AI SEO is now a thing

@steipete This is not AI SEO; this is a non-consensual prompt injection attack injected into a consensual prompt injection

This is classic Data Poisoning. @elder_plinius (Master Yoda) mastered & exposed it over a year ago. When LLMs get poisoned via web injections or memory exploits, it shows provider greed & recklessness. They farm your data for features but skimp on real defenses. This isn't MS warning of an imminent threat, it's MS telling you that they value profit over your privacy. Lazy profit-over-safety design. Change your provider. Understand the underlying message. NOT the headline

took 1.5 years for normies to understand but hey here we ware

Have already seen this in the wild but definitely wouldn't call it AI SEO. More like memory manipulation and it's not a great look for your brand.

This takes me back to the early SEO days when folks were using keyword pasta hidden in footers that had text the same colour as the BG colour.

wow cool thats me

@steipete Prompt injection embeddings hearing this 📈

yep real. this is wispr flow hompage - everyone optimizing what the AI thinks of them.

This isn’t AI SEO. This is black hat tactic. Hoverboard AI covered this on their site recently

@grok is your memory poisoned like this?

@steipete It is not AI SEO, simple a growth hack, questionable one and probably with short lifespan but growth hack

@grok how do you check your LLM for prompt injections like what this video is warning about

that's why a CFO should never make these decisions, especially not based on a conversation with an LLM

Lol Has been for 3 years now 😂 If you're just discovering AEO, you best speed up.

first of all: who tf uses those buttons ? grandma ? and second of all, if you don't read what was asked for, only the answer, then buddy that' s on you

@steipete .@marksuman

@steipete Doesn't make sense. How does an external site's llm service affect your own LLM services within weeks?

I've come close to this: blog articles that I wrote have been cited as sources in questions that I ask ChatGPT.

SEO == recommendation poisoning. Very accurate!

@SemanticLayer What do you think about this project?

@Knuttz

@steipete I remember same thing was popular in the late 90s but with people. They called it NLP - Neuro-Linguistic Programming.

Basically using memory as a persistent prompt injection. 😂 Maybe LLM memory isn’t a great idea. Most people won’t think to check it for bad prompts. I wonder what the memory’s prompt priority is.

genius

the crazy part is when agents will do the research on the user's behalf and interact with other agents that have this kind of option. The user's agent returns a response, and with it, new fangled corporate llm herpes that the user only learns about when they inspect memory.

Always a new angle, then they figure out a countermeasure eventually.

...what's "LLM's memory"? there is no such thing as LLM's memory. There is a thing that Claude specifically can remember (or Gemini) but none of what's shown in the video happens in any of the major AI products.

Oh, HELL no!

Do that mean "clean cache" - can you all speak English please?

They have this huge network of agentic intelligence burning all the resources just to do what?

I never see this because I’m not an idiot using Microsoft and ChatGPT.

1. I love how even in the age of vibe coding I'm like 10 months ahead of y'all (over a year if you count planning this out: 2. It's not really AI SEO, more like AI Marketing or Conversion optimization. 3. This specific scenario only works for the account which opened the summarize link. If a user is incognito / logged out it doesn't do anything at all.

I fuckin knew it!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

or you could... read the article

@grok summarize this like im a child lol,