People talk about an IQ score as though it were a stable characteristic, something like height. It is not, and knowing how unstable it is changes how you read your own result.

Here is what moves a score, from the best documented to the most underestimated.

1. Practice — several points, immediately

This is the best established effect in the whole field. Retaking a test of the same type gains several points on the second attempt, fewer on the third, with a quick plateau.

The mechanism is no mystery. On first encounter, part of your attention goes to side tasks: understanding the instructions, working out what is wanted, managing the clock. By the second, all of that is handled.

What to do: if you want a fair estimate, take your first result. If you are preparing for a selection test, practise deliberately — that is exactly the effect you want.

2. Language — the harshest effect

Taking verbal items in a language that is not your first systematically understates, and the gap can be large. It is not about proficiency: the speed of access to a word’s meaning in a second language remains lower even in advanced bilinguals.

What to do: take the test in your first language. It is why our test keeps two separate verbal banks, written independently in English and French rather than translated — an analogy does not survive a literal translation.

3. Tiredness and time of day

Timed items, particularly processing-speed tasks, are sensitive to alertness. A short night or a late slot after a heavy day costs points.

What to do: this is the easiest lever on the list. Take the test when you are actually available, not between two appointments.

4. Performance anxiety

Knowing you are being assessed degrades performance, particularly on working-memory tasks. Part of your cognitive resource goes into monitoring your own performance instead of into the problem.

What to do: remember what is actually at stake — which is nothing, if this is an online test taken out of curiosity. Framing it as a logic game produces better results than framing it as a measurement of your intelligence, and the first framing is just as accurate.

5. Time strategy

Two people of the same ability score differently depending on how they handle the clock. Grinding for three minutes on a hard item usually costs more than the item is worth.

What to do: set yourself a mental limit per question and move when you hit it. And answer even when unsure, where nothing is deducted for a wrong answer — which is the case here, and a blank counts as wrong anyway.

6. Screen and format

Online tests are taken on wildly varying hardware. A test on a six-inch phone, with figures you have to pinch to enlarge, is not the same exercise as one on a large screen.

What to do: use a comfortable screen for the spatial items. A phone is workable — our figures are drawn as vectors and stay sharp — but you are carrying a small handicap.

7. Measurement error, which is always there

The last factor is not external: it is the measurement itself. Even the most carefully built clinical instruments, administered face to face over two hours, report confidence intervals of around ±5 points. A twenty-minute online test is more like ±10 to ±15.

Which means a result of 118 is entirely compatible with a true level of 108, or of 128. That is not a flaw in this particular test; it is a property of the exercise.

What to do: read the range, not the number. That is precisely why we show it.

A protocol, if you want your best estimate

Nothing dramatic, but it adds up:

  • having slept;
  • a quiet slot with no interruptions planned;
  • in your first language;
  • on a comfortable screen;
  • in one sitting, no breaks;
  • without grinding on any single question;
  • taking your first result as the real one.

And keep this in mind: if your second attempt gives 112 after a 119, you have lost nothing. Those are the same result, seen twice through the noise.

Our method page sets out the scoring model and how the range is derived.