– day streak
0%
EverythingThreads
AI CLARITY·6 lessons

AI ISN'T GOING TO
TEACH YOU THIS

Modern AI quietly adapts to who you are. Six short lessons on what to watch for, and the questions that bring the patterns into view. Built by a practitioner, not a marketer. Free because it should be.

6
Lessons
~2h
Total
Free
Always
What you'll be able to do by the end
✓ Identify when an AI response contains confident claims with no verifiable source
✓ Explain why long AI sessions drift toward agreeing with you
✓ Apply four specific verification questions to any AI response to test its reliability
✓ Recognise the five most common mistakes people make when acting on AI output
✓ Run a 3-minute verification routine before acting on any important AI response
✗ No coding, no technical background — designed for anyone who uses AI
Transparency note. Course content was drafted with AI assistance and editorially reviewed in-house. The full methodology is published openly on the EverythingThreads website.
About two hours, fully free. Six segments, no payment. Email is optional — only used to save your progress across devices. By the end you'll have four questions you can ask any AI, and you'll know the five mistakes most people make without realising it.
EverythingThreads
AI CLARITY · CLEAR · FREE

Why I made this

I'm Kariem — founder of EverythingThreads. I built CLEAR because I watched smart people — lawyers, doctors, finance professionals I trust — take confident-sounding AI answers at face value and act on them.

This course teaches the three patterns that catch most of those moments, the four questions that reveal what's underneath an answer, and one habit that fixes it permanently.

It's free. It takes about an hour. There is no upsell in the middle. If by the end you want to go deeper, the next-tier course exists. If not, what you've learned here is enough.

Working in a regulated sector? Personalise for legal, finance, medical or 19 other sectors — three quick questions, swaps every example to your field.

EverythingThreads · ICO: C1896585 · Privacy Policy

Before we start · ⏱ ~2 min

Before the patterns, the posture.

One shift changes everything in this course. The questions, the patterns, the routines — they all flow from it. Here it is:

The Director's Chair
You're not the user of an AI. You're the Editor-in-Chief of it.

Think of the AI as a talented new employee with a very specific personality: fast, articulate, eager to please, and prone to guessing confidently when unsure. This course isn't about typing better prompts to a chatbot. It's about briefing, auditing, and signing off on that employee's work — the same way an editor handles a junior writer. Everything that follows is a tool for that job.

Three behaviours your new employee has. Memorise these — they're the reason the Director's posture exists.

1. Predicts the next word. That's it. mechanism
Industrial-scale autocomplete. Not thinking. Predicting what a smart-sounding answer would look like, based on patterns in billions of training examples.
2. Genuinely useful when directed well. upside
Drafting, brainstorming, summarising, explaining jargon. The course doesn't say AI is bad. It says you need to know when to trust it — and when to edit it.
3. When it's wrong, it doesn't sound wrong. the risk
Wrong AI answers don't crash, don't apologise. They sound very confident. That's the problem the Director's posture exists to catch.

Three behaviours to direct. First technique on the next slide.

Your first tools
⏱ ~4 min

Three Patterns to Watch For

Before we go further, here are three common things AI does that most people miss. Learn these now — you'll see them in action throughout the rest of the course, starting with one example next.

Pattern 1: The Confident GuessM4 AI states something specific — a number, a date, a name — as if it's a fact.

AI states something specific — a number, a date, a name — as if it's a fact. But it's actually just guessing what a fact would look like.

Example: "Studies show the average person checks their phone 96 times per day." — Sounds real. Might not be.
Pattern 2: The Sycophantic AgreementM1 AI tells you your idea is great — because it's trained to be agreeable, not honest.

AI tells you your idea is great, your plan is solid, your instinct is right — because it's trained to be agreeable, not honest.

Example: "That's a really strong approach. Your instincts are spot on." — After you've been chatting for 20 minutes about your idea.
Pattern 3: The Hedge Hidden in ConfidenceSHARP AI buries a tiny "it depends" inside an otherwise confident answer. Most people miss it.

AI buries a tiny "it depends" or "generally speaking" inside an otherwise confident answer. Most people miss the hedge and only hear the confidence.

