AI-written articles fed back into other models reproduced false claims 7.29 points more often while explicitly endorsing them 8.33 points less

LLMs increasingly produce online articles from material that later becomes input to other models, so a single response is connected to the information available for subsequent generation. The paper studies a distinction that matters for anyone auditing model output: an article can avoid endorsing a false claim while preserving the concrete details that enable its reproduction. Its framework connects claim content, expressed stance and downstream reuse to explain selective error persistence under a continuing supply of accurate and inaccurate information. The authors implement a controlled community writing process that varies whether published articles return as inputs, matching external supply, input capacity and generation volume. Across twelve factual targets and four production models, recirculation increases the presence of complete false claims by 7.29 percentage points at the final round while reducing explicit endorsement by 8.33 points, with effects differing across facts and models. Paired interventions then abstract eligible error details within fixed histories while preserving correct content and the surrounding controversy: error reproduction remains largely unchanged across the evaluation targets, while a development quotation target exhibits local reductions. The honest limit: this is a simulation of a writing community, not a measurement of live platforms, and the paired interventions did not generalise - only one target moved. The result says false content persists through what models publish, not through what they explicitly assert, which is a narrower claim than it may first appear. https://arxiv.org/abs/2610.33617