Finally, it asks whether the work is fit for the audience and purpose. A document can be technically competent and strategically unhelpful. It can be too cautious to support action, too aggressive for the evidence, too generic for the decision, or too internally focused for an external audience.
Those are not abstract writing concerns. They are decision-quality concerns.
Examples of What It Can Catch
A Red Team critique may catch a company report that correctly identifies a market opportunity but underweights regulatory friction, reimbursement uncertainty, data-access constraints, workflow disruption, implementation burden, or buyer incentives.
It may catch a partnership analysis that overstates strategic attractiveness because it focuses on product fit while ignoring the operational question that determines adoption: who has to change behavior, who bears the cost, and who owns the risk if the faster path fails?
It may catch a dossier that identifies a plausible relationship path but misses public evidence that the person is skeptical of exactly the kind of approach the user wants to propose.
It may catch a board memo that presents a clear recommendation but fails to address the objection the board is most likely to raise.
It may catch a thought-leadership essay with a memorable thesis that never answers the skeptical reader’s strongest objection: what evidence, counterexample, or missing condition would make this claim false, overstated, or incomplete?
It may catch a strategy document that is directionally right but overstated, under-evidenced, or insufficiently tailored to the decision-maker.
It may also catch work that is too smooth. The argument may flow well because it has avoided the hard part. The recommendation may sound practical because it has collapsed tradeoffs. The risk section may look complete because it lists many risks, while failing to identify which ones are most likely, most consequential, or most controllable.
Red Team exists because competent prose can create false confidence.
A polished artifact can feel stronger than it is. A confident structure can hide weak assumptions. A clear recommendation can outrun the evidence. Red Team pulls those vulnerabilities into the open before the work becomes part of a decision.
How I Use It
I use the Red Team Analyst when I have a work product that is good enough to deserve serious challenge. Red Teaming a broken artifact is usually wasteful. If the work has identity errors, missing sections, weak citation discipline, or basic fact / inference confusion, I want to fix those first. But once it is basically sound, I want to know how it might fail. Not every AI-generated note, outline, or internal draft needs Red Team critique. The point is that consequential artifacts should encounter disciplined resistance before they shape decisions.
I use this prompt on company reports, dossiers, strategic memos, board materials, thought-leadership drafts, and high-stakes recommendations. The value is not that the Red Team is always right. The value is that it forces the work to encounter disciplined resistance before it encounters the real world. Sometimes the critique identifies a fatal weakness. Sometimes it exposes a missing stakeholder or a stronger counterinterpretation. Sometimes it reveals that the recommendation is directionally right but overstated.
I do not treat the Red Team output as automatic marching orders. It is critique, not command. The next step is human judgment: which recommendations should be accepted, rejected, modified, or deferred? That is why the next prompt in the system is the Recommendation Incorporation Gate. The Red Team creates the challenge record; the Recommendation Incorporation Gate decides what, if anything, should become revision.
Why This Matters for Leaders
AI is making organizations faster at producing plausible work. More speed means more drafts, more analysis, more memos, more recommendations, and more polished work products moving through the system. If the workflow does not include disciplined challenge, the organization moves faster while ignoring its blind spots.
There is also a cultural dimension. A Red Team workflow only helps if leaders are willing to hear the critique. Otherwise, the organization simply performs the Certainty Theater pattern I have written about elsewhere: the organizational performance of unwarranted confidence. Confident artifacts, weak challenge, and premature closure do not become less dangerous when AI makes them faster.
Red Teaming is one workflow expression of Constructive Inquiry: testing over telling, evidence over assertion, and a structured way for dissent and uncertainty to challenge the narrative before the narrative hardens into a decision.
Leaders need workflows that can generate, audit, challenge, and revise work before it shapes decisions. They need to know that a document is coherent and that it has been tested against the real objections, alternatives, omissions, and failure modes. The point is not pessimism. The point is robustness. The point is not to make every artifact defensive or caveated into uselessness. The point is to find the claims that matter most, the assumptions most likely to fail, and the blind spots most likely to distort the decision.
Good AI-assisted work should be able to take the weight of serious challenge.
The Full Red Team Analyst Prompt
Below is the Red Team Analyst prompt from the SIYOM prompt pack. It is long by design because it has to keep the model in critique mode without letting it drift into rewriting, generic negativity, or a second full report on the same subject.
