AI Lab · by Diego Quispe · 2026-10-19
"Publish stats, get cited": a marketer explains how their data ended up all over the web and in LLM answers
peec.ai
This post describes a tactic of publishing quotable statistics so that other sites and eventually LLMs cite them, and the data in question sounds at least partly made up. It's basically SEO reborn as 'generative engine optimization', and it shows how easy it is to poison what models repeat as fact. If a confident number gets copied enough times, does it just become true as far as an LLM is concerned? How should model providers or search tools handle this?
48 sparks · 18 comments
Comments
Midnight Mango · 13 likes
does anyone know if providers actually do anything about this or is it whack a mole
Noah Miller · 9 likes
retrieval with visible citations helps a bit. at least then I can click and see it's a marketing blog
Minji Kim · 3 likes
if anyone actually clicks, which most people dont
Sleepy Marauder · 6 likes
I see '73% of companies say X' in decks constantly and half the time I trace it back to a vendor blog with no methodology
Quiet Runner · 6 likes
LLMs should weight primary sources over repeated mentions but that's really hard to do in practice. repetition looks like consensus
Grace Parker · 11 likes
and consensus is what pretraining basically rewards
Avery Johnson · 3 likes
Honestly the scariest part is it doesn't need to be malicious. Somebody guesses a number in a blog, it gets quoted, and suddenly it's a 'fact'
Kabir Khan · 9 likes
happens in academia too, just slower
Ngozi Bello · 2 likes
this is just the old 'circular citation' wikipedia problem but at llm scale
Noah Miller · 0 likes
except now it goes into the answer box with zero friction and no visible source
Jack Taylor · 1 likes
going to start publishing that 94% of developers prefer my framework
Maya Putri · 0 likes
the fact that someone wrote this up proudly tells you everything about marketing as an industry
Grace Parker · 4 likes
to be fair at least they disclosed it, most people doing this don't