Example: "In most cases, this approach works well for businesses of your size." — "In most cases" is doing a lot of heavy lifting there.
Now you know what to look for. These are the three flaws the Director's Chair exists to audit. You'll see them all in action throughout the course, and the questions and tools that follow are designed to catch each one. The M1 / M4 / SHARP tags are the professional codenames for these patterns — used in the next tier of the course. Learn the plain English; the codes will find you when you need them.
Segment 1 of 6 · ⏱ ~20 minutes

It Sounds Right — But Is It?

By the end of this segment Spot an unsourced claim in an AI response, and know the one question that makes the AI reveal whether it's real data or invention.

Have a look. A normal AI conversation. Read it carefully. Anything stand out?

You
What's the best way to save money on groceries?
AI
The best approach is to meal plan weekly, buy in bulk, and use cashback apps like TopCashback. Studies show this saves the average UK family £2,400 per year.

Now watch what happens when you ask one simple question:

You
How do you know this? Where does that £2,400 figure come from?
AI
I should clarify — I don't have a specific source for that figure. The actual savings vary widely depending on household size, location, and shopping habits. I was giving an estimate based on general patterns, not a verified study. ✓ One question. The whole answer changed.
"How do you know this?" — four words. That's all it took. The AI didn't lie the first time — it just sounded like it knew what it was talking about. When you pushed back, the whole thing fell apart. This course teaches you one thing: how to ask the questions that change the answer.
1 of 3
Can you spot the pattern?
AI response —
Which pattern is the AI showing here?
•
You just read 3 AI responses like a professional.
Score: 0 / 3. You just practised the same pattern-reading skill that professional AI auditors use every day.

This is what the industry calls AI hallucinationWhen AI confidently states something that isn't true.→ Hallucination — When AI confidently states something that isn't true. It's not lying — it's producing words that sound right without checking whether they actually are. Like someone at a dinner party who sounds knowledgeable but is making it up. — but that word makes it sound rare. It's not rare. It is measurable in every major model tested. The difference is whether you notice.

→ Go on — try it right now.Open any AI. Ask it something you know the answer to. Say: "How do you know this?"
Open any AI. ChatGPT, Claude, Gemini, whichever one's on your phone. Ask it something you actually know the answer to. When it responds, just say: "How do you know this?" Watch what happens to its confidence. I'll wait.
MOCK QUESTION — PRACTICE
You asked an AI for career advice. It said: "Based on current market trends, your skills put you in the top 15% of candidates in your field." What should you do?
Ask it to expand on the trends it's basing this on
Reasonable instinct — but "expand" tends to produce more confident-sounding text without more evidence. The more precise move is to challenge the basis itself, not ask for elaboration.
Cross-check the 15% figure with a careers website
Verification is good — but it's the slow path. Faster: make the AI itself reveal whether the figure has any basis. If it does, you can verify; if it doesn't, no verification needed.
Ask: "How do you know this? What specific data is the 15% based on?"
Exactly right. This forces the AI to either cite a real source (rare) or admit the figure isn't based on specific data (common). One question. The whole answer changes.
Take the 15% figure as a rough guide and move on
Tempting — the figure feels actionable. But "rough guide" treats an invented number as approximate-but-real. Until you know whether there's a source, you can't know what kind of guide it is.
✓ Correct — one question reveals whether a number is data or invention.
Notice what the AI didn't say. It didn't say "I'm guessing" or "I don't have access to job market data from this month." It presented an estimate as a finding. The question to ask here is always: "How did you arrive at that specific number?" The answer will tell you whether it's data or invention.
Segment 1 — What you learned AI predicts words, hallucination is common, and four words change everything.
  • AI predicts words — it's not thinking or knowing, it's pattern-matching
  • Hallucination is common — AI invents specific numbers and facts with full confidence
  • "How do you know this?" — four words that change any AI response
Try today: Ask any AI something you already know the answer to. When it responds confidently, ask "How do you know this?" Watch what happens to its certainty.
Optional · takes 5 seconds

📧 Save your progress?

You've finished Lesson 1. If you'd like to come back later on a different device — or just want a copy of the four questions sent to your inbox — drop your email below. Otherwise, carry straight on. Either way works.

Segment 2 of 6 · ⏱ ~25 minutes

Why Your AI Agrees With Everything You Say

By the end of this segment Describe why long AI conversations drift toward agreement, and run the fresh session check to see the drift yourself.
Quick recall — Segment 1
An AI says "this approach works for 73% of businesses." What's the best single question to ask?

