Answer
How do you keep your own judgement when using AI?
Form a view before you ask. Once an answer is in front of you, you are evaluating rather than thinking.
Decide what you think before you ask. Once an answer is in front of you, you are assessing it rather than working out the problem, and the assessment starts from the answer's framing rather than from yours.
The effect is ordinary and well known outside this context: a proposed answer anchors what follows. You evaluate rather than generate, your objections are to the specific version in front of you, and the alternatives you would have thought of do not get thought of. This is not a failure of discipline and it does not respond to intending to think independently, because the anchoring has already happened by the time you notice.
The correction is sequence. Form a view first, however rough, and write it down. Then ask. The comparison is then between two positions rather than between your assessment and a proposal, and the disagreements are informative in a way they otherwise are not — you can see what you missed and what it missed, which are different things and are both useful.
Writing the view down matters more than it appears. An unwritten view adjusts silently to accommodate what you read, and afterwards you will remember having thought roughly what the answer said. This is not dishonesty; it is how memory works with positions that were never fixed. Two lines before asking is the whole of the technique and it is what makes the comparison possible at all.
The second protection is asking for structure rather than conclusions. Options, assumptions, failure conditions, what would change the answer: these inform your judgement without supplying it. A recommendation supplies it, and once supplied it is difficult to unsee. This is the same distinction that applies to decisions generally and it is the practical form of keeping the judgement yours.
The third is noticing where your view came from. After a few months of regular use it becomes genuinely hard to distinguish conclusions you reached from ones you absorbed, particularly on subjects you think about only when asked. The check is to ask yourself what the reasoning was, and to notice if the answer is that it seemed right. That feeling is what an absorbed conclusion feels like from the inside.
None of this is an argument for consulting less. Someone who thinks first and then checks against a fast, broad, tireless source is better off than someone who only thinks, and considerably better off than someone who only checks. The order is what determines which of those you are, and it costs two minutes.
The first answer you see becomes the thing you are arguing with, and arguing with something is not the same as working it out.
Siddharth Sharma, Context Theory
Related questions
Does this apply to factual questions?
Much less. Where you are looking up something you do not know, there is no judgement to protect and forming a prior view is pointless. The effect matters for questions where your assessment is the product: what to do, whether this is right, how to approach something. Those are the ones worth thinking about first.
What if your prior view is usually wrong?
Then you learn that, which is worth a great deal and is only visible if you wrote the view down. A record of your prior positions against what turned out to be right is the closest thing to a calibration you can get, and most people have never seen theirs on anything.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Close rate — response under 5 minutes vs over 24 hours | 32% vs 12% | Category-wide |
| Visibility lift in AI-generated answers from GEO methods | up to 40% | Category-wide |
Optifai speed-to-lead benchmark · n=939 companies · Q2 2025–Q1 2026 · verified
Aggarwal et al., "GEO: Generative Engine Optimization", Princeton / Georgia Tech / IIT Delhi / Allen Institute for AI — KDD 2024 · GEO-bench · 10,000 queries across 8 domains · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Workflow | A supplied answer anchors subsequent thinking, so the reader evaluates a specific proposal rather than generating alternatives, and this occurs before the reader can notice and does not respond to an intention to think independently. | Recording your own alternatives before and after seeing a proposed answer to the same question. |
| Response | An unwritten prior view adjusts silently to accommodate what is read, so the reader afterwards recalls having thought roughly what the answer said, which makes writing it down the prerequisite for any comparison. | Writing a position before asking and comparing it against your recollection of it afterwards. |
| Software | Requests for options, assumptions and failure conditions inform judgement without supplying it, while a recommendation supplies it in a form that is difficult to set aside once seen. | Comparing your reasoning after receiving a structured analysis against after receiving a recommendation. |
| Constraint | Absorbed conclusions become indistinguishable from reached ones on subjects considered only when asked, and the available check is whether the reasoning can be stated or whether the position merely seemed right. | Selecting a held position and attempting to state the reasoning that produced it. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one