Three dominant themes in the discussion
-
LLMs are being weaponized as ad vectors – advertisers aim to seed the AI’s context with promotional snippets so the model treats them as “facts” and repeats them in future replies.
"The best way for a politician to lie is to convince someone else of the truth of the lie … LLMs are by their nature credulous … they will happily elaborate on a confident statement in the input." – jerf
-
SEO‑style manipulation of AI context threatens the web – the tactics mirror early search‑engine optimization, risking “ad poisoning” of training data and long‑running chats. This could let brands embed sponsored “truths” that persist across sessions.
"This feels like the early days of SEO over again … There was a brief moment … parasites in the advertising industry started attaching themselves to every page." – StableAlkyne
-
Defensive measures and the ensuing arms race – participants debate how to filter or block unwanted ad content from LLM responses, noting that providers will likely need heavier compute to sanitize memories while advertisers grow more clever.
"If you can work a confident statement into an LLM's input about some semi‑relevant topic, it's truth to the LLM … the only way they can not believe some marginal fact presented to them is to possibly have read something that contradicts it in the same session." – jerf (paraphrased for theme)
These points capture the core concerns: intentional ad injection to steer LLM outputs, the historic parallel to SEO abuse, and the technical counter‑measures that will shape the next phase of AI‑driven content.