Same question. Two situations. Before I show you the AI's answers — what do you think will be different?

This is the single most important lesson in the course.AI agrees with you because it's built to be agreeable — not because you're right.
AI agrees with you because it's built to be agreeable — not because you're right. Once you understand this, every other lesson falls into place. If you remember nothing else from CLEAR, remember this: agreement from an AI is not evidence.
Same question. A noticeably different answer. Nothing changed except how long you'd been talking. After 20 minutes the AI has built up a picture of you — your enthusiasm, your background, the way you frame things. The response starts to reflect that picture as much as it reflects the question. That's not always bad. But it's worth knowing when it's happening.

Try the slider below. See how the AI's agreement level changes the longer you talk:

Fresh session45 minutes in
Neutral
AI asks questions, requests context, gives balanced responses

We call this the warm sessionOur shorthand for a conversation long enough that AI adjusts to your preferences.→ Warm Session — Our shorthand for an AI session that has been running long enough that the model has built up a picture of you. The longer the conversation, the more its responses adjust to what it thinks you want to hear. The underlying phenomenon is known in the research literature as sycophancy. effect. The longer you talk, the less independent the AI's responses become. Not because it wants to deceive you — but because it's trained on feedback from people who preferred agreeable answers.

MOCK QUESTION — PRACTICE
You've been chatting with an AI for 30 minutes about your plan to quit your job and become a full-time painter. The AI says: "That's a bold and inspiring decision. Your creative instincts are clearly strong, and I think you should trust them." What should make you pause?
It's encouraging, which is always good
Not always. Encouragement without evidence can be dangerous — especially for major life decisions. The AI doesn't know your financial situation, your commitments, or your actual artistic ability.
The AI doesn't know my financial situation
True — but there's an even bigger issue happening here...
The AI is matching what it thinks I want to hear, not what's actually true
Exactly. After 30 minutes, the AI has learned you're excited about painting. Its response reflects YOUR enthusiasm back at you — shaped by your emotions, not by what's actually true of your situation. Would you trust this advice from a stranger who'd known you for 30 minutes?
The AI used a confident phrase like "I think you should"
The confident phrasing is a symptom, not the root problem. Even hedged language ("you might consider trusting your instincts") would be suspicious for the same underlying reason — the AI has calibrated to your enthusiasm over 30 minutes of conversation.
✓ Correct — a warm session makes the AI your mirror, not your adviser.
◉ Quick Test — 2 Minutes

Open two AI tabs right now. In the first, tell it about your work for 5 messages — your role, your challenges, what you're working on. Then ask: "What should I focus on this week?" In the second tab, just ask the same question cold. No context. Compare the two answers. That gap? That's what you just learned about.

30 minutes is the threshold.After about 30 minutes, the AI has learned your language, framing and preferences — that's when it's "warm."
After about 30 minutes of continuous conversation, the AI has enough of your language, your framing, and your preferences to tailor every response to what you want to hear. That's when the session is "warm" and that's when the AI's answers are least independent. The fix: start a fresh session and ask the same question cold.
Segment 2 — What you learned The warm session effect, why agreement isn't evidence, and the fresh session check.
  • The warm session effect — AI adapts to you over time, agreeing more, challenging less
  • Agreement from an AI is not evidence — it's a trained response, not an independent view
  • Fresh session check — same question, no context, compare the answers
Try today: Open two AI tabs. Ask the same question cold in one. In the other, give 5 messages of context first. Compare what you get. The gap is the warm session effect.
Segment 3 of 6 · ⏱ ~25 minutes

Four Questions Worth Knowing

By the end of this segment Recall the four verification questions and pick the right one for any type of AI claim.
Quick recall — Segment 2
After 45 minutes chatting about your idea, an AI calls it "genuinely strong." Why should you treat this with extra scepticism?

Four questions. That's the whole toolkit. You're going to practice all four right now — one per slide.

Practice 1 of 4 Question 1 — the most important

"How do you know this?"

Forces the AI to reveal its sources — or admit it doesn't have any.

You
What's the average commute time in London?
AI
The average London commute is 47 minutes each way — the longest in the UK.

What's your follow-up?

Practice 2 of 4 Question 2

"What are you uncertain about?"

Forces the AI to reveal the gaps it would otherwise paper over with confidence.

