Case Study: Analyzing 1,000 ChatGPT Brand Queries

"Does ChatGPT know my brand?"

This is the most common question we get from CMOs. The second is: "Why does it recommend my competitor?"

To answer this, we didn't just guess. We conducted a large-scale analysis of 1,000 transactional and comparative queries across three major LLMs: GPT-4o, Claude 3 Opus, and Gemini Advanced.

The goal: To reverse-engineer the "Recommendation Algorithm" of Generative AI.

The Methodology

We selected 10 verticals (SaaS, FinTech, E-commerce, etc.) and ran 100 queries per vertical. The queries fell into three intent buckets:

Discovery: "What is the best CRM for startups?"
Comparison: "HubSpot vs. Salesforce vs. [Brand X]"
Validation: "Is [Brand X] legit?"

We analyzed the output for:

Frequency: How often a brand was mentioned.
Sentiment: The adjectives used to describe the brand.
Order: The position in the list (1st vs. 5th).
Hallucination Rate: How often the model invented features.

Key Finding 1: The "Power Law" of Recommendations

In traditional SEO, being on Page 1 (Top 10) is good enough. In AI SEO, it's winner-take-all.

Our data showed a massive Power Law distribution:

The #1 recommended brand appeared in 82% of relevant queries.
The #2 brand dropped to 45%.
The #3 brand appeared in only 18%.

Insight: LLMs are designed to be "helpful assistants." They don't want to overwhelm the user with 10 options. They want to give the best option. If your brand isn't the clear category leader in the model's vector space, you are often excluded entirely.

Key Finding 2: "Brand Adjectives" Determine Ranking

Why does the model choose Brand A over Brand B? It comes down to the semantic adjectives associated with the brand entity.

We found that brands consistently recommended as "#1" had strong associations with specific "Superlative Adjectives" in the training data:

Reliable
Standard
Enterprise-grade
Fastest

Brands that were ignored often had "neutral" or "generic" associations (e.g., affordable, tool, platform).

The Takeaway: You cannot just optimize for "CRM." You must optimize for "The [Adjective] CRM." You need to own a specific attribute in the vector space. (See The Death of the 10 Blue Links for more on this).

Key Finding 3: The "Reddit Validation" Loop

This was the most surprising finding. For queries asking for "honest reviews" or "is X legit?", 68% of the AI's output correlated directly with sentiment found on Reddit and G2.

If a brand had a polished website but a negative Reddit sentiment (even from threads 2 years old), the AI would often add a caveat: "However, some users report issues with customer support."

The Mechanism: LLMs heavily weight user-generated content (UGC) from high-trust domains like Reddit and Stack Overflow during the "Reinforcement Learning from Human Feedback" (RLHF) phase. They are trained to trust "human" consensus over marketing copy.

Action Item: Reputation management on Reddit is no longer optional. It is core SEO.

Key Finding 4: Brand Co-occurrence

We noticed that brands were often recommended in "Clusters." If a user asked for "Best Email Marketing Tools," Mailchimp and Klaviyo almost always appeared together.

This is due to Co-occurrence. In the training data, these words appear near each other frequently. If you are a new challenger brand, your goal is to force Co-occurrence with the market leader. You want articles, press releases, and reviews that say "Brand X vs. Mailchimp." This trains the model to associate your vector with the leader's vector.

The "Invisible Brand" Problem

Of the 1,000 queries, 40% of the brands we tested (mostly mid-market) were never mentioned once. When we asked the model directly about them ("What is [Brand Y]?"), the model often hallucinated or gave a generic "I don't have enough information" response.

This is the Entity Gap. The model knows the words on your website, but it hasn't formed a stable Entity in its Knowledge Graph. It doesn't know what you are, only that you are.

How to Fix It: The 3-Step Plan

Based on this data, here is the playbook to move from "Invisible" to "Recommended":

Define Your Superlative: Choose the one adjective you want to own. "The Simplest CRM." "The Secure VPN."
Force Co-occurrence: Launch a "Vs." campaign. Create comparison pages, get listed in "Best X Alternatives" listicles.
Seed the Knowledge Graph: Use Wikidata and Schema markup to explicitly tell the model who you are. (Read: Why Schema Markup is Your API to the AI).

Conclusion

The battle for attention is moving from the search bar to the chat window. The brands that win will be the ones that understand they are optimizing for a neural network, not a crawler.

The data is clear: You are either the answer, or you are noise.

Curious about your own visibility? Learn How to Check Your Brand's Visibility in AI.