A Stella Rising Study of 524 LLM Users
The keyword is dead—or at least, the way SEO teams have spent a decade thinking about keywords no longer matches how people actually search.
When we asked 524 active LLM users to write the prompts they'd type into ChatGPT for an everyday task, only one in five wrote a short, keyword-style query. Sixty percent phrased their prompts as full questions. Nearly half wrote between 6 and 15 words; another quarter wrote 16 to 30 words. The average LLM prompt looks nothing like the three-word search-bar query that built traditional SEO.
That matters because LLM search isn't a future shift—it's where consumer behavior already lives. Half of our respondents use ChatGPT and similar tools daily for tasks they used to do on Google. Sixty-eight percent say they trust an LLM recommendation more than a Google result. And 85% click through to cited sources at least sometimes, meaning citations aren't just brand impressions—they're traffic.
In January 2026, we surveyed 524 active LLM users through Stella Rising’s research community to understand not just how often people use these tools, but how they're using them—the actual words, prompt structures, and intent patterns that shape what AI shows them. Below, we break down what we learned and the six shifts SEO teams need to make to stay visible in a GEO world.
|
50%
Use LLMs daily+ to replace traditional search
|
68%
Trust ChatGPT more than Google
|
85%
Click cited sources at least sometimes
|
72%
Use voice chat with LLMs monthly+
|
50% ARE REPLACING GOOGLE DAILY
Half of LLM users now turn to tools like ChatGPT on a daily basis to accomplish tasks they previously used traditional search engines for. Breaking that down further: 21% use LLMs several times per day, while another 29% use them daily. Only 4% of active LLM users say they "never" use these tools for search-like tasks.
This isn't theoretical displacement. It's happening now, at scale.
| Several × Day |
|
21% |
| Daily |
|
29% |
| Weekly |
|
30% |
| Monthly |
|
16% |
| Never |
|
4% |
THE TRUST SHIFT: 68% TRUST CHATGPT MORE THAN GOOGLE
| 68% |
The Trust Shift
Trust ChatGPT more than Google.
When asked directly, two-thirds of LLM users say they trust an AI recommendation more than a search result. The reasons: more detailed answers, no ads, faster results, personalized responses.
|
Perhaps the most striking finding: when asked directly whether they trust recommendations from ChatGPT more than Google, 68% said yes.
This trust gap isn't about brand loyalty—it's about user experience. In open-ended responses, users cited reasons like more detailed answers, no ads to sort through (yet!), personalized responses, and faster results.
For SEO professionals, this trust differential has massive implications. If two-thirds of your brand audience trusts an AI recommendation more than a search result, optimizing for AI visibility isn't optional, it's essential.
CITATIONS MATTER: 85% CLICK SOURCES AT LEAST SOMETIMES
A common assumption is that LLM users accept AI-generated answers at face value. Our data says otherwise.
| Always |
|
22% |
| Sometimes |
|
63% |
| Never |
|
15% |
That means 85% of users engage with citations at least some of the time. In GEO terms, being cited isn't just about brand awareness—it also drives actual traffic.
VOICE IS MAINSTREAM: 72% USE VOICE CHAT MONTHLY
Voice interaction with LLMs has gone mainstream far faster than many predicted. Our survey found that 72% use voice chat with LLM tools at least monthly, with 34% using it daily or more.
| Several × Day |
|
14% |
| Daily |
|
21% |
| Weekly |
|
25% |
| Monthly |
|
13% |
| Never |
|
28% |
This has significant implications for content optimization. Voice prompts tend to be more conversational, longer, and more question-oriented than typed queries.
WHAT ARE PEOPLE USING LLMS FOR?
