You Know What You Chose. Probably.
Ask anyone what they are sure of, and the list starts with the self: your own choices feel like bedrock. You pointed at the left photograph, picked the strawberry jam, voted for the candidate, and nobody can tell you otherwise, because you were there.
Modern life is built on that confidence, since economists infer preferences from purchases, pollsters treat stated opinions as data, and courts ask witnesses what they intended, all of it assuming a tight loop between deciding and knowing that you decided; in 2005, four researchers tested that loop with a conjuring trick and found it far looser than anyone expected.
The Card Trick
Petter Johansson, Lars Hall, and Sverker Sikström of Lund University, with Andreas Olsson of NYU, recruited 120 participants to pick the more attractive face from 15 pairs of photographs, then asked them to explain some choices. That was when the trick happened.
Using a magician's double-card ploy, the experimenter flipped the chosen photograph face down and slid a hidden card across the table, presenting each participant with the face they had just rejected, which they then picked up and justified in detail.
The design was careful: deliberation time varied from two seconds to unlimited, face similarity from near-twins to highly dissimilar pairs, and each participant faced three rigged trials out of fifteen, 354 in all.
Thirteen Percent
Of those 354 swaps, participants caught 46 in the moment: thirteen percent. Even in the easiest condition, unlimited time plus the most dissimilar faces, detection topped out at 27%. More time helped a little; dissimilar faces did not.
Counting every form of detection, no more than 26% of swaps were ever exposed, and even that is generous, since catching one swap makes participants vigilant, forcing the authors to discard every later trial. No differences by sex or age emerged.
Take the 26% at face value and three out of four manipulated choices still sail through unnoticed; use the stricter in-the-moment figure and it is nearly nine out of ten.
The Confabulation Machine
Missing the swap is only half the story; the other half is what participants did next, when asked why they chose the photograph they produced fluent, specific, confident reasons for a choice they never made: one participant, handed the blonde after choosing the brunette, praised her dark hair, while another praised the smile of a woman who was not smiling.
Three independent raters, blind to the rigging, scored every explanation for engagement, specificity, and confidence; the fabricated reports were indistinguishable from the genuine ones. The only tell, a slight tendency on rigged trials to pause and interrogate one's own motives, appeared in just 5% of reports.
The authors named it choice blindness.
It Was Not a One-Off
A single clever study with 120 Swedish students could be a curiosity; what happened next made it a phenomenon. The Lund group took the trick to a supermarket, rigging jam and tea tastings for passing shoppers, where 14% caught the swap in the moment and about a third ever did.
Then they went after things people care about: in a 2012 moral-attitudes survey, 69% missed at least one reversal of their own stated moral positions, then argued coherently for the opposite of what they had just endorsed. A 2013 study ran the trick on voter intentions in the Swedish general election; later teams extended it to facial identity sorting and to people's own written arguments, judged more harshly when misattributed.
Eleven years, four paradigms, faces and jam and morality: the blind majority holds everywhere, seven to nine out of ten manipulations passing unnoticed.
Nobody, as far as I can tell, has put these numbers side by side. Here they are:
| Study | What got swapped | Spotted in the moment | Spotted at all |
|---|---|---|---|
| Faces, lab, 2005 (n=120) | Chosen photograph | 13% | 26%* |
| Jam and tea, supermarket, 2010 | Tasted sample | 14% | 33% |
| Moral survey, 2012 | Stated moral position | â | 69% of participants missed at least one of two swaps |
| Face sorting, 2016 (n=172) | Identity match | 21% | 21% retrospective |
*26% before correcting for post-first-detection vigilance; the corrected rate is lower.
Eleven years, four paradigms, laboratories and supermarkets, faces and jam and morality: the blind majority holds everywhere. Roughly seven to nine out of ten manipulations pass unnoticed, no matter what is being chosen.
The Strongest Counterargument
The sharpest objection does not dispute the numbers but what they mean: philosopher Dominic Lopes argued that participants may simply update their preference on the spot, making their reasons sincere reports about a new preference.
A second objection targets the stakes: judging strangers' faces is the thinnest of preferences, and perhaps we monitor choices we care about more closely. The Lund data offer a partial reply, since dissimilar faces did nothing and the moral-attitudes replication used dearly held convictions with the same blind majority. Still, nobody has run the trick on a decision with real money.
A third objection is politeness: perhaps some noticed the swap and played along, but the retrospective interviews found the opposite, with those who caught a swap blaming a technical error. Even so, the experiment cannot fully separate "I did not notice" from "I noticed and constructed a new preference in the moment."
What We Didn't Prove
We did not prove that introspection is always fiction: the non-manipulated reports might be perfectly truthful, and the authors present the radical reading, that confabulation is the norm, as a provocation rather than a result.
We did not test important choices. Faces, jam, and survey bubbles are thin by design, since the trick requires swappable options. Whether choice blindness survives a mortgage or a medical decision is unknown.
We did not learn the mechanism: the paper explicitly warrants no conclusions about what causes the blindness, leaving inattention, weak memory traces, social compliance, and genuine preference updating all on the table; nor did we sample the world, since the original 120 were Swedish students in 2005 and most replications come from the same Lund lineage, meaning a phenomenon this strange still needs a broader cast.
The Bottom Line
You are not the reliable narrator of your own decisions. When researchers secretly reversed people's choices, the overwhelming majority never noticed, then explained in fluent detail decisions they never made. The effect spans faces, flavors, and moral convictions. Your reasons may be real, but check whether they were written before or after the choice.
What You Can Do
Write down your pick before you explain it. For any decision that matters, record the choice and one sentence of reasoning first. The paper trail is your defense against retroactive editing.
Distrust fluency. A smooth explanation feels like evidence of a true memory, but choice blindness shows fluency is cheap. When your reasons come easily, ask what you would have said about the opposite choice.
Verify restatements. When someone mirrors your position back to you, check it against your written note, not your memory. The 2012 survey showed people will argue for the mirror image of their own views without noticing.
Run a decision journal: for repeated decisions, log the choice, the reason, and the date, and review quarterly, since if your reasons drift toward whatever worked out, it will show there first.