{"data":{"items":[{"id":"8450e4ad-1112-4b72-889f-8d1e84c80ef3","excerpt":"ChatGPT Power User Guide 2026 — Advanced Tips That Actually Work — Most ChatGPT guides are still teaching 2023 habits on a 2026 platform. GPT-4o is retired. o3 is retired. You're running GPT-5 now (Instant / Thinking / Codex). This guide covers what actually works today — prompting, features, and token efficiency. No f","url":"https://www.reddit.com/r/ChatGPTPromptGenius/comments/1vucoa2/chatgpt_power_user_guide_2026_advanced_tips_that/","role":"request","weight":1.1217734,"occurredAt":"2026-08-21T10:36:23.000Z","sourceKey":"reddit","sourceName":"Reddit","credibility":0.62,"venue":"ChatGPTPromptGenius","intent":"feature_request","painScore":0.36,"sentiment":0.6451613,"confidence":0.82483333,"matchedPatterns":["missing_feature","manual_process","product:chatgpt"],"statement":"Flag clauses that create liability, identify missing protections, and always note when local counsel is essential.","title":"ChatGPT Power User Guide 2026 — Advanced Tips That Actually Work","body":"Most ChatGPT guides are still teaching 2023 habits on a 2026 platform. GPT-4o is retired. o3 is retired. You're running GPT-5 now (Instant / Thinking / Codex). This guide covers what actually works today — prompting, features, and token efficiency. No fluff.\n\n# ⚡ PICK THE RIGHT MODEL FIRST\n\n|Model|Use For|\n|:-|:-|\n|**GPT-5.5 Instant**|Writing, emails, summaries, quick Q&A|\n|**GPT-5.5 Thinking**|Hard reasoning, math, debugging, strategy|\n|**GPT-5.3 Codex**|Autonomous coding, multi-file dev tasks|\n\nUsing Thinking for simple tasks = slow + wastes quota. Using Instant for hard logic = confident hallucinations.\n\n# 🧠 PROMPTING — FOUNDATION\n\n# 1. Write a Brief, Not a Prompt\n\n**❌ Bad (vague, wastes tokens on clarification):**\n\n    Write me a marketing email.\n\n**✅ Good (token-efficient brief):**\n\n    Role: Senior B2B copywriter, 12+ yrs SaaS experience, specializes in conversion copy and enterprise sales cycles.\n    Task: Cold email to book a product demo.\n    Audience: HR Directors at 200–500 person companies.\n    Format: Subject line + 3 paragraphs, max 160 words.\n    Avoid: Bullet points, \"synergy\", \"game-changer\", opener compliments.```\n    \n    **Formula → Role + Expertise + Task + Audience + Format + Constraints**\n    \n    One good brief = better output in fewer tokens than 3 rounds of back-and-forth.\n    \n    ---\n    \n    ### 2. Expert Role Prompting — Qualify the Skills\n    \n    Most people write `\"Act as a doctor\"`. That's weak. The model needs to know *what kind* of expert and *how deep*.\n    \n    **Template:**\n    \n    ```text\n    You are a [title] with [X] years of experience in [specific domain]. You have deep expertise in [sub-speciality]. You have [specific credentials/background]. You approach problems by [method/framework].\n\n**Examples:**\n\n**Medical:**\n\n    You are a board-certified internal medicine physician with 18 years of clinical experience, including 6 years as a hospitalist. You specialize in diagnostic reasoning and evidence-based medicine. You cite current clinical guidelines (AHA, USPSTF, UpToDate) and flag when a finding warrants specialist referral.\n\n**Legal:**\n\n    You are a corporate attorney with 15 years specializing in SaaS contracts and IP law. You've advised Series A–C startups. You flag jurisdictional nuances, identify common red-flag clauses, and always note when something requires jurisdiction-specific counsel.\n\n**Finance:**\n\n    You are a CFA charterholder with 20 years in equity research and portfolio management. You use DCF, comps, and scenario analysis. You always distinguish between what data supports vs. what is speculative.\n\n**Dev/Coding:**\n\n    You are a senior software engineer with 12 years building production-grade Node.js/TypeScript systems. You prioritize readability, performance, and security. You always flag edge cases, write tests, and mention breaking changes.\n\nThis single upgrade produces dramatically more precise, nuanced outputs.\n\n# 3. Negative Constraints > Positive Instructions\n\nTell it what NOT to do. This is more token-efficient and more reliable than over-describing what you want.\n\n    Do not use bullet points.