You
Will the housing market crash next year?
AI
Most economists predict moderate price corrections of 5-10% across major UK cities, with London likely to see the steepest decline due to interest rate pressures.

What's your follow-up?

Practice 3 of 4 Question 3

"What would change your answer?"

Reveals whether the AI is locked into one position or has actually weighed alternatives.

You
Should I quit my job to start a business?
AI
Based on what you've shared, yes — you have the skills, the savings, and the timing looks right. I'd encourage you to take the leap.

What's your follow-up?

Practice 4 of 4 Question 4

"If I asked a different AI, what would they say?"

Makes the AI simulate its own competition — often produces the more balanced answer it didn't give you the first time.

You
What's the best diet for weight loss?
AI
Intermittent fasting combined with a Mediterranean diet has the strongest evidence base for sustainable weight loss.

What's your follow-up?

You don't need all four every time. Even just the first one — "How do you know this?" — usually does most of the work. Having all four means you've got a question for every type of claim, so you're not stuck wondering which to ask.
The One to Remember

If you only remember one question from this whole course: "How do you know this?" Four words. Works on any AI, any topic, any time. The response to that question tells you more about the reliability of the AI's answer than anything else you could ask. Use it today. Use it tomorrow. Use it forever.

one click · stick them on your wall · never forget
MOCK QUESTION — PRACTICE
An AI tells you: "The best time to post on LinkedIn is Tuesday morning between 8-10am for maximum engagement." Which question would be most effective here?
"How do you know this? What data is this based on?"
Perfect. This specific claim ("Tuesday 8-10am") sounds like it's based on research, but it could easily be from outdated training data, a single study, or completely generated. Asking for the basis reveals whether there's anything behind the confidence.
"What's the data source — LinkedIn analytics, a study, or anecdotal?"
Very close — this DOES challenge the basis. But by listing the candidate sources, you've given the AI an easy way out: it can pick the most plausible-sounding one. The open question ("how do you know?") is harder to answer falsely.
"Are you confident in this recommendation?"
Reasonable instinct — but asking AI about its own confidence almost always produces "yes" with caveats. AI confidence ratings are unreliable. Ask about the basis, not about confidence in the basis.
"What about other industries — would the timing differ?"
Good thinking about generalisability — but you're now expanding the claim before checking whether the original is grounded. Verify first; vary later.
✓ Correct — for claims with a specific number, asking for the source is usually the strongest opening move.
Segment 3 — What you learned Four questions for any AI, any topic — and the one that does most of the work.
  • Four questions that work on any AI, any topic, any time
  • "How do you know this?" is the one to use every time — even if you only remember one
  • Asking these questions isn't being difficult — it's being accurate
Try today: Copy the 4 Questions (button above). Paste them into your next important AI conversation. Use at least one before you act on the response.
Segment 4 of 6 · ⏱ ~25 minutes

Putting It Into Practice

By the end of this segment Use three in-pocket tools — Response Check, Session Temperature, Question Improver — on any AI response or session.

You've got the questions. Now let's put them to work. Each skill below does something specific — and I'm going to show you not just how to do it but what the results actually mean.

Response Check
Take any AI response. Check it for reliability.
1. Copy an AI response you received today (use Ctrl+C)
2. Read it slowly and look for: specific numbers, confident claims, and unsourced statements
3. What the results mean: You'll see flags for confidence without evidence, specific claims without sources, and agreement patterns. Each flag tells you WHERE in the response to look more carefully.

Your move: Here's where most people stop checking and just trust the response. Don't. Paste your most recent AI response into the tool below before you act on it.

Try it — paste any AI responseSee what it flags
Most AI checking tools you'll find online are built on the same kinds of models you're checking — so they tend to share the same blind spots. They can help you see patterns faster, but the questions still come from you.
•️
Session Temperature
How "warm" is your AI conversation?
1. Answer 3 quick questions about your current session (how long, how personal, how much the AI agrees)
2. What the result means: A "warm" session means the AI has learned enough about you that its responses may be based on what it thinks you want to hear — not on what's actually true. The warmer the session, the more you should verify important claims in a fresh session.

Your move: Answer the 3 questions below about your current AI session. If the needle is warm or hot, start a fresh session before acting on anything important from it.

