# A Persona Bot Reply Is Not the Person's Current Judgment ## Claim A chatbot trained on a real person's public output and framed as letting users "chat with" them needs an explicit boundary marking replies as generated, or users will read them as that person's current position. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds specifically for bots framed around a named, real individual's identity or voice — a generic branded assistant with no claimed personal identity doesn't carry the same risk. - The risk scales with how authoritative or newsworthy the person is; a persona bot for a low-visibility figure carries lower misattribution stakes than one for a public founder or executive. - Doesn't argue persona bots shouldn't exist — it argues the disclosure infrastructure (marking, dating the corpus, citing sources) is a load-bearing part of the product, not an optional add-on. ## Evidence Matt Mullenweg's AutoMattBot announcement is a single short paragraph: a Telegram bot trained on his blog posts and public talks, framed as chatting "as" him. Nothing in the announcement states how a reply is marked as generated rather than a live statement, what corpus cutoff applies, or whether the bot cites its sources. The gap is the point: the product ships before the disclosure norms that would make its outputs trustworthy. ## Links - [[AI & the Future of Knowledge Work MOC]] — the attribution boundary knowledge work needs when a model speaks in a named person's voice - [[3. Resources/Permanent Notes/Working Locally With No-Push Intent Limits AI Contribution Risk]] — because contrasts: both are about drawing a boundary around AI-generated output before it's mistaken for something it isn't, one in code contribution, one in personal voice. - [[3. Resources/MOCs/AI Agent Safety MOC]] — because applies-in: persona-simulation disclosure is a specific, nameable subcategory of agent-safety governance. ## Tests / falsifiers - Survey users of a persona bot after a session: do they correctly identify which statements were generated versus which reflect a verified, current position from the real person? A high misattribution rate would confirm the risk; a low rate (e.g., because the interface already marks replies clearly) would weaken the claim for that implementation. - If a persona bot with no disclosure markers produces measurably no more misattribution than one with clear markers, the claim that explicit boundaries are necessary would be undercut. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Readwise/Articles/Automattbot]] — the announcement's brevity and absence of stated disclosure, corpus, or citation practice.