Continued Influence Effect

You can repeat the correction. You believe the correction. An hour later, when someone asks why the thing happened, the first story is still doing the explaining. The retraction updated what you would say if quizzed. It did not fully update the model you reason from.

Researchers call that the continued influence effect. Misinformation is encoded, then discredited, and it still shows up in inferences. People are not simply "forgetting the update." In the classic studies they often remember that the early claim was withdrawn. They reach for it anyway when they need a cause, a motive, or a lesson.

A bare negation leaves a hole. "The closet was not full of flammable material" tells you what to delete and not what to put in its place. If the fire still needs a cause, the deleted cause is the one already linked to the smoke, the timeline, and the questions you were asked. An alternative, "the fire started this other way," gives the model something to use. Negation alone often does not.

This is not the claim that corrections backfire and make the false version stronger. That stronger claim has not held up as a general rule. Corrections usually help. The bug is the residue: less belief, not clean belief. You walk away able to say "that was wrong" and still explain the next decision with the wrong story.

• "I know they retracted it, but it still fits."
• "The update said the number was wrong. I keep planning as if it were right."
• "They cleared him. I just wouldn't leave him alone with the account."
• "The article was corrected. The headline is still how I remember the issue."
• "We pulled that feature request. I keep designing as if it were in scope."
• "The rumor was false. I notice I still trust them less."
• "I read the footnote. My summary doesn't include the footnote."

The rumor at work: someone says a teammate missed the deadline on purpose. The next day you learn they were out sick and the handoff failed upstream. You apologize, and you also notice you are still interpreting their slack messages as evasive. The correction is in memory. The motive is still in the story.

The health headline: a scary claim spreads, then a careful correction arrives. You can say the claim was overstated. When a friend mentions the topic, the scary version is the one you narrate first, with the correction as a trailer. Advice you give still tracks the original alarm.

The product postmortem: an early theory blames a vendor. A later log shows the fault was internal. The slide deck still has a bullet about "vendor risk" because that bullet organized the whole meeting. People remember the log. The decision list still behaves as if the vendor were the cause.

The group chat: a forwarded screenshot says a public figure did something. A reply, three messages later, says the screenshot was cropped and the line was never said. By then the jokes have started. The next day's opinion is about the line, and the crop is the part people "know" and do not use.

Judgments about cause, blame, and risk stay tilted. You might not repeat the false sentence when asked directly, and still hire, avoid, vote, spend, or diagnose under its influence. That gap is why the effect matters more than a trivia error. The cost shows up in the inference, not always in the quiz.

Corrections that arrive as "disregard that" leave audiences with a story-shaped hole. They fill it with the best causal material they have, which is often the material you just told them to disregard. Public arguments then talk past each other: one side cites the retraction, the other side is still explaining the world with the retracted cause, sincerely.

There is a kindness cost too. A cleared person, a fixed bug, or a withdrawn accusation can remain socially "a bit true." You are not trying to be unfair. The old cause is simply still available, and available causes get used.

A coherent story is easier to think with than a gap. If the misinformation supplied the cause, deleting it without a replacement leaves the rest of the event under-explained. Later questions ("why did it spread?", "who was careless?") pull the old cause back because it still connects the pieces.

Johnson and Seifert tested one comforting theory and it did not hold: that the problem is only failing to edit inferences you drew before the correction arrived. People who were corrected immediately, before they had time to build those inferences, still used the discredited information about as much as people corrected later. The influence is not only leftover early conclusions. New conclusions, made after you know better, can still lean on the old claim.

Repetition and fluency make the original line easier to retrieve than the update, especially when the update was a negation. "Not the closet" is a worse handle than "the closet." What you can picture gets into the explanation.

Wilkes and Leatherbarrow showed in 1988 that people used invalidated details to answer questions about an event even after those details had been identified as wrong. Johnson and Seifert's 1994 paper named the continued influence effect and pressed on where it comes from. Using a warehouse-fire story in that line of work, people read that flammable materials had been stored carelessly, then read a correction that the closet had been empty. They still drew inferences from the discredited cause. An immediate correction did not prevent that. Offering another plausible cause, rather than only negating the first one, reduced the leftover influence.

Lewandowsky, Ecker, Seifert, Schwarz, and Cook's 2012 review treats the effect as stubborn across studies: people can understand and remember a correction and still rely on the original misinformation when they reason. The review's practical theme matches the experiments. A correction that supplies an alternative account does more than a correction that only says the first account was false. None of this requires the claim that corrections increase belief. The reliable finding is residual use.

• You can state the correction, and your explanation of the event does not use it.
• You say "I know that's not true, but..." and the "but" carries the decision.
• The headline or the first version is the one you would retell.
• A person or a product stays slightly guilty after the charge was dropped.
• You have no replacement cause, so the withdrawn one still organizes the story.

Do not rely on "I saw the correction" as proof the old claim is out of your reasoning. Ask what is actually playing the role of cause.

• When you correct yourself or someone else, replace the cause. "Not X" needs a "Y instead," even a tentative one.
• Retell the event once, out loud or in a note, using only the corrected version. If you cannot, the old claim is still structural.
• Before a decision that rests on a story you saw updated, write the one sentence you are treating as the cause and check it against the latest source.
• Be slower to act on first versions about people. Blame is sticky precisely because it explains so much.
• If you publish or forward a claim, plan the correction as a full sentence people can remember, not as a footnote under the original drama.

If I remember the correction, hasn't it worked?

Memory of the correction and use of the original claim come apart. The studies found people referring to discredited causes in their explanations while the correction was available to them. The test is the explanation you produce, not whether you can pass a true-or-false item.

So corrections are pointless?

No. Corrections reduce influence. They often do not erase it. A correction with a replacement cause works better than a correction that only deletes. Treat "we issued a fix" as incomplete until the new story is the one people can tell.

Is this why a lie still feels familiar after it is debunked?

Familiarity is a neighbor, not the whole effect. Repetition can make a line feel true on its own. Continued influence is what happens after you have accepted that the line is false and it still guides inferences. You can have either piece without the other, and bad information often arrives with both.

Demote the first story from "the explanation" to "a draft that was withdrawn." Then put a cause in the draft's place before you decide.

The teammate

The headline

The postmortem

Practice this pattern in the Reframing App - capture the trigger, label it (like Continued Influence Effect), check evidence, and write a more balanced thought.