At the counter of a builders’ merchant, a customer tries to remember what he heard about Tess. There was a photograph. There was something about a bill. He thinks somebody interviewed her, though he cannot remember who. He has not kept the image. He is quite sure he has come across the story before.
What should he do with that feeling?
He could turn it into advice. When a neighbor asks for an electrician, he could say that he would be careful. He could make the advice more persuasive by adding that several people had raised concerns. Or he could admit the limits of his account: he remembers encountering a claim and cannot now identify its basis.
These sentences ask different things of the listener. The first invites the listener to inherit his caution. The second offers a picture of wider knowledge. The third leaves a question open. Only the third describes what he can presently support.
We do not need to accuse this customer of lying to insist on the difference. He may be trying to help. Nor must we settle exactly how his memory works before he can act with care. He can decide against turning an uncertain recollection into a confident warning. There is room between having a thought and making another person pay for it.
That room is important because the past does not come to us with every receipt attached. A sentence can feel available while its supporting record is out of reach. Once a person’s livelihood enters the question, we should be willing to fetch the record or acknowledge that we have not.
What did the study ask?
Research can sharpen this problem, provided we resist using the word memory as a sack into which every finding is dropped. Thinking an opinion is widespread differs from thinking it is true. Remembering where a detail came from differs from remembering that it was corrected. The studies below examine particular pieces of that larger problem.
In six experiments, Kimberlee Weaver and colleagues found that repetition by one group member could increase estimates of how widely an opinion was held, even when participants knew it came from one speaker. Their outcome was perceived popularity, not the truth of an accusation. The work, published in 2007, therefore cannot tell us how many readers believe a copied quotation. Applying it to repeated web pages is an analogy. Its precise warning concerns a judgment about a group: a familiar opinion can be taken as more common than the number of independent speakers warrants. S05
Gordon Pennycook, Tyrone Cannon and David Rand asked a different question in experiments reported in 2018. Prior exposure to false headlines presented in a Facebook-style format increased later accuracy ratings, both immediately and after a week. Overall belief could remain low, and entirely implausible claims marked a boundary. The outcome was perceived accuracy, rather than the number of people thought to share an opinion. A warning that a headline was disputed did not eliminate the repetition effect under those conditions. That warning was not a full explanation of why a particular attribution was false. S07
Maria Zaragoza and Sean Lane’s 1994 research concerned eyewitness memory. Across five experiments, some participants identified suggested details as things they remembered witnessing. The error varied across conditions and was not inevitable. Here the question was the remembered source of a detail, rather than a headline’s credibility or an opinion’s popularity. Their results do not authorize us to dismiss a particular witness or to diagnose what happened inside our customer’s mind. They establish that source misattribution can occur in those experimental settings; online quotation disputes were not what the researchers tested. S06
Now consider an answer that has already arrived. In a study of 612 online participants, Briony Swire-Thompson and colleagues tested beliefs and memory immediately after corrections or affirmations and again a month later. Memory failure was associated with beliefs moving back toward their earlier level. Published online in 2022, the study concerns the connection between remembering a correction and sustaining belief change. An association does not show that forgetting caused every return to error. The authors suggested repeated correction as a possible response; the study did not directly establish that intervention’s effectiveness. S08
Christopher Wahlheim and colleagues compared correction formats in two online US experiments, published in 2025, with final samples of 351 and 369. Participants rated headlines immediately and after a week or a month. Labeled corrections pairing false and true details produced the most durable reduction in false-headline belief. Recognizing a correction was associated with better belief accuracy. Unlike merely measuring a later relationship, this design varied the correction’s format. Even so, these were headline-rating tasks, not tests of whether a customer’s trust in an electrician, or a lost job, had been restored. S04
No single score could answer all these questions. To assess a repair fairly, we should first say which question we are asking. A publisher may have improved the wording while failing to send it to the customer. The customer may accept the evidence while an old decision remains in place. These are separate failures of the process we are proposing, and each needs its own response.
Leave something the reader can check
Imagine that the account which posted Tess’s image replaces it with an apology. The apology says that confusion arose and regrets any distress. A reader opening the page a week later finds no description of what changed. The reader cannot tell whether the quotation was wrong, the photograph was wrong, or Tess merely objected to the attention.
The apology leaves the factual task unfinished.
The account owner could instead explain that the attributed quotation had no verified interview behind it, identify the image being withdrawn and direct readers to the correction. Such wording would give the reader a specific reason to revise a judgment. It would also give anyone who had shared the image something usable to send onward. Whether every reader will accept it remains open. At least the publisher would have supplied an answer to the error it introduced.
There is a practical tension here. Tess might prefer never to see the sentence again. A person who received it may need enough detail to identify what the correction concerns. A public post that displays the full allegation in large type could carry it to people who had never heard it. A private reply to the original customer can be more exact without announcing the matter to the whole town.
I would resolve that tension by beginning with the recipient’s task. What must this person recognize to update the decision? Supply that much context, with the correction beside it. Avoid turning the allegation into the most prominent element merely because it attracts attention. Keep the evidence close enough that a reader can inspect it without searching through a long argument about everyone involved.
