Why “Rankings” Don’t Apply to AI Search

SEO consultants should NOT use the keyword “RANKINGS” applied to AI answer engines such as ChatGPT, Gemini, Claude or Perplexity…

Do you realize that with AI, you do not track RANK but VISIBILITY?

AI engines aren’t search engines that scan a live index or fetch pre-written content. AI engines predict responses from patterns baked into their datasets!

AI answers in large language models (LLMs) are predictive by nature because these systems are designed to generate responses based on patterns and probabilities learned from incredible amounts of data.

During training, LLMs are exposed to enormous datasets: articles, web pages and social conversations.

They analyze these texts to learn relationships between words, phrases, and ideas. The output (the answer) of ChatGPT or Google’s Gemini depends on the input (your prompt).

Did you ask

💬 “What is LinkedIn known for?”

The LLM’s training data will use associations:

↳ professional networking

↳ business profiles

↳ recruiters

↳ resumes

↳ jobs

These words aren’t random: they’re the most frequent and coherent pairings your favorite LLM has seen with the brand “LinkedIn” across millions of examples…

ChatGPT might respond:

LinkedIn is known for being a professional networking platform where people connect with colleagues, find jobs and showcase their careers.

I didn’t check but that’s a high-probability answer because of how Linkedin’s portrayed in corporate blogs, tech news and social networks…

Now, suppose you tweak the question and ask:

💬 “What do people say about LinkedIn?”

The prediction shifts slightly.

The model might lean into sentiment analysis or recurring themes from the dataset. Maybe the LLM has seen LinkedIn tied to phrases like “𝘨𝘳𝘦𝘢𝘵 𝘧𝘰𝘳 𝘫𝘰𝘣 𝘩𝘶𝘯𝘵𝘪𝘯𝘨” but also “𝘴𝘱𝘢𝘮𝘮𝘺 𝘈𝘐 𝘤𝘰𝘮𝘮𝘦𝘯𝘵𝘴” (hello).

In that case, the LLM will predict:

People say LinkedIn’s a top spot for job opportunities and professional connections, though some complain about annoying spammy AI comments

That’s the LLM mirroring the mixed bag of narratives it’s absorbed, weighted by what’s most prominent.

So, when you search for brands, it will trigger specific contextual predictions.

It is still all probabilities, but they’re shaped by the brand’s unique footprint in the text the LLM has been trained on…

🛑 PLEASE: stop saying “RANKINGS”!

When it comes to large language models, the idea of “rankings” does NOT fit because we’re not dealing with a fixed hierarchy or a measurable scoreboard the way we rank search engine results.

With AI answers, it’s a lot more messy than that!

In LLMs, brands do not compete in a straight line where one can be “ranked” above the others.

With AI, there are NO RANKINGS! 🥇

The answer depends on what you’re asking!

Think VISIBILITY and BRANDING but not ranking!

My ironic conclusion?

AI models just echo brand fame.

That’s why after winning in search engines, BIG BRANDS will also win in AI answers! 🏆
The unfair competitive advantage they had in search engines is also an advantage in AI answers.

© Feb 27, 2025 – Elie Berreby

DOWNLOADABLE PDF: Why-Rankings-do-not-Apply-AI-Search-Elie-Berreby.pdf