The one idea
You cannot feel a bias from the inside. It does not arrive labelled. It feels exactly like thinking clearly — which is why "I'll be careful" is not a countermeasure and never has been.
The only things that work are external: a record made before you started, a question you are obliged to answer, a person whose job is to disagree.
Write down what you expect to find, before you look. Then check whether you found exactly that.
That single habit catches more errors than knowing the name of every bias in the list below.
You cannot notice your own blind spots by trying harder. You need a written record of what you thought before you started.
Log your prediction before research. Assign someone to argue the other side. Ask "what would change my mind" and write the answer down.
Bias mitigation that relies on introspection fails, because the bias is not available to introspection. Effective countermeasures are procedural: they change what you are required to produce, not how hard you concentrate.
The five, and what they look like in you
Skip the definitions. Here is what each one actually feels like on a Tuesday.
Confirmation bias
What it feels like: research going well. Sources agreeing. The picture coming together quickly.
What it looks like in your behaviour: you open eight tabs and close the two that disagree without reading them, because they "seemed low quality". You search "why is X good" rather than "problems with X". You stop when you have enough to support your view — and that point arrives earlier when the view is one you already held.
The countermeasure: before searching, write I expect to find that ___. Then run one search phrased against yourself — "X problems", "regret buying X", "migrated away from X" — and give it the same time as the first. If you found no credible criticism of anything, you have not searched, you have shopped.
Anchoring
What it feels like: having a sensible starting point.
What it looks like: the first number said in a meeting shapes every number after it. A vendor quotes ₹8 lakh and the negotiation now happens around ₹8 lakh, whatever the thing is worth. Someone estimates three months and every later estimate clusters near three months — including yours, including after the scope doubled.
The countermeasure: produce your own number before hearing theirs. Know your salary figure before the range is mentioned. In estimation, have everyone write privately before anyone speaks: spoken-first estimates converge on the first speaker, not the truth.
Authority bias
What it feels like: deferring to experience. Which is sometimes correct.
What it looks like: the senior person says it and you stop evaluating, including on topics where you know more. You cite a name instead of an argument. You treat a confident, well-produced source as a correct one — which is also how a polished vendor page or a fluent AI answer gets believed.
The countermeasure: separate the claim from the claimant. Ask for the reasoning, not the source. When you notice you would have challenged the same sentence from a junior colleague, that is the signal. Expertise is real and worth weighting; it is domain-specific and does not transfer sideways.
Sunk cost
What it feels like: commitment, follow-through, not being a quitter.
What it looks like: "we have already spent four months on this" as an argument for continuing. Finishing a course you no longer need. Staying in a role because of the two years you already gave it.
The countermeasure: If I were arriving today with no history here, would I choose to start this? Money and time already spent are gone in every branch of the decision. They cannot favour one branch, though they will feel like they do.
Availability
What it feels like: knowing what is important.
What it looks like: the risk you worry about is the recent or vivid one. One furious customer email reshapes the roadmap while a hundred quiet cancellations do not. After an incident, everyone over-invests in preventing that exact incident. You judge frequency by how easily an example comes to mind — and examples come to mind because they were recent, dramatic, or yours.
The countermeasure: count instead of recall. Before acting on "we keep getting complaints about X", count how many. The gap between remembered and real frequency is routinely large in both directions.
The mental model
| Bias | The tell | The countermeasure |
|---|---|---|
| Confirmation | Research going smoothly | Written prediction, then a search against yourself |
| Anchoring | A number arrived before your thinking did | Your estimate in writing before you hear theirs |
| Authority | You stopped evaluating when a name appeared | Ask for the reasoning, not the source |
| Sunk cost | "We've already invested…" | Would I start this today from scratch? |
| Availability | A vivid recent example | Count it |
You are asked whether the team should move to a new project tool. You already want to.
Before opening a browser you write: I expect to find it is faster for task tracking and that teams like ours report good results.
Then you search "why we moved away from" plus the tool name and read three results from people who left. You discover a weak reporting export — exactly what your manager will ask about in month three.
Your recommendation does not change. But you now give it with the weakness named, and when someone raises it later you are the person who already knew.
Two job offers. One company impressed you at the interview; the office was nice, the interviewer was warm.
Write down first: I expect to find Company A is a better place to work. Then search both the same way — reviews, exit posts, employee comments — time-boxed equally.
The discipline is symmetry of effort. Most biased research is not falsified evidence; it is twenty minutes on one option and four on the other.
