We analyzed 32,961 queries on SearchGPT, Gemini and Perplexity - Q2 2025 Study

Q2 2025 study on LLM ranking factors based on 32,961 queries across three LLMs. For the latest insights, see our Q3 2025 study with 184,128 queries.

Research

When we first launched Qwairy, our GEO tool, our mission was to help marketers improve their brand visibility in LLMs like ChatGPT, AI Overview (Gemini), and Perplexity.
Search is evolving. The way we've browsed the web for the past 20 years is slowly being replaced by a new paradigm.
I notice it every day. Just last weekend, I was with my mom.
When she wanted to find a coffee shop, she immediately said:

"Let's see what ChatGPT recommends."
While building Qwairy we quickly understood that marketing managers needed data to backup their analysis.
After a lot of hard work, we're proud to present this comprehensive study, based on 32,961 queries across three LLMs (SearchGPT, Gemini, and Perplexity).
Let's explore what we can learn to boost your brand's presence in LLM results.

Context, Data, and Disclaimer

Before we start we want to share with you how we generated and gather the data.
All data was generated on our GEO tool Qwairy. Qwairy is a tool that helps you improve your brand presence on LLMs.
Data comes either from our own brand analyses or from clients who have added their brands to the platform.
In total, we generated 32,961 answers across three LLMs (SearchGPT, Gemini, and Perplexity), all between late 2024 and early 2025. We asked the same set of questions to each model.
Each answer also includes the sources used by the LLMs - totaling 59,992 sources (since one answer can include multiple sources).
Like any dataset, there are inherent biases. We aim to be as transparent as possible.
All responses were generated via the LLMs' APIs. The models used were GPT-4-o-mini Search (also known as SearchGPT), Sonar (Perplexity), and Gemini 2.0.
Our dataset includes a variety of brands (e-commerce, SaaS, blogs, etc.).
We plan to conduct this analysis quarterly to support SEOs, marketers, and CMOs. As more people use Qwairy, each report will become more comprehensive.

Part 1 : Answers Analysis

The first aspect we analyzed was the responses generated by the LLMs.
We aimed to identify recurring patterns across these answers.

Number of Competitors Mentioned per Query

With Answer Engines and LLMs, a new paradigm is emerging in search marketing.
It's no longer about appearing in the classic "top 10" Google results.
Today, the goal is to be explicitly mentioned in the LLM's answer.
If you're Tesla, and someone searches for the best electric car using an LLM, you want to be the first brand mentioned.

Characters analyzed by answer

Most used expression by LLMs

Most used expression by SearchGPT

Most used expression by Gemini

Most used expression by Sonar

Way of talking by LLMs

How does SearchGPT talk?
✅ Impersonal form (e.g., "it is recommended") : Present in 17.8% of all SearchGPT answers.
✅ Use of imperative (e.g., "try", "take advantage") : Present in 25.3% of all SearchGPT answers.

How does Gemini talk?
✅ Impersonal form (e.g., "it is recommended") : Present in 38.42% of all Gemini answers.
✅ Use of imperative (e.g., "try", "take advantage") : Present in 32.69% of all Gemini answers.

How does Sonar talk?
✅ Impersonal form (e.g., "it is recommended") : Present in 11.68% of all Sonar answers.
✅ Use of imperative (e.g., "try", "take advantage") : Present in 16.35% of all Sonar answers.

Conclusion : Response Analysis

We observed clear differences between the models: Gemini tends to provide longer, more comprehensive answers, while SearchGPT is more concise and selective in its citations.
The most frequently used expressions and discourse structures also reveal consistent patterns. LLMs favor cautious, structured, and often impersonal communication.
Finally, the near-systematic use of bullet points (with "-") shows that LLMs prefer clear, organized, and easily digestible formats.

Part 2 : Sources Analysis

The second part of this analysis is to analyze all the sources used by those LLMs.
Over the 32,961 queries that were added on Qwairy, the LLMs quoted 59,992 different sources.

What are the most used Website by Gemini?

What are the most used urls as a Source by ChatGPT?

Common words in Source URL By ChatGPT

Conclusion

With over 32,000 queries analyzed and nearly 60,000 sources examined, this is the most comprehensive study to date on LLM ranking factors.
It confirms what many marketers already suspected: Search as we know it is evolving - fast.
This study is just the beginning. We plan to update it quarterly with more data and sharper insights.