Elie Berreby AI search February 25, 2025

AI Answers Over Search: Is Google Digging Its Own Grave?

Google Search decided to become an AI answer engine.

Google destroyed its own strategic moat against competitors.

Years ago, Google used to be a search engine satisfying most users.

And then Google decided to self-sabotage…

Google made itself more vulnerable to competition by killing its own competitive advantage: being a search engine.

I tested Grok 3 extensively thanks to early access.

It is a powerful multi-modal AI model with Deep Search and decent reasoning abilities.

Nothing revolutionary contrary to what some X users claim but is it really good? Yes! 🙂

Grok has access to distinct datasets and among those:

One dataset is made up of the crawled and indexed web pages (yes web scraping + RAG)
One dataset is made of the mind-blowing X index, which is the social equivalent of Google’s index and currently grows at a rate of about 350,000 posts per minute according to the latest public X data available as of 2025.

Assuming this figure of 350,000 X posts per minute is accurate over the long run, that would translate into 504 million X posts per day…

And keep in mind that while X posts are mostly short texts written by users of the platforms, they also consist of long-form videos (with subtitles), images, memes, and a not-so-secret weapon that almost nobody seems to pay attention to: X Spaces!

The quality of some X Spaces is absolutely incredible. The people who join and speak live, unfiltered, are fantastic. But everything is recorded and simple voice-to-text models can transform those lively spaces into yet another UGC dataset: a unique training corpus made up of user-generated discourse.

If you understand the value of data, you know why having this tied to usernames is extremely precious…

“Grok isn’t the new Google Search but at this rate, the new Google will become Grok.“

— Elie Berreby

Alphabet achieved something incredible with Google Search: making their search engine obsolete in the eyes of millions of users who now rely on ChatGPT, Perplexity, Claude and of course Grok to “search”.

What about Google’s Gemini? It is a good language model but it isn’t great and the monthly cost isn’t justified.

Despite an unbelievable dataset (mostly Google’s index), Gemini has all the typical structural flaws I uncovered in every other Large Language Model I’ve tested in the past 3-4 years. That’s because the quality of a language model isn’t just in the dataset, but in the way it is coded and how “smart” the algorithms are…

Google’s AI overviews rely on a Gemini version tailored for search.

Regular people don’t know what they are called and they don’t care. For users, AI overviews are part of Google Search (and they actually correct: AIOs are part of Google Search). Users hate them partly because they’ve been profoundly misleading and because individuals prefer having the freedom to choose rather than being compelled. Shocking, I know 😀

Dissatisfied users are increasing and Google/Alphabet should remove their blindfolds…

Google Search is transitioning into an AI answer engine and is killing its competitive moat.

By transitioning so violently to AI, Google Search faces an existential crisis of its own making.

Google Search, Googlebot and Indexing issues

The Google Search vs ChatGPT screenshots I’ll show you now were taken at the same time, 3+ hours after this exact URL was confirmed as indexed in Google Search Console (my European timezone).