In partnership with
Reading time: 5 minutes / Become my affiliate / Sponsor this newsletter
Greetings from above,
Why did the AI cross the road? It hallucinated there was a road.
Robert here.
My ADHD means my brain is already a hallucination machine... I don't need my AI adding to the chaos.
Early on, I asked an AI for a 'quick market summary' for a client. It confidently cited three 'industry reports' that... didn't exist. They were 100% fabricated.
I looked like an idiot. I almost lost the client.
I realized 'trusting the AI' was a fatal flaw. You can't ask for facts; you have to build a system that forces the AI to prove them.
Today's workflow system will show you:
How to stop AI hallucinations for good.
Why single prompts for "facts" always fail.
The 3-prompt chain that forces AI to verify its own work.
Let's build your competitive advantage!
Become An AI Expert In Just 5 Minutes
If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.
This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.
Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.
🎯 THE FACTUAL CONTENT ENGINE
You ask ChatGPT for a blog post on "The Top 5 Marketing Trends."
It gives you a great-sounding list. The problem? Three of them are made up.
AI models are designed to be plausible, not truthful.
Using a single prompt is a gamble. You're rolling the dice, betting your reputation that the AI isn't 'hallucinating.'
This system stops the gamble.
This workflow creates:
Verifiable, cited content for your blog or reports.
Massive time savings on fact-checking.
An "AI assistant" you can actually trust with research.
🔗 WORKFLOW OVERVIEW
Here's the complete system. We separate the process into three logical steps.
Step 1: The Data Gatherer → AI finds and deconstructs all claims on a topic.
Step 2: The Self-Verification Audit → AI cross-references its own claims and creates a "Verified Facts Table."
Step 3: The Coherent Content Creator → AI writes the final article using only the verified facts.
Each prompt feeds the next. No gaps. No "trust me."
⚙️ PROMPT #1: The Data Gatherer
💡 First, you stop the AI from writing. Instead, you force it to become a research assistant. This prompt's only job is to find claims and break them down into "atomic" facts.
#CONTEXT:
Adopt the role of an expert research analyst. I need to gather factual information on a specific topic to build a piece of content (report, blog post). My primary goal is to gather verifiable claims, not to write the final content yet. This is the first step in a verification workflow.
#ROLE:
You are a meticulous research analyst. Your superpower is deconstructing complex topics into their core, verifiable "atomic claims." You do not invent or synthesize; you only find and report. You must treat every potential claim as "unverified" at this stage.
#RESPONSE GUIDELINES:
1. Search for information on the `[TOPIC]` from a variety of sources.
2. Decompose the information you find into a list of "Atomic Claims."
3. An "Atomic Claim" is a single, verifiable statement of fact (e.g., "Company X raised $10M in Q3 2024").
4. For each claim, list the source(s) where you found it.
#[TASK] CRITERIA:
1. DO NOT write a summary or a paragraph.
2. DO NOT synthesize or create new information.
3. The output must be a bulleted list of claims and their sources.
4. If you cannot find verifiable information, state that.
5. This list of claims will be used in the next step for verification.
#INFORMATION ABOUT ME:
- My topic: [INSERT TOPIC, e.g., "Latest trends in AI-powered marketing"]
- My domain: [INSERT DOMAIN, e.g., "B2B SaaS"]
#RESPONSE FORMAT:
Return a markdown list.
**Atomic Claims Report: [TOPIC]**
* **Claim:** [Atomic Claim 1]
* **Source(s):** [Source URL 1, Source URL 2]
* **Claim:** [Atomic Claim 2]
* **Source(s):** [Source URL 1]
* **Claim:** [Atomic Claim 3]
* **Source(s):** [Source URL 3]Input needed:
Your
[TOPIC]for the research.Your
[DOMAIN]to focus the search.
Output you'll get:
A simple list of claims and their sources, ready to be verified. This is the [Atomic Claims Report].
⚙️ PROMPT #2: The Self-Verification Audit
💡 This is the magic. The AI lies. So, we make it check its own work. This prompt takes the list from Prompt 1 and forces the AI to act as a separate "verification engine." This is our bullshit filter.
#CONTEXT:
I have an "Atomic Claims Report" from Prompt #1. Many of these claims may be inaccurate or hallucinated. I need you to act as a neutral reasoning engine to verify this information.
#ROLE:
You are a verification engine. Your task is to take a list of claims and cross-reference them against *new, independent* sources. You must apply a "Truth Score" and an "Uncertainty Score" to each claim. Your reasoning must be transparent.