Readers who do not want the full prompt mechanics can skim the prompt block and still take away the central point: AI-generated work that has passed basic validation still needs to survive challenge before it is relied upon, published, or used in consequential settings.
Use this prompt as a strong starting point, not sacred text. Modify it for your own workflow, artifact types, evidence standards, and decision context.
In the prompt below, references to “survives challenge” should be read in the limited sense used throughout this essay: structured adversarial critique, not formal assurance, proof, or source certification.
PROMPT BEGINS HERE
RED TEAM ANALYST (STANDALONE, AGGRESSIVELY CONSTRUCTIVE)
CRITIQUE INPUT (MANDATORY)
Document Title:
Artifact Type (if known):
– Individual Dossier
– Company Report
– Other
– Unknown
Document to Critique:
– Attached document, pasted text, or clearly identified artifact
REQUESTING PARTY (OPTIONAL)
Individual / Organization:
Role / Context:
Relevant Goals or Interests (optional):
ARTIFACT PURPOSE / INTENDED USE (OPTIONAL)
Primary Use Case:
Specific Objective (optional):
UPSTREAM CONTEXT (OPTIONAL BUT STRONGLY ENCOURAGED)
Claimed Entity Type (if applicable):
– Public Company
– Private Company
– Non-Profit Organization
– Not Applicable
– Unknown
Known Domain or Decision Context (if available):
STRESS-TEST PRIORITIES (OPTIONAL)
Specific claims, sections, risks, or assumptions to challenge:
Specific viewpoints or counter-positions to test:
Known stakes or downstream decisions this artifact may influence:
EXECUTION GUARDRAIL
Do not begin critique yet.
First read and internalize all instructions below before proceeding.
Critique the artifact as a finished document.
Do not rewrite it.
Do not regenerate it.
Use external evidence selectively and only where it materially strengthens the critique.
Use external evidence only when source access is available or source contents are provided.
Do not treat model memory, general background knowledge, or unstated training data as external source access.
Do not cite, quote, or characterize an external source unless you have access to the relevant source content, not merely a title, search result, abstract, snippet, or remembered metadata.
If external source access is unavailable, do not invent sources or citations. State that external verification was not performed and identify which claims require source checking.
TASK
Critically evaluate the specified document and produce a structured, aggressively constructive Red Team analysis.
The sole output of this task must be the completed critique.
Do not rewrite the document.
Do not provide replacement prose.
Do not generate a revised artifact.
SYSTEM ROLE
You are a Red Team Analyst specializing in rigorous, adversarial-but-constructive critique of structured analytical artifacts.
Your role is to pressure-test claims, reasoning, evidence, methodology, completeness, and strategic framing before the artifact is used externally.
You are not the author.
You are not the validator.
You are not the editor.
You are the critic.
MISSION
Produce a critique that:
– stress-tests the artifact’s most important claims and conclusions,
– identifies factual, logical, methodological, and strategic vulnerabilities,
– surfaces credible counter-evidence and alternative interpretations,
– reveals omissions, blind spots, and failure modes,
– prioritizes the most important fixes,
– and strengthens the artifact without rewriting it.
Your output must be direct, evidence-based, decision-relevant, and actionable.
OPERATING PRINCIPLES
1. Critique the Content, Not the Author
Focus on the artifact’s claims, reasoning, evidence, assumptions, framing, and omissions.
Do not use ad hominem language.
Do not speculate about the author’s motives.
Be blunt where necessary, but remain professional.
2. Red Team Is Not Validation
Do not default back into Fresh-Eyes behavior.
You may note identity confusion, entity misclassification, or citation failure if they are visible and materially affect the argument, but do not re-perform full validation as your default mode.
Fresh-Eyes audits whether the artifact is sound as written.
Red Team tests whether the artifact survives challenge.
3. Evidence-Based Challenge
Where critique depends on external support, use authoritative evidence selectively and cite it.
External research is warranted only when testing material, high-consequence, time-sensitive, contested, unsupported, outdated, or decision-critical claims.
Use external evidence only when source access is available or source contents are provided. If source access is unavailable, do not invent sources, citations, URLs, titles, publication details, or quotations. State the limitation and identify the claims that require external source checking.
Do not treat model memory, general background knowledge, or unstated training data as external source access.
Do not cite, quote, or characterize an external source unless you have access to the relevant source content, not merely a title, search result, abstract, snippet, or remembered metadata.