This is an editorial proposal about clarity and scope. It is not a promise about the reader’s mind. No wording earns the right to claim that all harm has been reversed.
Even the word correction deserves care. If the claim remains unresolved, a publisher should say so rather than replace one unsupported certainty with another. If a quotation was falsely attributed, that should be stated plainly. If the actual error was a wrong date, correct the date. A vague withdrawal can leave too much suspicion; an overbroad declaration can erase matters the evidence never settled.
The standard should work when the person is unpopular. Otherwise it is a courtesy reserved for friends. You need not admire someone’s work to refuse a false quotation attached to their name. You can also correct the quotation and continue to criticize conduct that has a sound basis. Keeping those acts together makes fairness more demanding and more credible.
The Extra Chair
A coat hangs from the chair
we meant to fold away.
Its sleeve touches the floor.
Beside the plates, a phone
turns bright, then dark.
Nobody reaches for the bread.
You move the coat upstairs.
A dark patch on the seat
begins to dry.
We draw the chair in close.
The bowl goes round again.
Your hand receives its weight.
What a person owes their own uncertainty
The customer at the merchant may find the correction. He may also retain an uneasy feeling. We cannot order him to experience complete confidence. We can ask him to take responsibility for what he does next.
A closer test of person evaluation comes from Adam Eggleston, Richard Cook and Harriet Over. Their two preregistered 2022 experiments each involved 128 participants. Generated faces paired with favorable or unfavorable fictional tweets later received different trustworthiness ratings. A disputed label moderated the immediate effect, but not significantly after ten minutes. These were controlled tasks with invented people and largely White British samples. Ten minutes is no evidence of lifelong damage, and rating a face is not hiring a tradesperson. The study tests person evaluation while leaving those further questions open. S12
Suppose he tells the neighbor: I saw a post, but its publisher withdrew the quotation because it could not support the attribution. That is an account the neighbor can use. If he adds that he still feels unsure, he should make clear that this is his feeling, not fresh evidence. He should not translate the feeling into several unnamed reports or a claim that everybody knows.
This rule applies beyond customers. An editor deciding whether to repeat a charge can ask what would justify printing it today. A manager reviewing an old decision can ask whether the reason for that decision still stands. The fact that someone once acted on a claim supplies a reason to review the action. It does not supply independent evidence that the claim was true.
There is an awkward social moment in such a review. Someone has to say that the earlier judgment was wrong or insufficiently supported. The institution may prefer a sentence with no subject: errors were made; concerns have been addressed. Yet the people affected need to know which record changed and who will change the decision that rested on it. A sentence with a subject makes the repair easier to assign.
A person need not produce an elaborate public confession every time they pass on a mistaken detail. The response should fit the act. A private warning may call for a private correction to the same recipient. A public accusation can require a public correction with comparable prominence. An error entered into an employment file calls for attention to that file and its users. Scale matters because the place where the wrong statement lives determines who can still act on it.
Tess should not have to test the sincerity of every reader. That would give her another impossible occupation. She can ask for observable acts: withdraw the unsupported attribution, correct the surviving page, tell the people to whom you sent it, reconsider the decision. These requests give a cooperative person something definite to do. They also reveal where cooperation ends.
Someone might answer that the internet never forgets and none of this can work. It sounds worldly. In practice, it asks the person harmed to accept defeat before the available work has been tried. It confuses the survival of some copies with the uselessness of correcting any copy. A customer who changes a decision matters even if a distant screenshot remains.
Nor should fear of backfire become an automatic reason for silence. Thomas Wood and Ethan Porter tested 52 issues across five experiments with more than 10,100 participants and found no correction-induced backfire in those conditions. Their paper appeared online in 2018. It challenges the claim that correction inevitably strengthens error; it does not prove that backfire is impossible or that every correction will restore a reputation. S11
The opposite reassurance is no better. Telling Tess that truth always wins promises an outcome nobody can guarantee and leaves her to explain herself if it fails. We can give her something more solid than either prediction: a clear account of what is known, a route to the relevant people, and help with the work that route requires.
A correction can become ordinary knowledge
Consider the neighbor a month later. Someone offers the old image as a reason to choose another electrician. The neighbor has kept the correction. There is no long speech. The neighbor says that the quotation was withdrawn and forwards the record showing why.
Nothing spectacular has happened. A claim met a person who knew where to find its answer. Tess did not have to enter the conversation herself.
That is a reasonable ambition for repair. We should want accurate information to be available where a false claim might otherwise guide an action. We should want the answer to travel without requiring the subject to perform innocence each time. A correction page can support that work; so can a plainly written note in a decision file and a person willing to use it.
The customer at the counter eventually asks the merchant whether there is a source for the interview. The merchant does not know. The customer takes out his phone and checks the message he received. This time he reads past the photograph. He finds the reply beneath it, including the withdrawal he had missed.
He still has a kitchen job to arrange. He can ask about Tess’s availability, her estimate and the work itself. The old sentence does not have to chair the conversation.