Try this
You have decided your team's productivity problem is caused by too many meetings. Your manager asks you to investigate.
Write down three things that, if you found them, would tell you that you are wrong.
Your challenge
Level 3 · IndependentTake a decision you are currently facing where you already lean one way.
- Write your prediction in one sentence: what you expect the evidence to show.
- Write what you would have to find to change your mind. Be specific — a named finding, not "if there were problems".
- Spend twenty minutes researching only the side you do not favour. Time it.
- Write what you found, and whether it matched step 1 exactly.
Success criterion: if step 4 says "I found exactly what I predicted", you have to be able to point at a specific piece of evidence that could have gone the other way and didn't. If you cannot, you did not research — you gathered support.
What people usually get wrong
- Learning the names and stopping there. Being able to define anchoring does not reduce anchoring. The written countermeasure does.
- Believing that knowing about a bias protects you. It does not. People who study these errors professionally still make them, and say so.
- Using the vocabulary as a weapon. Telling a colleague "that's just sunk cost fallacy" ends the conversation without improving the decision.
- Auditing yourself only when the answer is inconvenient. The decisions most in need of a check are the ones that felt easy.
- Over-correcting into paralysis. Refusing to conclude anything is not rigour. Decide, but write down what would change your mind.
How someone experienced does it
Experienced people install the countermeasure into the process rather than relying on themselves to remember it. A written prediction at the top of the research document. A required "what would change my mind" line in every recommendation. A named person asked to argue the opposite case before a significant decision, so that disagreeing is their assigned job rather than a social risk.
That last is the highest-value habit here. Groups are worse at this than individuals, not better, because agreement feels like validation and the first opinion voiced anchors the room. A team where nobody has ever disagreed in a meeting is not aligned; it is uninformative.
The other senior move is keeping dated predictions and going back to them. Not to feel bad about being wrong, but because it is the only way to learn how wrong you typically are and in which direction. Most people find a consistent tilt — chronically optimistic on timelines, chronically pessimistic on other people's competence — and knowing your own tilt beats knowing all five biases.
When not to use this
Do not run a full bias audit on reversible, low-stakes decisions. Which font, which lunch, which of two acceptable phrasings — spending twenty minutes de-biasing a decision that costs nothing to reverse is itself a mistake.
Reserve the procedure for decisions that are expensive to undo, that you will have to defend, or that you notice you feel unusually certain about. Unusual certainty on a complicated question is a reason to slow down, not speed up.
Prove it
Start a prediction log. One line per entry: the date, the question, what you expect, and your confidence in plain words (fairly sure / could go either way / guessing).
Add an entry each time you start any piece of research or make any call whose outcome you will eventually see. Review it monthly. Within three months you will know something about your own thinking that no lesson can tell you.
Keep learning this
Paste this into any AI assistant. It turns the assistant into a tutor that tests you instead of just answering you.
Act as an experienced practitioner who is good at teaching. I have just learned designing a bias check for a specific decision I am making. Assume I am intelligent but relatively new to this — treat me as intermediate level. Work through this in order, and wait for my reply at each step: 1. Ask me 5 questions that test whether I actually understood designing a bias check for a specific decision I am making. Do not reveal the answers yet. 2. After I answer, tell me which parts I got right, which I got wrong, and which I only half-understand. Explain only what I misunderstood — do not re-teach what I already know. 3. Give me one practical challenge based on something I could genuinely encounter at work or in daily life. Do not solve it for me. 4. Evaluate my solution the way an experienced person would judge it, including what a professional would have done differently. 5. Tell me what to learn next, and why that comes next. 6. Give me trustworthy sources for deeper study — prefer official documentation, primary research or standards bodies over blogs and videos. Rules for you: no buzzwords. No motivational filler. Say "I'm not certain" when you are not certain, and tell me which parts of your answer I should verify myself. Clearly separate facts from your recommendations and your opinions.
Become independent at this
Use this when you want a path from where you are to actually good, with checkpoints you can test yourself against.
I want to become independently capable at checking your own reasoning — not permanently dependent on AI, tutorials or step-by-step guides. Design a progression for me with five stages: Beginner, Guided practice, Independent practice, Real-world application, Professional level. For each stage tell me: - what I must know - what I must be able to do without help - the mistakes people make at this stage - one practical challenge - one real project that would prove I reached this stage - one way I can test myself honestly Then tell me the signals that I am ready to move to the next stage, and the signals that I have skipped ahead too early. Keep the theory to the minimum I actually need. Focus on ability I can transfer to situations you and I have not discussed.