Every communications team has a version of the same fear. Someone publishes a false claim about the brand. The room debates whether to answer it. Responding means repeating the accusation to deny it, and repetition breeds familiarity, and familiarity breeds belief. So the disciplined move — the one that sounds like restraint — is to say nothing and let it die.
That fear has a name in the literature. It is called the familiarity backfire effect, and it has just been tested more carefully than it has ever been tested in a brand context. It did not survive.
Which is the good news. The bad news is that the thing communications teams are not afraid of is the thing that actually costs them.
A 2026 paper in the International Journal of Research in Marketing ran five studies designed to separate two forces that had always been tangled together: repetition, which makes a false claim feel more true, and correction, which makes it less believed. Every previous debate about whether to respond assumed those two forces were in tension and that repetition might win.
It doesn't. Across all five studies, the researchers found no familiarity backfire effect. Repeated exposure did make brand misinformation more credible, exactly as expected. But in no study did that repetition raise belief more than the correction lowered it. Even weak corrections neutralized the repetition. Clear and unambiguous ones reversed it outright.
This is consistent with what the broader field has been finding for several years. Attempts to reproduce the backfire effect in standalone corrections have largely failed to replicate it. The correction almost always nets out ahead.
So the strategic instinct to stay silent is wrong on its own terms. Answering works. Answering plainly works better than answering carefully. The cost of the correction is not that it spreads the lie.
The cost is that it doesn't finish the job.
In 2016, Emily Thorson ran a set of experiments with an unusual design. Rather than measuring only whether people believed a false claim after it was corrected, she measured what they thought of the target afterward. Belief was corrected — cleanly, immediately, to the standard that fact-checking aspires to. Attitude was not. Participants who saw a false negative claim and then saw it convincingly debunked still rated the target worse than people who never saw the claim at all.
She called the residue belief echoes. The belief is gone. The echo persists.
A 2026 replication and extension in Political Communication found the same structure holding even under what the authors describe as the gold standard of correction: immediate, unambiguous, and successful at the level of stated belief. Attitudes did not return to baseline.
Thorson identified two routes by which the residue forms, and the distinction matters for anyone writing a response.
The first is affective. The negative information lands as feeling before it lands as fact, and the correction operates on the fact. Emotion encoded fast is not undone by information delivered slow. The person now believes the claim was false and continues to feel what the claim made them feel.
The second is cognitive, and it is more uncomfortable. A person accepts that the specific accusation was untrue, and then reasons — not unreasonably — that a company against which such a thing was said is more likely than average to be a company about which other such things are true. The correction settles the instance. It does not settle the inference.
That second mechanism is the one that should keep a communications lead awake. It means a successful correction can leave behind a rational, articulable, entirely defensible negative belief that you never addressed because it was never stated.
There is a third layer, and it is the oldest finding of the three.
Most brand responses to a hostile claim lean on the source rather than the substance. The report came from a short seller with a position. The review was posted by a competitor. The account has four hundred followers and was created last month. This is efficient — it is faster to discredit a speaker than to litigate a claim — and in the short run it works.
In the long run it evaporates. The sleeper effect describes what happens when a message arrives attached to a discounting cue: audiences are less persuaded immediately, and more persuaded later, as the message detaches from the reason they had for dismissing it. Memory keeps the claim and drops the attribution. Kumkale and Albarracín's meta-analysis of the whole literature found the delayed rise in persuasion is strongest under two conditions: when the message and the cue both landed hard initially, and when the discounting cue arrived after the message rather than before it.
Read that second condition again, because it describes crisis response precisely. The claim goes out on Tuesday. The statement explaining who published it and why goes out on Thursday. That sequence — message first, discounting cue second — is the exact configuration under which the meta-analysis finds the sleeper effect at its strongest.
A defense built on the source is not a defense. It is a delay with a maturity date.
Stack the three findings and a measurement problem appears.
Standard crisis measurement takes a pulse at seventy-two hours and, if the organization is diligent, again at two weeks. At seventy-two hours, belief in the false claim has dropped — the correction did its work — and the discounting cue is still fresh enough to be doing its work too. Both effects are at maximum. The number looks like recovery.
The belief echo is not visible in that number because the instrument asked the wrong question. It asked whether people believed the claim. The residue lives in favorability, in consideration, in whatever measure captures how the brand feels rather than what is known about it. And the sleeper effect has not begun to show, because dissociation of source from message takes weeks, not days.
Two months later, the tracker moves and nobody can attribute it, because the event that caused it was marked closed in the second week. This is the same failure of instrumentation covered in the case against dashboard attribution: the measurement window was chosen for operational convenience and then treated as though it were chosen for validity.
There is a related trap in asking audiences directly. Post-crisis research that asks people whether the incident affected their view of the brand collects a reconstruction, not a cause — the same problem examined in the work on choice blindness. Affective residue is precisely the kind of influence people cannot report on themselves. Asking them to is a way of confirming that the correction worked while the damage continues underneath.
Four things follow.
Correct fast and correct plainly. The reason to stop hedging is not courage. It is that the strength of the correction is the variable that determines the outcome, and the feared cost of stating the claim clearly has now failed to appear across five studies. Ambiguity buys nothing. A correction written to avoid restating the accusation is a weak correction, and weak corrections only neutralize. Clear ones reverse.
Replace the account, do not negate it. The continued influence literature is consistent on this point: a correction that removes a causal explanation and puts nothing in its place leaves a gap, and the retracted claim keeps filling it because nothing else can. Saying what did happen outperforms saying what did not. A denial is a hole. An account is a substitute.
Build the defense on the claim, not the speaker. Naming the source's motive is legitimate and should be part of the record. It cannot be the load-bearing element, because it is the element with the shortest half-life. Whatever survives twelve weeks has to be a fact about the thing that was alleged, not a fact about who alleged it.
Separate belief from affect in the instrument, and extend the window. Measure the claim and the feeling as two different quantities, at ninety days rather than fourteen. Expect belief to recover and affect not to. The gap between them is the actual damage, and it is invisible to any instrument that treats them as one number.
The last point is the one that changes how the work is resourced. If the residue is affective, statements do not clear it. Statements operate on belief, and belief was never the problem. What moves affect is accumulated behavior observed over time — the same asymmetry that makes an apology weaker than a demonstration of control, examined in the case against contrition.
This reframes what a hostile claim costs. Most organizations price a reputational event as a communications expense: a war room, a statement, some paid amplification, a fortnight of senior attention. That is the price of correcting belief, and it is roughly right for that job.
The residue is a different line item entirely. It is a small permanent drag on affect that no statement will lift, distributed across everyone who saw the claim, decaying on its own schedule and not on yours. It does not appear in the crisis budget because it does not appear in the crisis measurement.
Which means the honest version of a post-incident report contains a sentence most of them do not: belief has been restored, favorability has not, and the difference is a cost we are carrying rather than a problem we have closed.
Correction is not repair. Correction ends the argument. Repair is what happens afterward, in behavior, over a period long enough that most organizations have already stopped watching.