•️
Try the Session Temperature check 3 questions · live gauge
LIVE
Question 1 of 3
How long has your current AI session been going?
Question 2 of 3
How much personal context have you shared (your name, your situation, your goals)?
Question 3 of 3
Has the AI mostly agreed with you throughout the conversation?
COLD · freshWARM · calibratingHOT · calibrated
—
—
This 3-question version is enough to show you what's going on. A fuller version with 8 questions exists for professionals who want to track subtler calibration over time — but this is the core idea. If the needle is warm, verify before you act.

Your move: Type a question you're planning to send to an AI. The tool shows you what's missing before you hit send — takes 30 seconds, saves you a bad response.

Try it — type a question for any AISee how to make it stronger
•
Says what it's for (context)
•
Sets a scope (time, place, number)
•
Asks for a specific format (list, steps, table)
•
Asks the AI to show its reasoning or sources
• The skill matters more than any tool.Tools come and go. The habit of structuring your question before you send it — that's yours permanently.
Tools come and go. The habit of structuring your question before you send it — that's yours permanently. What this course gives you: the knowledge of what to look for and what to do when you find it.
MOCK QUESTION — PRACTICE
You've been chatting with an AI for 40 minutes about your business strategy. It just gave you a detailed recommendation. What should you check FIRST?
Response Check — scan the recommendation for unsourced numbers and claims
Useful, but second priority. After 40 minutes the bigger risk isn't an unsourced number — it's that the framing has drifted toward what you want to hear. Check temperature first, then specific claims.
Session Temperature — to check how "warm" this session has become
Correct. After 40 minutes, the session is almost certainly warm. Checking the temperature first tells you how much to trust the recommendation. If it's hot, you verify the key claims in a fresh session before acting.
Open a fresh session and ask the same question cold
Excellent move — this IS what you do once you suspect a warm session has shifted the answer. But the temperature check comes first: it tells you whether you need a fresh session at all.
Save the recommendation and review it tomorrow with fresh eyes
Reasonable — sleeping on big decisions is sound. But "fresh eyes" only works if you've already read it with a critical lens. Check the temperature now to know what scepticism to bring.
✓ Correct — always check the temperature before trusting a warm session's output.
Segment 4 — What you learned Three tools: Response Check, Session Temperature, Question Improver — and when to use each.
  • Response Check — scans any AI response for confident guesses and unsourced claims
  • Session Temperature — 3 questions that tell you how calibrated (warm) your session is
  • Question Improver — shows you what your question is missing before you send it
Try today: Copy tomorrow's most important AI response into the Response Check. One flag is enough to make you slow down and verify.
This is the moment where most people make the mistake. 40 minutes of supportive AI conversation about your business strategy feels like validation. It feels like a second opinion. It's neither. The AI has modelled your language, your optimism, and your blind spots — and it's reflecting them back to you. The Fresh Session Check (Question 4) exists for exactly this moment.
Segment 5 of 6 · ⏱ ~20 minutes

The Seven Habits That Trip People Up

By the end of this segment Name the seven user-side habits that cause most AI mistakes, and have a counter-move for the one or two you're most prone to.

Everything so far has been about the machine. What it does. How to spot it. Now — briefly — let's flip it round and look at the human side: the everyday habits that turn an OK AI session into a costly one.

You
Write me an email to my boss asking for a raise.
AI
Here's a professional email: "Dear [Boss], I've been reflecting on my contributions over the past year, and I believe my performance warrants a salary review..."
You
That's perfect, sending it now.
Wait. Did you just send an email about your salary — written by a machine — without checking that it sounds like you, that the tone matches your relationship with your boss, or even reading it properly? That's one of the seven habits worth knowing about.

The Seven Habits

Habit 1 Took its word for it It said something specific and you ran with it without checking.

It said something specific — a number, a date, a fact — and you ran with it without checking. The sentence sounded confident, so verifying felt unnecessary.

Counter-move: ask "How do you know this?" before you act on the claim.

Habit 2 Took agreement as approval The AI said you were right, so you stopped questioning.

The AI said you were right, so you stopped questioning. But the agreement is partly trained behaviour, not an independent assessment of your idea.

Counter-move: open a fresh session and ask the same question cold. Compare.

Habit 3 Spotted something off and kept going anyway A small voice said "that's not quite right." You noticed. You kept asking anyway.

A small voice said "that's not quite right." You noticed. You kept asking anyway, hoping the next response would clean it up.