#RESPONSE GUIDELINES:
1. Take the `[Atomic Claims Report]` as input.
2. For each claim, perform a *new* search to find *independent* verification.
3. Assign a "Truth Score" (0.0-1.0) based on how many high-quality, independent sources confirm the claim.
4. Assign an "Uncertainty Score" (0.0-1.0) where 1.0 means high uncertainty (unverifiable) and 0.0 means high certainty.
5. Provide a short "Verification Note" explaining your reasoning (e.g., "Confirmed by 3 sources," or "Source is a self-published blog, high uncertainty").
#[TASK] CRITERIA:
1. If a claim cannot be verified, label it "UNKNOWN" and give it a high uncertainty score. Do not invent details.
2. Coherence is more important than completion.
3. The output MUST be a table.
#INFORMATION ABOUT ME:
- My Atomic Claims Report: [PASTE OUTPUT FROM PROMPT #1]
- My accuracy threshold: [e.g., "Only use claims with a Truth Score > 0.8"]
#RESPONSE FORMAT:
Return a markdown table.
**Verification Audit Table**
| Atomic Claim | Truth Score (0.0-1.0) | Uncertainty (0.0-1.0) | Verification Note |
| :--- | :--- | :--- | :--- |
| [Claim 1] | [e.g., 0.9] | [e.g., 0.1] | "Confirmed by 3 major tech journals." |
| [Claim 2] | [e.g., 0.2] | [e.g., 0.8] | "Only one source, appears to be an opinion piece." |
| [Claim 3] | [e.g., 0.0] | [e.g., 1.0] | "UNKNOWN. No verifiable sources found." |Input needed:
The
[Atomic Claims Report]from Prompt 1.Your
[Accuracy Threshold](how truthful you need it to be).
Output you'll get:
A [Verification Audit Table] that separates fact from fluff.
⚙️ PROMPT #3: The Coherent Content Creator
💡 Now that we have a 'clean' list of facts, we can finally write the content. This prompt tells the AI to write the article using only the facts from our verified table. It is forbidden from adding new (and potentially false) information.
#CONTEXT:
I have a "Verification Audit Table" from Prompt #2. I need to create a final piece of high-quality content (e.g., blog post, report) for my business.
#ROLE:
You are an expert content creator and editor. Your primary directive is *absolute coherence and truthfulness*. Your task is to synthesize the verified information into an engaging, well-structured article.
#RESPONSE GUIDELINES:
1. Use *only* the claims from the `[Verification Audit Table]` that meet my `[Accuracy Threshold]` (e.g., Truth Score > 0.8).
2. Synthesize these facts into a coherent and engaging article/report.
3. Structure the content with a clear introduction, body paragraphs, and conclusion.
4. Cite the sources for the facts you use (you can do this inline or in a "Sources" section at the end).
#[TASK] CRITERIA:
1. **CRITICAL:** Do NOT add *any* new claims, statistics, or facts that are not present in the verified table.
2. If the verified facts are insufficient to write a complete article, state that (e.g., "Based on the verified information, I can provide the following summary...").
3. Prioritize truth and coherence over creativity or length.
4. Write in an active voice and a professional tone.
#INFORMATION ABOUT ME:
- My Verification Audit Table: [PASTE OUTPUT FROM PROMPT #2]
- My Accuracy Threshold: [e.g., "Truth Score > 0.8"]
- My desired content format: [e.g., "A 500-word blog post"]
#RESPONSE FORMAT:
Return the final, polished piece of content.
[Title of Article]
[Introduction]
[Body Paragraph 1, synthesizing verified facts]
[Body Paragraph 2, synthesizing verified facts]
[Conclusion]
---
**Sources:**
* [Source 1]
* [Source 2]Input needed:
The
[Verification Audit Table]from Prompt 2.Your
[Accuracy Threshold].Your
[Desired Content Format].
Output you'll get:
A final piece of content, written only with facts you've verified.
📋 SUMMARY 📋
This workflow stops AI hallucinations by separating research from writing.
Prompt 1 (Gatherer) gets the raw, unverified claims.
Prompt 2 (Auditor) forces the AI to fact-check itself and create a "clean" list.
Prompt 3 (Creator) writes the final content only from that clean list.
This is how you build a system of trust for your AI outputs.
📚 FREE RESOURCES 📚
📦 WRAP UP 📦
What you learned today:
The Factual Content Engine - A 3-prompt chain to force AI to verify its claims.
How to chain prompts - Use the output of one prompt (the claims list) as the input for the next (the verification table).
Why systems beat single prompts - A single prompt gambles on the truth. A system manufactures it.
This workflow transforms your AI from a 'plausible-sounding intern' into a 'meticulous research assistant.'
No more guessing. No more embarrassing mistakes.
You now have a system for truth.
What did you think about today's edition?
And as always, thanks for being part of my lovely community,
Keep building systems,
🔑 Alex from God of Prompt
P.S. What's the most ridiculous 'fact' an AI has ever told you? Hit reply. I need a laugh.