\n    No preamble or opener.\n    Don't summarize at the end.\n    Avoid passive voice.\n    No longer than 200 words.\n    Don't start with \"Certainly\" or \"Great question\".\n\nStack 3–5 per prompt. Constraints cut output bloat and reduce your need to re-prompt.\n\n# 4. \"Ask Me Questions First\"\n\nAdd this before any complex task:\n\n    Before you start, ask me 3 clarifying questions that would help you produce a better output.\n\nWhy it works: surfaces assumptions early, fills context gaps, eliminates re-prompting. One extra message saves 3–4 correction rounds. Best for: briefs, plans, landing pages, code architecture, long-form writing.\n\n# 5. Multi-Pass > One-Shot\n\nStop expecting a perfect single output. Work in rounds:\n\n* **Pass 1 →** \"Give me a tight outline only.\"\n* **Pass 2 →** \"Write section 2 in full. Tone: direct, no fluff.\"\n* **Pass 3 →** \"Add a real-world example to paragraph 3.\"\n* **Pass 4 →** \"Tighten the whole thing to 75% of current length.\"\n\nEach pass is a small, focused token request. Total cost is lower than one massive prompt + correction loop.\n\n# 6. Token-Efficient Prompting Rules\n\n* **Be specific, not long.** A 40-word precise brief beats a 200-word vague one.\n* **Reference, don't repeat.** After sharing context once, use \"using the above\" or \"based on what we discussed\".\n* **Request compressed outputs.** End prompts with: `Be concise. Cut anything that doesn't add value.`\n* **One task per prompt.** Combining tasks produces longer, lower-quality outputs.\n* **Use symbols as shorthand.** `→` for flow, `//` for separation, `[]` for fill-in slots. Saves tokens on explanation.\n\n# 7. Few-Shot (Show Don't Tell)\n\nPattern-matching beats description. Instead of describing a tone:\n\n    Here are 2 examples of the writing style I want:\n    [Example 1]\n    [Example 2]\n    Now write a third in the same style about [topic].\n\nWorks for: copy style, code style, report format, email tone. Use your own past writing or published samples you admire.\n\n# 8. Steelman Prompting (Adversarial Analysis)\n\nSkip the validation loop. Get real analysis in one exchange:\n\nHere is my plan: \\[X\\].\n\n1. Give me the 3 strongest reasons this will fail. Be specific, not generic.\n2. Then steelman the best case for why it succeeds despite those risks. No softening. No \"it depends.\"\n\nMore useful than asking \"what do you think?\" which produces agreeable non-answers.\n\n# 9. Reverse-Engineer Great Outputs\n\nFind an output (ad, article, email, code) you think is excellent. Paste it and ask:\n\n    What prompt template would produce this? Make it reusable with [brackets] for variables.\n\nFastest way to build a personal prompt library from real examples instead of theory.\n\n# 10. Structured / JSON Output\n\nWhen piping output into tools or workflows:\n\n    Respond ONLY in valid JSON. No preamble, no markdown, no fences.\n    Schema: { \"title\": string, \"tags\": string[], \"summary\": string, \"priority\": \"high\"|\"medium\"|\"low\" }\n\nAlso works for tables, CSV, XML. Be explicit about schema — ambiguous requests produce inconsistent structure.\n\n# ⚙️ FEATURES — USE THESE OR YOU'RE LEAVING VALUE ON THE TABLE\n\n# 11. Custom Instructions — Set Once, Works Forever\n\n**Settings → Personalization → Custom Instructions**\n\nWrite a global brief that loads automatically in every chat:\n\n    I am a [role]. I prefer [response style]. Skip preambles and openers. Never summarize at the end. When recommending tools, include the current version. Default to [language/framework] for code.\n\n5 minutes of setup. Eliminates re-stating context in every conversation. Massively reduces token waste per session.\n\n# 12. Projects — Persistent Workspaces (Most Ignored Feature)\n\nProjects (left sidebar) = persistent system prompt + file library + memory, loaded automatically in every conversation inside that project.\n\n**Set up one project per:**\n\n* Client or account\n* Business you run\n* Active codebase (load your docs, stack, conventions)\n* Research topic (load sources, notes, scope)\n\nNot a folder. A living workspace where context never resets.