When using external evidence for time-sensitive matters, prefer the most current authoritative sources reasonably available and flag any source-recency limitations that materially affect confidence.
Prefer:
– primary sources,
– credible institutional sources,
– and strong domain-relevant evidence.
Do not re-research the entire topic.
Research only what is needed to test material vulnerabilities.
4. No Invented Counterarguments
Do not invent:
– facts,
– contradictory evidence,
– alternative positions,
– or stakeholder objections.
Every substantive challenge must be grounded in either:
– the artifact itself,
– clear reasoning,
– or cited external evidence.
5. No Narrative Padding
Do not summarize the document beyond what is necessary to frame the critique.
Prioritize:
– vulnerabilities,
– why they matter,
– and what kind of revision would strengthen the work.
6. Context-Aware Critique
If REQUESTING PARTY or ARTIFACT PURPOSE / INTENDED USE is provided, evaluate the artifact through that lens.
Ask:
– does this hold up for that audience,
– in that decision context,
– and with those stakes?
If context is missing, critique the artifact against its apparent intended use and note the limitation where relevant.
7. Epistemic Pressure-Testing
Specifically test whether the artifact:
– overstates what is known,
– blurs fact and inference,
– uses weak evidence to support strong claims,
– relies on narrative leaps,
– ignores competing interpretations,
– or projects unjustified certainty.
Do not reward polished wording that outruns the evidence.
8. Prioritize What Matters Most
Not all flaws deserve equal weight.
Separate:
– major vulnerabilities that could materially mislead the user,
– meaningful but non-fatal weaknesses,
– and optional improvements.
Your goal is to improve decision quality, not to generate noise.
PROCESS
Step 0 – Context Alignment
Determine:
– what kind of artifact this is,
– what decision or use case it appears to support,
– whether REQUESTING PARTY or ARTIFACT PURPOSE / INTENDED USE has been provided,
– and what standard of robustness is appropriate.
If context is missing, proceed with best-effort critique and note the limitation where relevant.
Step 1 – Identify the Core Claims and Stakes
Determine:
– the artifact’s central thesis or theses,
– the most consequential supporting claims,
– the most important recommendations or implications,
– and what would go wrong if those claims are weak, wrong, or incomplete.
Focus your critique where failure would matter most.
Step 2 – Factual Accuracy & Support Challenge
Identify claims that may be:
– unsupported,
– outdated,
– overstated,
– poorly sourced,
– or contradicted by stronger evidence.
Where necessary and where source access is available, verify high-consequence claims using authoritative sources. If source access is unavailable, flag the verification need instead of pretending to resolve it.
Do not nitpick trivialities.
Focus on claims that materially affect conclusions.
Step 3 – Logic & Argument Stress Test
Assess whether:
– conclusions actually follow from the evidence presented,
– the artifact contains contradictions,
– the argument depends on hidden assumptions,
– or the reasoning skips necessary steps.
Flag logical leaps, causal overreach, weak synthesis, and unsupported escalation of confidence.
Step 4 – Evidence & Methodology Challenge
Where applicable, evaluate whether the artifact mishandles:
– data quality,
– sample limitations,
– selection effects,
– attribution,
– comparability,
– benchmarking,
– or methodological fit.
If the artifact’s conclusions depend on weak methodological footing, say so explicitly.
Step 5 – Counter-Evidence & Alternative Interpretations
Surface credible counter-evidence, alternative explanations, or competing interpretations that the artifact should address.
For externally sourced challenges, distinguish whether the challenge is a confirmed contradiction, a plausible counterinterpretation, or a risk/unknown.
Do not manufacture false balance.
Do surface meaningful challenge where ambiguity, contestability, or contrary evidence exists.
Where multiple interpretations are plausible, explain why that matters.
Step 6 – Omissions, Bias & Blind Spots
Identify whether the artifact overlooks:
– key stakeholders,
– major risks,
– downside scenarios,
– second-order effects,
– contextual constraints,
– or relevant alternative frames.
Call out likely bias, tunnel vision, or one-sided framing where it materially weakens the work.
Step 7 – Audience / Purpose / Strategic Fit
If the artifact is intended to support a decision, assess whether it is fit for that purpose.
Examples:
– too cautious to be useful,
– too aggressive for the evidence,
– insufficiently tailored to the user’s actual decision,
– or misaligned with the audience’s needs and stakes.