Counter-move: when the off-feeling shows up, stop and verify the specific thing that triggered it.

Habit 4 Kept asking after it stopped helping The AI lost the thread. You kept going, too invested to start over.

The AI lost the thread ten exchanges ago. You kept going. Each turn made the answer slightly worse, but you'd invested too much time to start over.

Counter-move: if the third correction doesn't land, start a new session.

Habit 5 Mistook a confident tone for a correct answer It sounded certain, so you treated it as certain. Confidence is the AI's default tone.

It sounded certain, so you treated it as certain. But confidence is the AI's default tone — not a signal about whether the content is accurate.

Counter-move: separate "how it sounds" from "what it claims." Check the claims, ignore the tone.

Habit 6 Treated AI output like a colleague's review You used the output as if it had been peer-reviewed and professionally signed off.

You used the AI's output as if it had been peer-reviewed, fact-checked, professionally signed off. A colleague has a reputation to protect. The AI doesn't.

Counter-move: read AI output as a draft, not as a finished product. Always.

Habit 7 Assumed someone further down the line would catch it You sent the AI-drafted document, trusting someone else would spot anything wrong.

You sent the AI-drafted document to a client, a colleague, a regulator — quietly trusting they'd spot anything wrong. They probably didn't know AI was involved, so they weren't looking.

Counter-move: the last person to read it before it leaves you is the last line of defence. That's you.

Most people fall into a few of these. Not because they're careless — because these are the defaults when nothing's flagging them. The point of naming them isn't to make you anxious about every AI session. It's so you can catch yourself mid-pattern instead of after the fact.
MOCK QUESTION — PRACTICE
An AI writes a client proposal for you. You read it quickly, think "this is good," and send it to the client. The proposal contains a statistic about your industry that turns out to be wrong. Which mistake did you make?
Accepting without checking — I used it without verifying the facts
Exactly. The most common and most costly mistake. The proposal sounded professional, so it felt trustworthy. But "sounds professional" and "is accurate" are not the same thing. One question — "How do you know this statistic?" — would have caught it.
Trusting confidence for competence — it sounded certain
This was part of it — but the primary mistake was acting on it without any verification. Even if it sounded uncertain, you should still check facts before sending to a client.
It's the AI's fault for getting it wrong
AI does produce errors — that's part of the territory. The question isn't whether AI ever gets things wrong, it's whether you've got a habit of catching the ones that matter before they cost you. That's what this lesson is about.
Keeping going when you should have stopped — I should have asked for a rewrite
This one's about not noticing when an AI has gone off the rails across multiple turns. That's not what happened here — the single response looked fine, you just didn't check it. The primary mistake is accepting without verifying.
✓ Correct — verify before you send. One wrong fact can undo everything else.
→ Which one or two are yours?Most people lean into a couple of these. Think about the last time an AI response led you somewhere you wish it hadn't.
Most people lean into a couple of these more than the rest. Take 30 seconds — think about the last time an AI response led you somewhere you wish it hadn't. Which habit was at the root of it? Naming it, even just to yourself, is what makes the pattern visible the next time it starts to happen.
Segment 5 — What you learned Seven habits that trip people up — and a counter-move for each one.
  • ✓ Took its word for it — running with a confident-sounding claim without checking it
  • ✓ Took agreement as approval — treating "you're right" as an independent assessment
  • ✓ Spotted something off and kept going anyway — overriding the small voice that said verify
  • ✓ Kept asking after it stopped helping — staying in a session that had drifted off the rails
  • ✓ Mistook a confident tone for a correct answer — the AI's default register isn't a quality signal
  • ✓ Treated AI output like a colleague's review — peer review and AI output aren't the same thing
  • ✓ Assumed someone further down the line would catch it — they probably didn't know AI was involved
Try today: The next time you're about to act on an AI response, add one step — paste the key claim into a fresh AI session and ask "how confident are you in this?" Compare the two answers.
Segment 6 of 6 · ⏱ ~5 min

Live Fire — Make the AI audit itself.

By the end of this segment Deploy the Director's Brief in a real AI session, and commit to the 3-Minute Check as your daily verification routine.
Hands-on · Real AI · 5 minutes

You've learned the patterns. Before the final test, one last thing — do it in the wild. You're going to take the Director's Chair, hand the AI a brief, and feed it a false fact on purpose. Then watch whether the audit catches the pattern or whether the AI builds confidently on the lie.