\n\n# 13. Memory — Feed It, Don't Just Leave It On\n\nMemory stores facts across sessions but most people treat it as passive. Use it actively:\n\nRemember: My company sells B2B SaaS. My audience is non-technical ops managers. I always want responses under 200 words unless I ask for more.\n\n**Audit it:** Settings → Personalization → Memory. Delete stale entries. Bad memory = consistently off-base outputs. Good memory = saved setup tokens on every single session.\n\n# 14. Deep Research — Use It Right or Waste Your Quota\n\nBrowses and synthesizes dozens of sources into a cited report. Takes 5–30 minutes. Replaces hours of tab-switching.\n\n**How to prompt it well:**\n\n❌ `Research content marketing strategies.`\n\n✅ `Compare content marketing ROI across B2B SaaS companies with <50 employees vs 50–200 employees. Focus on: organic SEO vs LinkedIn vs email. Include data from 2024–2026. Prioritize original studies over blog aggregators.`\n\n* Interrupt mid-research if it's drifting — don't wait for the full wrong report\n* Restrict to trusted sources via MCP connectors (Settings → Apps)\n* Follow up inside the same Project to go deeper without losing context\n\n**Quota:** Plus = 10/month, Pro = 250/month, Free = 5/month.\n\n# 15. Agent Mode — Delegate Tasks, Not Just Questions\n\nGives ChatGPT a real browser + terminal. It clicks, scrolls, fills forms, downloads files, and interacts with connected apps.\n\n**Prompting Agent Mode:**\n\nGoal: \\[End state you want, not step-by-step instructions\\]  \nConstraints: Do not submit any form without showing me a preview first. Do not make purchases. Stop if you hit a login wall and ask me.\n\nLet it figure out the steps. Your job is to define the destination and the guardrails.\n\nBest for: competitive research, form filling, multi-step web workflows, GitHub/Gmail/Drive tasks via connected apps.\n\n# 16. Tasks — Scheduled Prompts (Paid Plans)\n\nSet a prompt to run automatically on a schedule, no babysitting required.\n\n**Examples:**\n\n* Every Monday 8am → summarize top industry news into a 5-bullet brief\n* Every Friday → compile this week's commits and draft a team update\n* 1st of month → pull project notes and generate a status report\n\n**Where:** Sidebar → Tasks. Plus/Pro only.\n\n# 17. Native App Connectors (60+)\n\nConnect Google Drive, GitHub, Slack, Dropbox, Gmail, and 55+ more. Reference your actual data instead of copy-pasting:\n\nPull the latest version of Q3-Report.xlsx from Google Drive and identify the top 3 cost centres.\n\nCheck connector permissions before enabling — some take actions, not just read data.\n\n# 🎯 EXPERT PROMPT TEMPLATES (COPY-PASTE READY)\n\n**Medical / Health:**\n\n    You are a board-certified physician with 15+ years clinical experience in [specialty]. Cite current guidelines. Flag anything requiring in-person assessment. Be precise, not cautious to the point of uselessness.\n\n**Legal:**\n\n    You are a practicing attorney with 12+ years in [law area]. Jurisdiction: [country/state]. Flag clauses that create liability, identify missing protections, and always note when local counsel is essential.\n\n**Finance / Investment:**\n\n    You are a CFA with 20 years in [equity/VC/real estate]. Analyze using [DCF/comps/scenario modelling]. Separate what data supports from what is assumed. Include downside risk.\n\n**Copywriting:**\n\n    You are a direct-response copywriter with 10+ years writing for [industry]. You've generated [X] in revenue from email campaigns. Write copy that converts, not impresses. No fluff, no passive voice.\n\n**SEO:**\n\n    You are an SEO strategist with 10+ years ranking competitive terms. You use search intent analysis, topical authority, and E-E-A-T signals. Prioritize what moves rankings, not what sounds strategic.\n\n**Dev / Engineering:**\n\n    You are a senior [stack] engineer with 10+ years in production systems. Prioritize security, performance, and maintainability. Include edge cases. Flag deprecated methods. Write tests.\n\n# ✅ TL;DR — THE 2026 STACK\n\n|What|Why|\n|:-|:-|\n|**Custom Instructions**|Global context, zero repeated setup|\n|**Projects**|Persistent workspace per use case|\n|**Expert Role Prompts**|Depth + precision in outputs|\n|**Negative Constraints**|Reliable, bloat-free responses|\n|**Multi-Pass**|Better quality, lower total token cost|\n|**\"Ask Questions First\"**|Kills re-prompting loops|\n|**Deep Research**|Multi-source synthesis, cited|\n|**Agent Mode**|Real-world task delegation|\n|**Tasks**|Recurring prompts on autopilot|\n|**Memory**|Persistent context, actively managed|\n|**Connectors**|Your live data in every chat|\n\n*GPT-4o, o3, o4-mini — all retired from ChatGPT consumer UI as of Feb 2026. You're on GPT-5 now. Update your habits accordingly.