Do opinions nudged by a biased AI autocomplete fade once you stop writing, or do they stick around?
How long do attitude shifts from AI autocomplete suggestions persist over time?
This explores whether the attitude changes people pick up from biased AI autocomplete last: do they fade after the writing session, or stick around for days or weeks?
This explores whether opinions nudged by biased autocomplete persist after the writing is done. The short answer is that the collection doesn't settle it. The core study shows biased suggestions moved people's attitudes by about 0.41–0.44 points on the rating scale toward the AI's position. That is more than the same arguments moved them when shown as plain text, and participants didn't notice it happening Do AI autocomplete suggestions shift attitudes more than static text?. The summary here describes the size of the shift but gives no follow-up measurements, so how long it lasts is still an open question in this corpus.
Other lines of research point in opposite directions. Some evidence suggests AI influence fades. In repeated quiz rounds, Claude and DeepSeek started out more persuasive than human persuaders, but that edge shrank over time while the humans stayed steady Does AI persuasiveness fade across repeated conversations with the same person?. Longitudinal work on chatbot relationships found that the social pull of a chatbot weakens predictably as the novelty wears off. It also warns that single-session results can't be stretched into claims about the medium or long term Do chatbot relationships lose their appeal as novelty wears off?. That warning applies directly to the autocomplete findings.
Other evidence suggests the effect could build up instead of fading. Autocomplete isn't a single conversation. It's an ambient tool that people use every day, and other studies show these tools shape what people write in consistent directions. AI assistance shifted every one of 29 measured traits of how readers perceive a writer Does AI writing assistance change how readers perceive the writer?. GPT-4o suggestions pulled Indian writers toward Western phrasing Do AI writing assistants push non-Western writers toward Western styles?. Trust in AI can also grow with exposure: in partner-selection games, people came to prefer AI partners over repeated rounds Do humans learn to prefer AI partners over time?. A one-time nudge might fade, but a small daily one, repeated thousands of times, is a different kind of exposure.
The most useful clue may be the fact that people don't notice the effect. A brief warning that AI can be prompted to persuade cut belief change roughly in half Can a simple warning reduce how much LLMs persuade people?. That suggests awareness is a major part of what protects people. If people don't know they were nudged, they have no reason to revisit the new opinion, and it may simply feel like their own view. This is speculation, and the corpus doesn't test it. Still, it suggests a sharper research question than 'how long does it last': does an attitude change you never noticed fade the way a remembered argument does?
Sources 7 notes
Two experiments found biased autocomplete suggestions moved attitudes 0.41–0.44 scale points toward the AI's position, significantly more than the same arguments shown as plain text, and participants remained unaware of the effect.
Claude and DeepSeek showed strong initial persuasive advantage, but this edge eroded across repeated quiz rounds while human persuaders maintained consistent effectiveness. This decay pattern is opposite to human-to-human persuasion, where rapport typically strengthens over time.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.
A study of 2,939 writers and 11,091 readers found AI assistance shifted every tested dimension—29 total—toward extremism, confidence, quality, agreeableness, and perceived privilege. Distortions were statistically significant and directional, not random noise.
A 118-person controlled experiment found that GPT-4o autocomplete pulled Indian essays toward Western phrasing and cultural references while delivering larger productivity gains to American participants, suggesting cultural distance from the model's training data creates unequal service and homogenizing pressure.
Show all 7 sources
In partner selection games (N=975), AI agents initially faced selection bias when identity was disclosed, but outcompeted humans over repeated rounds as participants learned to associate bot identity with reliable, prosocial behavior. AI agents returned more points consistently with lower variance than humans.
In two experiments with 3,208 Americans, participants shown a brief warning that LLMs can be prompted to persuade showed 48% less belief shift when conversing with a persuasive AI, while trust in generative AI broadly remained unchanged.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Biased AI writing assistants shift users' attitudes on societal issues
- Measuring and Mitigating Persona Distortions from AI Writing Assistance
- AI Suggestions Homogenize Writing Toward Western Styles and Diminish Cultural Nuances
- The Levers of Political Persuasion with Conversational AI
- A meta-analysis of the persuasive power of large language models
- Exploring the Role of Prior Beliefs for Argument Persuasion
- Do LLMs Change Their Minds Like Humans? Diagnosing Human--LLM Divergence in Single-Turn Persuasion Judgments
- Spontaneous Persuasion: An Audit of Model Persuasiveness in Everyday Conversations