Retracted misinformation continuing to shape beliefs and decisions even after the correction has been accepted.
Imagine someone tells you a story about a house fire and says it was caused by fireworks left in the garage. Later they say, 'Actually, there were no fireworks in the garage.' Even though you heard the correction, when someone asks you why there was so much smoke, your brain still whispers 'fireworks.' The first explanation got stuck like glue in your brain's story, and pulling it out leaves a hole your brain doesn't like.
The Continued Influence Effect describes how information that was initially accepted as true but later retracted or debunked continues to contaminate reasoning and judgment. Even when people explicitly remember and accept a correction, they still draw inferences based on the discredited information, particularly when it served a causal or explanatory role in their understanding of events. The effect is driven by the brain's preference for coherent narratives: removing a piece of causal information leaves an uncomfortable gap in one's mental model, so the mind defaults to using the old, false explanation rather than tolerating incompleteness. This makes the CIE extremely robust and difficult to eliminate, with research showing it persists across fictional and real-world scenarios, across cultures, and even when corrections are repeated multiple times.
The same glitch looks different depending on the terrain. Finance, medicine, a relationship, a team — same mechanism, different costume.
Investors continue to factor in debunked earnings forecasts or retracted analyst reports when making buy/sell decisions, even after corrections are issued. Initial misinformation about a company's financial health creates a persistent anchor in mental models of that company's value.
Patients who initially hear an incorrect diagnosis or a debunked health claim (such as vaccines causing autism) continue to let that misinformation influence their treatment decisions and risk assessments, even after accepting the correction from their physician.
The phenomenon was first empirically demonstrated by A. L. Wilkes and M. Leatherbarrow in 1988, using the warehouse fire paradigm. The term 'continued influence effect' was formalized by Hollyn M. Johnson and Colleen M. Seifert in their 1994 paper in the Journal of Experimental Psychology: Learning, Memory, and Cognition.
In ancestral environments, first-received information about threats or resources was usually accurate and acted as a survival cue. Rapidly encoding and retaining initial causal explanations — such as 'that bush rustled because of a predator' — and being slow to overwrite them provided a safety margin against false negatives. The cost of clinging to a slightly outdated belief was usually lower than the cost of discarding a true warning.
Language models trained on large corpora absorb misinformation alongside corrections. Because debunked claims often appear in more sources and with greater frequency than the corrections, models may assign higher probability to false claims. Additionally, when fine-tuned on outdated data that includes since-retracted information, models can perpetuate and even amplify the continued influence effect by confidently restating corrected falsehoods without the retraction context.
This page taught you the name. The deck turns the name into reflex. 1,100+ swipeable scenarios, 1,100+ defenses, 650+ detection prompts — spaced-repetition Swipe Deck, unlimited Spot-the-Bias Quiz, Defense Playbook, Pre-Flight, My Blindspots, Cheat Sheets, Field Guide e-book. $39.53$59.