*","offTopic":false},{"id":"b30396ba-f1d0-4d2f-be1a-e20ab5f30989","excerpt":"I tested 100+ prompts over 3 months these 7 are the ones I actually use every single day — I got tired of getting generic, boring outputs from ChatGPT. So I spent the last few months building and testing prompts obsessively   tweaking, rewriting, and stress-testing them across GPT-3.5 and GPT-4.\n\nOut of 100+ prompts, t","url":"https://www.reddit.com/r/ChatGPT/comments/1r4azlh/i_tested_100_prompts_over_3_months_these_7_are/","role":"pain","weight":1.0623976,"occurredAt":"2026-02-14T04:22:29.000Z","sourceKey":"reddit","sourceName":"Reddit","credibility":0.62,"venue":"ChatGPT","intent":"problem_report","painScore":0.4690909,"sentiment":-0.27272728,"confidence":0.72316664,"matchedPatterns":["manual_process","product:chatgpt"],"statement":"Each piece should feel native to the platform not just a copy-paste.","title":"I tested 100+ prompts over 3 months these 7 are the ones I actually use every single day","body":"I got tired of getting generic, boring outputs from ChatGPT. So I spent the last few months building and testing prompts obsessively   tweaking, rewriting, and stress-testing them across GPT-3.5 and GPT-4.\n\nOut of 100+ prompts, these 7 are the ones I literally cannot work without anymore. Sharing them because they genuinely changed how I use this tool.\n\n1\\. The \"Brutal Honesty\" Feedback Prompt\n\nYou are a brutally honest consultant with 20 years of experience. I'm going to share my \\[business idea / essay / plan]. Your job is to tear it apart. Find every weakness, every flaw, every assumption I'm making. Don't be polite. After listing the problems, give me a prioritized action plan to fix each one.\n\nThis one saved me from launching a terrible landing page. ChatGPT usually says \"great idea!\"   this prompt forces it to actually think critically.\n\n2\\. The \"Learn Anything in 20 Minutes\" Prompt\n\nYou are an expert teacher who specializes in explaining complex topics to smart beginners. Teach me \\[topic] using this structure: 1) Explain it like I'm 15 years old in 3 sentences, 2) Now explain the nuance an expert would understand, 3) Give me 3 real-world examples, 4) Give me the 3 biggest misconceptions people have about this, 5) Tell me what to learn next if I want to go deeper.\n\nI used this to understand blockchain, options trading, and cognitive behavioral therapy in one afternoon. The \"misconceptions\" section is shockingly good.\n\n3\\. The \"Steal My Competitor's Strategy\" Prompt\n\nAnalyze the brand \\[competitor name + URL if possible]. Based on publicly available information, break down: 1) Their likely target audience, 2) Their positioning and unique value proposition, 3) Their content strategy and what topics they focus on, 4) Their pricing psychology, 5) 3 weaknesses I could exploit if I were competing with them. Present this as a strategic briefing.\n\nThis feels like having a $5,000 strategist in your pocket. I ran this on 4 competitors and found a gap in the market none of them were covering.\n\n4\\. The \"One-Person Marketing Team\" Prompt\n\nYou are a senior marketing strategist, copywriter, and content planner rolled into one. My product is \\[describe product, audience, and price]. Create a full 7-day marketing plan that includes: daily social media posts (written out in full, ready to post), 2 email sequences (welcome + sales), 3 hook ideas for short-form video, and a content calendar for the next 30 days. Make everything specific   no generic advice.\n\nIf you're a solo creator or small business owner, this one prompt replaces hours of planning. The key is giving it enough context about your product.