We asked respondents to select all use cases that apply. The results paint a picture of LLMs as comprehensive information assistants:
|
Use Case |
Percentage |
|
Answer questions |
73% |
|
Learn or understand a concept |
48% |
|
Compare options |
40% |
|
Help troubleshoot an issue |
36% |
|
Find the best prices |
35% |
|
Creative generation |
29% |
|
Find location / where to buy |
29% |
|
Check specifications |
28% |
|
Create lists |
20% |
|
Aggregate reviews |
15% |
|
Complete a purchase |
14% |
HOW PEOPLE PROMPT: THE DEATH OF THE KEYWORD
One of the most valuable aspects of our research was asking people to write prompts. We gave them a scenario ("you need a new pair of shoes") and asked them to write what they would prompt in ChatGPT.
| 1–5 Words |
|
20% |
| 6–15 Words |
|
44% |
| 16–30 Words |
|
25% |
| 31–50 Words |
|
9% |
| 51+ Words |
|
3% |
|
60% Question ("What is the best…")
31% Mixed
9% Command ("Compare X vs Y")
|
60% of users phrase prompts as questions, a significant shift from traditional search behavior.
INSIDE REAL PROMPTS: PATTERNS THAT MATTER FOR GEO
We analyzed hundreds of free-text prompts to identify patterns. Here's what we found:
|
25%
"Best"
Superlative queries dominate.
|
28%
Price
Budget & value language.
|
16%
Near Me
Local intent migrated to LLM.
|
32%
Personal
Size, condition, lifestyle.
|
13%
Brand
Brand + attribute combos.
|
1. "Best" Is the Magic Word
25% of prompts include the word "best." This validates what many GEO practitioners have observed: superlative queries dominate LLM interactions.
2. Price Sensitivity Is High
28% of prompts mention price or budget constraints. Users aren't just looking for recommendations—they're seeking value.
3. "Near Me" Translates to LLM
16% of prompts are location-based. The "near me" query pattern has successfully migrated from Google to ChatGPT.
4. Personal Context Changes Everything
32% include personal attributes like size, foot conditions, job requirements, or lifestyle factors.
5. Brand Mentions Are Common
13% mention specific brands. With traditional SEO, branded searches are their own category. With LLM prompts, brands appear alongside attributes and requirements.
WHAT THIS MEANS FOR GEO STRATEGY
1. Optimize for Questions, Not Just Keywords
With 60% of prompts in question format, your content needs to explicitly answer questions.
2. Claim Your "Best" Positioning
A quarter of all prompts include "best." If your brand is not appearing in "best [category]" responses, you're missing the highest-intent queries.
3. Build for Voice
With 72% using voice at least monthly, conversational query patterns are no longer edge cases.
4. Citations = Traffic
85% of users click sources. Being cited isn't just a vanity metric—it's a traffic driver.
5. Price and Availability Matter
Over a quarter of prompts mention price; 16% are location-based. Product pages need to surface pricing and location data clearly.
6. Context Is King
Users provide rich context in their prompts. Content that addresses specific use cases and user contexts will outperform generic category content.
THE BOTTOM LINE
The data is unambiguous: LLM-based search isn't a future state, but the present reality for a majority of users. Half are using these tools daily. Two-thirds trust them more than Google. And new consumer behavior patterns require a fundamentally different optimization approach than traditional SEO.
For brands with existing SEO initiatives, this isn’t cause for alarm but for adaptation. The brands that move now, and treat GEO as a core discipline rather than an experiment, will have a significant advantage as the new search behaviors become more dominant. The window for early-mover advantage is closing!
Stella Rising is a full-funnel marketing and media agency specializing in beauty, health, food, and lifestyle brands. Our GEO practice helps brands navigate the shift from traditional search to AI-assisted discovery.
Key Statistics At-A-Glance
|
50%
Use LLMs daily+ to replace search
|
68%
Trust ChatGPT more than Google
|
85%
Click cited sources at least sometimes
|
|
72%
Use voice chat monthly+
|
60%
Phrase prompts as questions
|
25%
Include "best" in prompts
|
|
28%
Mention price / budget
|
16%
Are location-based
|
n=524
Active LLM users (Jan 2026)
|
Methodology
This survey was conducted in January 2026 with 524 respondents who had used an LLM search tool in the past 30 days. Respondents were sourced through Centiment's consumer panel. The survey included both multiple-choice questions about behavior and open-ended questions asking respondents to write actual prompts.
This research builds on Stella Rising's August 2025 survey of beauty-focused consumers (n=178), allowing us to track evolution over a six-month period and compare general audience behavior to category-specific patterns.
COMMENTS