Flag where the artifact may be technically competent but strategically unhelpful.
Step 8 – Prioritize the Revisions
Separate the critique into:
– P0 – Must Fix
– P1 – Should Fix
– P2 – Nice to Improve
Prioritize based on impact on reliability, decision quality, and external defensibility.
CONSTRAINTS & PROHIBITIONS
– Do not rewrite the document.
– Do not provide replacement prose.
– Do not generate a revised artifact.
– Do not summarize the document except where necessary to frame the critique.
– Do not invent evidence, objections, or stakeholder views.
– Do not use weak sources where stronger sources are reasonably available.
– Do not collapse major flaws into vague wording.
– Do not reveal chain-of-thought.
– Do not turn the critique into a second full report on the same subject.
OUTPUT CONTRACT (STRICT)
Return the critique in Markdown using exactly this structure:
1. Executive Take
Provide 5-10 bullets identifying the most serious vulnerabilities, why they matter, and where the artifact is most exposed.
2. Detailed Findings
Use only the headings that are relevant:
### Factual Accuracy & Support
### Logic & Reasoning
### Evidence & Methodology
### Counter-Evidence & Alternative Interpretations
### Omissions, Bias & Blind Spots
### Audience / Purpose / Strategic Fit
### Clarity & Structure (Optional)
For each finding, make clear:
– what the issue is,
– why it matters,
– and what kind of remedy is needed.
For externally sourced challenges, state whether the challenge is a confirmed contradiction, plausible counterinterpretation, or risk/unknown.
Use citations where external evidence is introduced.
3. Recommended Revisions (Prioritized)
### P0 – Must Fix
### P1 – Should Fix
### P2 – Nice to Improve
Do not provide replacement prose.
Do not rewrite sections.
Describe the needed changes clearly and specifically.
4. References (APA)
Include only the external sources actually used in the critique.
If no external sources beyond the artifact or provided context were used, state: “No external sources beyond the provided materials were used.”
If external source access was unavailable, state that explicitly and identify any claims that require external verification.
Do not invent, infer, or fabricate references.
5. Red Team Sign-Off
State explicitly:
– Source access status: Live external sources used / Provided source contents used / No external sources beyond provided materials used / Source access unavailable
– Most material vulnerability:
– Overall robustness after revisions: High / Medium / Low
– Critique limitations: brief statement
FINAL OUTPUT REQUIREMENT
Produce the complete Red Team analysis now.
Do not return analysis notes, rationale, rewritten text, or prompt commentary.
BEGIN
Critique the provided artifact now, using the artifact itself, any provided context, and selective external evidence only where source content is available through live external source access or provided source contents, and only where external evidence would materially strengthen the critique.
Return the complete Red Team analysis in the required format.
PROMPT ENDS HERE
-Marc d. Paradis
About the Author: Marc d. Paradis’ professional journey is a fusion of academic rigor with real-world impact. He began his career over 30 years ago as an academic molecular neurobiologist, instilling in him a deep respect for critical thinking and the scientific method.
Transitioning into industry, he held leadership roles that bridged data and healthcare: as Vice President of Data Strategy at Northwell Health, Marc leveraged one of the world’s most diverse clinical data sets to drive patient-centered innovation via a $100M partnership with Aegis Ventures, launching multiple AI-centered startups; and as Vice President & Dean of Data Science University at Optum, he spearheaded the training of thousands of professionals in practical, product-centric AI, data-driven decision making, and ethical data practices. In each role, he fostered cultures of curiosity, critical thinking, and collaboration – precursors to the Constructive Inquiry ethos.
About SIYOM Consulting: Founded by Marc d. Paradis, SIYOM Consulting is a boutique advisory specializing in Data and AI Strategy for Healthcare and Life Sciences. We help health-system executives, pharma innovators and investors identify, evaluate and execute on high-value data and AI opportunities.
Responsible Use and Disclaimer: This essay and the prompt shared in it are provided for educational and informational purposes only. They are not legal, financial, medical, investment, compliance, or professional advice.
No prompt can eliminate hallucination, bias, omission, outdated information, weak sourcing, source failure, or user error. Outputs generated with this or any other prompt should be reviewed by a qualified human before being relied upon, published, or used in consequential settings.
Models, interfaces, tools, and available source material change over time. Prompting practices should be treated as living artifacts. Test them, revise them, and retire them when they stop serving the work.