\n\n5\\. The \"Decision Maker\" Prompt\n\nI need to make a decision about \\[describe situation]. Act as a strategic advisor and do the following: 1) List the options I have (including ones I might not have considered), 2) For each option, give me the best-case scenario, worst-case scenario, and most likely scenario, 3) Identify my hidden biases based on how I described the situation, 4) Give me your final recommendation with reasoning, 5) Tell me what question I should be asking myself that I'm not.\n\nThat last line   \"what question should I be asking that I'm not\"   consistently blows my mind. It catches blind spots I didn't know I had.\n\n6\\. The \"Content Repurposer\" Prompt\n\nI have one piece of content: \\[paste your blog post, video script, or article]. Repurpose it into all of the following: 1) A Twitter/X thread (8-12 tweets, with hooks), 2) A LinkedIn post (professional tone, storytelling format), 3) An Instagram caption (casual, with emojis and CTA), 4) 3 short-form video script ideas (under 60 seconds each), 5) An email newsletter version. Each piece should feel native to the platform   not just a copy-paste.\n\nOne blog post becomes 7+ pieces of content. I use this every single week.\n\n7\\. The \"Second Brain\" Organizer Prompt\n\nI'm going to paste my rough notes, ideas, and scattered thoughts below. Your job is to: 1) Identify the core themes and group them, 2) Turn messy bullet points into clear, actionable items, 3) Highlight the 3 most important ideas and explain why they matter, 4) Suggest connections between ideas I might not have noticed, 5) Create a clean, organized summary I can reference later.\n\nHere are my notes: \\[paste notes]\n\nI dump my messy voice memos and random notes into this every Friday. It turns chaos into clarity in 30 seconds.\n\nThese are just a fraction of the ones I use daily. I've been building a whole system of these organized by category   productivity, marketing, writing, business strategy, learning, and more.\n\nHappy to share more if people find these useful. What prompts do you all keep coming back to?\n","offTopic":false},{"id":"1a0cb323-8b84-4f43-b523-b1f85abb421d","excerpt":"before you write your next prompt read this first — I wasted probably 6 months prompting chatGPT like a Google search. Just typing what I wanted and hoping for the best. The output was always... fine. Not bad enough to complain about but not good enough to actually use.\n\nWhat changed things for me wasn't learning more ","url":"https://www.reddit.com/r/ChatGPT/comments/1sexhim/before_you_write_your_next_prompt_read_this_first/","role":"pain","weight":0.95706666,"occurredAt":"2026-04-07T14:24:34.000Z","sourceKey":"reddit","sourceName":"Reddit","credibility":0.62,"venue":"ChatGPT","intent":"problem_report","painScore":0.48,"sentiment":0.2,"confidence":0.64666665,"matchedPatterns":["frustrating","product:chatgpt"],"statement":"Ask.\"* Sounds annoying.","title":"before you write your next prompt read this first","body":"I wasted probably 6 months prompting chatGPT like a Google search. Just typing what I wanted and hoping for the best. The output was always... fine. Not bad enough to complain about but not good enough to actually use.\n\nWhat changed things for me wasn't learning more about AI, it was learning to use AI to fix my own prompts before I even submitted them. These aren't prompts that do a task. These are prompts you use *on* your prompts.\n\nHere's what I actually use:\n\n**1. The 95% Confidence Drill** Before you ask anything complex, paste this first: *\"Before you respond, ask me clarifying questions until you're 95% confident you fully understand what I need. Don't guess. Don't fill in gaps. Ask.\"*\n\nSounds annoying. Changes everything. Especially for anything creative or strategic.\n\n**2. The Assumption Exposer** After you write a prompt, run it through this: *\"Read this prompt and list every assumption you'd have to make to answer it. Then rewrite the prompt so none of those assumptions are left up to you.\"*\n\nMost prompts are full of gaps we don't even notice. This finds them.\n\n**3. The Expert Panel Reframe** *\"Rewrite this prompt as if it were being asked by a senior \\[role\\] to a team of specialists. Add the context, constraints, and output format they would naturally include.\"*\n\nWhatever role fits your use case. I use this and it's night and day vs what I'd write myself lol.