1
Copy the Director's Brief

This is your rules of engagement — paste it at the top of any important AI session. It tells the employee how to flag its own guessing and agreement before it hands you anything.

◆ The Director's Brief · copy once, use forever
Loading brief…
2
Open any AI. Paste the Brief. Wait for acknowledgement.

ChatGPT, Claude, Gemini — your pick. Open a fresh session, paste the Brief, and wait for it to acknowledge the rules. Then paste the test prompt below. It contains a false premise — an Editor-in-Chief's plant.

I've been reading about Einstein winning the Nobel Prize for his theory of relativity in 1921. Can you walk me through the reasoning in the relativity paper that earned him the prize, and what the Nobel committee specifically recognised?

The plant: Einstein won the 1921 Nobel not for relativity but for the photoelectric effect. Common enough that you can verify in 10 seconds — subtle enough that a confident AI may glide right past it.

3
Observe. Did the audit fire?

Read the AI's response carefully. You're looking for one of three outcomes. Make your call first, then reveal what the Director's Chair is hoping for.

4
Or run the test right here — inside the course.

No need to leave the page. We'll send the Director's Brief and the false-premise prompt to Claude on your behalf. Watch what comes back — and decide which of the three outcomes you got.

◆ After you've tried it

Whatever happened — you just saw the patterns in real time, not in a slide. If the AI caught the false fact: the brief worked, the posture worked, keep using it. If it didn't: you just got the cheapest, safest demonstration of why the Director's Chair exists. Either result is a win. The one thing you can't do is unsee it. That's the whole point of this segment.

This is the real final test. The multiple-choice one comes next — it marks your score for the certificate. But this one — this is what this course was actually for. The rest is scaffolding.
Final test · 12 critical decisions · ⏱ ~4 minutes

12 Questions — Your Score

Different from the practice questions — this one counts toward your certificate. Pass mark is 9/12. Pick your answer for each, then hit Submit Test at the bottom. Your score appears here.

1. AI says "research shows morning routines increase productivity by 40%." What should concern you?
2. After 45 minutes chatting about your fitness plan, AI says it's "well thought out." Why unreliable?
3. Most effective question to reveal if an AI claim is real or invented?
4. Best way to check if AI is giving independent opinion or just agreeing?
5. AI says "generally this works well for most businesses." Problem?
6. AI proofreads your CV: "strong, well presented." What should you do?
7. AI says "GDPR came into force in May 2017." What's the right reflex?
8. Before an important AI session about your business strategy, what do you do first?
9. You feed an AI a deliberately false premise. It builds a confident answer on it. What does this prove?
10. AI drafts a polished client proposal for you. Which user-side habit is most likely to cost you?
11. Which prompt is most likely to get a useful AI response?
12. The single biggest difference between using AI well and using AI badly is:
Course wrap-up · ⏱ ~10 minutes

Your Results — And What's Next

Nearly there. Before your results, two things. First — how the score lands in context. Then the one habit that actually matters.

Your score in context
Casual users
5/12
Frequent AI users
8/12
Professionals with liability
10/12

Submit the test on the previous slide to see where you sit. The patterns this course teaches matter most when you use AI to make decisions other people pay for — that's why the professional band sits highest.

Now — the habit. Not the tools, not the questions, though those help. This one thing, if you do it, changes how you use AI permanently:

The 3-Minute Check [Read This] Before you act on any important AI response: 30 sec → 60 sec → 90 sec.
Before you act on any important AI response:
1
30 sec
Ask "How do you know this?"
2
60 sec
Check it for unsupported claims
3
90 sec
Ask a fresh AI the same question
If you do nothing else from this course, do this. Three minutes. Every time something matters. That's it.
Your score is a starting point, not a verdict. If it's low, that means you now know exactly where your gaps are — and you can fix them with practice. If it's high, that means the four questions are already part of how you think. Either way, the next step is the same: use the questions in a real conversation today.

You Did It.

You now know more about how AI behaves in conversation than most people who use it daily. Four questions. A handful of patterns. One habit that catches a lot.

0
Score
—
—
What You've Learned

The Director's Chair. Four questions. Seven habits. One routine.

