AI‑powered overview” is a mathematical contradiction: It’s not a summary, it’s an amnesiac chatbot

Gemini API
1

Google’s “AI‑powered overview” (AI Overview) marketing is a technical deception.
It is not an overview.
It is not a snapshot.
It is simply a plain, stateless LLM.

To prove this, I attached screenshots taken seconds apart, using the exact same input.
The system generates completely different answers every single time.

Technical reality

Because of token limits and sampling mechanics, the system allows only one run, then restarts.
On every run, it:

  • generates a new token stream

  • uses new sampling

  • runs on a new model instance

Therefore it produces random, stochastic outputs for the exact same input.
It is simply an amnesiac Gemini chatbot disguised as a search feature.

The mathematical contradiction

Google calls this an “Overview” or “Snapshot”.
These concepts have strict logical definitions that are mathematically incompatible with how generative AI works.

What is a real overview / snapshot?

  • stable

  • reproducible

  • deterministic

  • unchanged under identical conditions

What is Google’s AI Overview instead?

  • constantly changes its output

  • unstable

  • not reproducible

  • stochastic (random)

  • a stateless generative process

Summary

You cannot call a stochastic text‑generation process a “snapshot” or “summary”.
A summary means deterministic extraction of facts.
AI Overview is by design incapable of this.

Selling an unstable, amnesiac chatbot as the web’s official “Overview” is:

  • false advertising

  • deceptive UI

  • and a fundamental lie to users

A search engine must be deterministic (the same query must return the same fact).
If it isn’t, then it is not a search engine and not a summary, but a random number generator.

This is incompatible with its claimed role as a web summarizer.

The Screenshot Evidence

The fact that the system gives different answers seconds apart to the same prompt is the clearest physical proof of “hallucination”.
They cannot claim “the web changed”, because the web does not update in seconds.

A token‑limited LLM is technically incapable of producing consistent web summaries.
It is not a serious information tool — it is a robot with short‑term memory loss.

Consumer protection laws penalize products (like AI Overview) that are marketed with properties (objective, summarizing, snapshot) that the product is by design incapable of delivering.

FireShot Capture 428 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com
FireShot Capture 428 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com1920×3736 774 KB

FireShot Capture 429 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com
FireShot Capture 429 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com1920×3468 638 KB

FireShot Capture 430 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com
FireShot Capture 430 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com1920×3460 671 KB

FireShot Capture 431 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com
FireShot Capture 431 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com1920×3756 765 KB

FireShot Capture 432 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com
FireShot Capture 432 - summary Google Isn’t Silent Because They Don’t See My Case — They’r_ - www.google.com1920×3570 687 KB

2

https://archive.org/details/ai-powered-overview-is-a-mathematical-contradiction-its-not-a-summary-its-an-amn

3

I asked Google’s AI‑mode a simple question: “What qualities are needed to make a good summary?” And the model immediately started acting like some conscious expert system, talking about analytical thinking, the ability to see the essence, objectivity, critical reasoning, deep text comprehension, precise writing skills, and logical structuring. All of this is hilarious, because these are human cognitive abilities, and a stateless, memory‑less LLM has none of them. The chatbot behaved as if it actually understood what summarization is, while in reality it’s just a random token generator pretending to be a search engine.

Then I pasted my post into the chat — the one that tears LLMs apart for being amnesiac, unstable, non‑reproducible, and mathematically incapable of producing a real summary. And the moment it saw that, the chatbot instantly backed off. All the big talk vanished. No more “analytical thinking,” no more “objectivity,” no more “critical reasoning.” Suddenly the mighty AI‑mode shrank down to a tiny disclaimer: “AI responses may contain errors.”

That line is Google’s own UI admitting that the entire “AI Overview” is basically a slot machine. You get a different answer every time, and then you have to pick which one you like, just like gambling. A stateless, unconscious, memory‑less LLM cannot “make mistakes,” because it doesn’t know anything in the first place — and that’s exactly why it cannot summarize the web. Google’s AI Overview is not a summary. It’s a random token lottery dressed up as one.

FireShot Capture 433 - összefoglalót ki tud készítsweni milyen tulajdonságok kellenek hozz_ - www.google.com
FireShot Capture 433 - összefoglalót ki tud készítsweni milyen tulajdonságok kellenek hozz_ - www.google.com1920×7438 1.18 MB

4

https://archive.org/details/ai-powered-overview-is-a-mathematical-contradiction-its-not-a-summary-its-an-amn_202607

https://archive.ph/wDKE5

https://archive.ph/wDKE5/image