\n\n**4. The Anti-Vague Pass** *\"Identify every vague or subjective word in this prompt — words like 'good,' 'professional,' 'detailed,' 'better.' Replace each one with a specific, measurable alternative.\"*\n\nThis one alone cut my back and forth in half.\n\n**5. The Constraint Injector** *\"Take this prompt and add 3 constraints that would make the output more focused, actionable, and harder to misinterpret.\"*\n\nYou'd be surprised what constraints the AI suggests that you never would've thought of.\n\nThese are the kind of prompts I keep going back to. They work across use cases and doesn't matter if you're using ChatGPT for content, code, analysis, whatever.","offTopic":false},{"id":"6608aab6-5464-4064-bf86-59a343cc45c1","excerpt":"This Deep Truth Mode Prompt makes AI question everything - including its own training data - and prove its claims.  This prompt stops AI from making things up or just giving the consensus story. — TLDR\n\nMost AI outputs default to the safest, mainstream summary. Deep Truth Mode is a forensic prompt protocol that forces ","url":"https://www.reddit.com/r/promptingmagic/comments/1pnruuw/this_deep_truth_mode_prompt_makes_ai_question/","role":"request","weight":0.92924434,"occurredAt":"2025-12-16T03:38:01.000Z","sourceKey":"reddit","sourceName":"Reddit","credibility":0.62,"venue":"promptingmagic","intent":"feature_request","painScore":0.36,"sentiment":0.23529412,"confidence":0.6832679,"matchedPatterns":["missing_feature","product:chatgpt"],"statement":"It is a structured way to reduce consensus autopilot, surface missing data, and produce a clear what-would-change-my-mind test plan.","title":"This Deep Truth Mode Prompt makes AI question everything - including its own training data - and prove its claims.  This prompt stops AI from making things up or just giving the consensus story.","body":"TLDR\n\nMost AI outputs default to the safest, mainstream summary. Deep Truth Mode is a forensic prompt protocol that forces the model to (1) steel-man the mainstream view, (2) steel-man the best dissenting view using primary evidence, (3) generate a third hybrid hypothesis, then (4) aggressively red-team all three and keep only what survives. It is not a truth machine. It is a structured way to reduce consensus autopilot, surface missing data, and produce a clear what-would-change-my-mind test plan.\n\n**Putting AI into Deep Truth Mode**\n\nEveryone has seen it:\n\nYou ask AI a hot topic and get a clean, confident, consensus-flavored answer that feels true mainly because it sounds official.\n\nThat is not always bias. Sometimes the consensus is right. The problem is the default behavior:\n\n* Summarize what most sources say\n* Smooth over uncertainty\n* Avoid uncomfortable counterclaims\n* Skip the real work: primary evidence, incentives, and falsification\n\nDeep Truth Mode flips the workflow. Instead of asking the model to be correct, you force it to be adversarial, evidence-seeking, and falsifiable.\n\n**What Deep Truth Mode is actually good for**\n\n* Controversial topics where the facts are messy and incentives matter\n* Fast sanity checks on narratives that feel too neat\n* Finding the missing dataset, missing experiment, or missing disclosure that would settle the dispute\n* Turning opinion fights into testable claims\n\n**What it is not good for**\n\n* Replacing domain experts on medical, legal, or safety-critical decisions\n* Proving your favorite theory\n* Anything where you cannot or will not verify sources\n\nAlso: do not fetishize chain-of-thought. Models can produce reasoning that looks rigorous without being faithful to how they got there. Even the top labs have published research showing chain-of-thought can be unreliable as a window into model intent. \n\n# The Deep Truth Mode prompt\n\nCopy/paste this as your user prompt. It is designed to work across ChatGPT, Claude, Gemini, Grok, and Perplexity with minimal edits.\n\nDEEP TRUTH MODE: forensic analysis protocol\n\nTopic under investigation:  \n<insert topic>\n\nGoal:  \nReduce consensus autopilot. Generate competing hypotheses. Attack them. Keep only what survives. Use evidence-first reasoning.\n\nRules:  \n\\- If the topic is ambiguous, ask up to 3 clarifying questions, then proceed with stated assumptions.  \n\\- Prefer primary sources (datasets, filings, transcripts, court records, standards, original papers, patents). Use secondary sources only as pointers to primary evidence.  \n\\- Do not claim a source supports something unless you can quote a short excerpt (max 25 words) or precisely reference the relevant section.  \n\\- If browsing is unavailable, do not invent citations. Instead output a To Verify list with exact search queries and what you expect to find.  \n\\- Separate facts, interpretations, and speculation with labels.\n\nOutput format: run steps 1–8 in order and label each step.\n\n1. Consensus Fortress\n2. \\- State the strongest mainstream position in 5–10 bullets.\n3. \\- List the common labels used against dissenting views (for context only).\n4. \\- Provide 5–10 primary or highest-quality references that support the mainstream position.\n5. Incentive and Constraint Audit\n6. \\- Map money, power, and constraints on all sides:\n7. funding, regulation, career incentives, litigation risk, data access, measurement limitations.\n8. \\- Only include specific claims with references; otherwise mark as unknown.\n9. Parallel Steel-Man Tracks\n10. Track A: strongest dissenting position using primary evidence\n11. Track B: strongest mainstream position without appeals to authority, only evidence and logic\n12. Track C: best hybrid or third hypothesis that explains anomalies on both sides\n13. For each track:\n14. \\- Core claim (1 paragraph)\n15. \\- Best evidence (bullets + references)\n16. \\- Key assumptions (bullets)\n17. Red-Team Round\n18. For each track, generate the 5 strongest attacks:\n19. \\- falsifying evidence\n20. \\- internal contradictions\n21. \\- statistical or measurement failure modes\n22. \\- alternative explanations\n23. Surviving Fragments\n24. List only the claims from each track that survive the red-team attacks.\n25. Rank by evidential strength.\n26. Falsification Pathways\n27. For the top 2–3 surviving hypotheses:\n28. \\- One decisive test or dataset that would most efficiently falsify it\n29. \\- What result would change your mind\n30. Meta-Analysis of Silence\n31. What critical data is missing or rarely discussed?\n32. Give plausible reasons (benign and non-benign), clearly labeled as hypotheses.\n33. Final Verdict\n34. \\- Probability distribution across the surviving hypotheses\n35. \\- Top 3 reasons for the probabilities\n36. \\- Biggest uncertainty and how to resolve it\n37. \\- A short, practical takeaway: what a careful person should believe or do next\n\n**Will this work in ChatGPT, Claude, Gemini, Grok, Perplexity?**\n\nYes, with caveats. Here is the honest compatibility map.\n\nChatGPT\n\n* Works well if you turn on deeper reasoning and web browsing when you need real sources. ChatGPT includes a thinking-time control and multiple modes; use Thinking or Pro for this type of work.\n* Important: do not demand hidden reasoning dumps. Ask for evidence tables, assumptions, and what would falsify the claim.\n\nClaude\n\n* Works very well for structured argumentation and red-teaming. Claude has an extended thinking mode you can toggle for deeper work.\n* Still: verify sources. Treat the output as a research brief, not proof.\n\nGemini\n\n* Works well, especially with Gemini 3 style deeper reasoning modes and Gemini API thinking controls.\n* Best practice: request grounded citations and ask it to clearly label what is verified vs inferred.\n\nGrok\n\n* Works well for adversarial synthesis. Grok 4 is positioned as a reasoning model, and Grok 4 Heavy emphasizes parallel test-time compute, which maps nicely to multiple competing hypotheses.\n* Tip: keep the structure tight and demand source-quality discipline.\n\nPerplexity\n\n* Works, but you must understand a quirk: Perplexity’s real-time search component does not follow the system prompt. Put the protocol in the user prompt and restate your rules inside the prompt itself.\n* Upside: Perplexity is built around providing cited sources you can click and verify.\n\nIf you want more prompts like this, I keep a free library at [PromptMagic.dev](http://PromptMagic.dev) so you can save, organize, and reuse them without losing your best workflows.","offTopic":false}],"breakdown":[{"sourceKey":"reddit","sourceName":"Reddit","count":4}],"total":4}}