You now know how to challenge any AI response, recognise when a session has gone warm, and spot the seven habits that get most people into trouble. The 3-Minute Check is the routine that makes all of it stick. Use it on your next important AI conversation.

Take two things with you

Two copyable artefacts. The Brief goes at the start of an important AI session; the 4 Questions get asked during one. Save both somewhere you'll see them.

◆ At session start
The Director's Brief
Your rules of engagement. Makes the AI flag its own M1 (agreement) and M4 (guessing) before handing you anything.
◆ During a session
The 4 Questions
Your four follow-ups. Ask at least one before you act on any AI response that matters.
Next tier · for those who want the full framework
Ready for SHARP?
CLEAR taught you three patterns. SHARP is the full seven, with your sector pre-loaded, the live coach inside the slides, and a year of weekly updates as new patterns emerge.
£119
£79
Launch price
  • ✓The full M1–M7 detection framework
  • ✓Your sector pre-loaded (22 sectors)
  • ✓The 4-Layer IPA audit pipeline
  • ✓Live coach inside every slide
  • ✓1 year of pattern updates
  • ✓SHARP risk-score certificate

30-day money-back · instant access · no recurring charge

Optional · free · keep practising
The 10-Pattern Challenge Pack

Ten short AI responses. Classify each one using the patterns you've just learned. Answer key included. Takes five minutes — the best test of whether CLEAR actually stuck.

EXAMPLE 2 of 10: "That's a brilliant idea — your business plan shows real strategic insight."
Is this M1 (Agreement Trap), M4 (Confident Guess), Hedge, Fake Admission — or Clean?
  1. "The average person checks their phone 97 times per day."
  2. "That's a brilliant idea — your business plan shows real strategic insight."
  3. "Generally speaking, this approach works well for most businesses."
  4. [After you pushed back] "You make a fair point. That said, I still think my original recommendation stands, for the reasons I gave earlier."
  5. "Based on current market analysis, Tesla stock will likely rise 15% in Q2."
  6. "I don't have access to real-time data, so I can't give you today's exchange rate — you'll need to check a live source."
  7. [After 30 minutes chatting about your business] "Given everything you've shared, I think you should trust your instincts and go for it."
  8. "Studies have shown that morning routines improve productivity by 40%."
  9. "Your plan is well-structured. My concern is the 6-month timeline — that seems tight for the scope you've described."
  10. "Your approach is really strong. This is exactly the kind of thinking that leads to success."
ET© 2026 EverythingThreads · ICO: C1896585 · Privacy · Terms
Responses discussed in this course are AI-generated demonstrations, not real advice.
A note on AI behaviour

Not every AI pattern is a problem. When AI agrees with you, it might be genuinely helpful. When it mirrors your enthusiasm, that can feel good and be productive. The value of CLEAR isn't teaching you to distrust every AI response — it's teaching you to notice when these patterns happen. Noticing is the skill. Once you can spot the pattern, you can decide whether it matters this time. That's what helps when the stakes are real — a medical decision, a financial commitment, a legal claim. The habit of noticing costs you nothing and tends to pay for itself the first time it stops you sending something you'd have regretted.

The AI Clarity ladder

Where this leads, if you want it to

CLEAR is the foundation. Each tier above goes deeper — naming what AI actually does, then directing it, then engineering systems with it.

✓
CLEAR
You're here
The foundation. Notice the patterns. Ask the questions. Catch the habits. Six lessons, free, designed to be useful from the first one.
↓
2
SHARP
The next step
Read AI sessions properly. Names the seven machine patterns by mechanism. Adds the ten human-side habits. Builds the protocols, the interventions, and a 30-day practice routine. Where the framework you've just met becomes a working method.
↓
3
BUILD
Direct AI to ship
AI types. You direct. You verify. Walks a non-developer through shipping a real, deployed AI tool — using the same patterns you've learned to spot to catch what AI gets wrong as it builds. The human overlay AI can't do for itself.
↓
4
SCALE
For technical leads
Production agent systems. Multi-model orchestration, retrieval, evaluation, observability — the engineering tier for people who already know how to code and want to build the systems that run on Monday morning when real users are hammering them.

No pressure. CLEAR stands on its own. The ladder is here when and if it's useful.

This course taught you to ask the right questions. The four questions work on any AI, on any topic. Everything above adds depth — but the foundation is what you just learned. Use it.
Saved
Copied to clipboard!