Most AI chats have no commercial intent, users usually ask short questions, and most conversations end after just two turns. Those findings come from a recent analysis by Dan Petrovic, director of AI SEO agency Dejan, who examined millions of conversational turns to show how people actually use AI assistants.
Why we care. As SEOs and marketers race to “optimize for AI,” Petrovic’s analysis suggests the industry is misreading how people actually use AI assistants. Most chats function as multi-step tasks, not keyword-style queries. And users aren’t flooding AI with “buy” queries – they’re exploring problems and comparing options.
By the numbers. Petrovic analyzed 4.4 billion characters, 613 million words, and 3.9 million conversation turns:
- Median chat: 2 turns (a quick question, quick answer).
- Averages hide a long tail of heavy sessions driven by users pasting documents for summarization or analysis.
- Median words per session: 430
- More than 80% of chats are under 1,000 words.
- Only 4.2% exceed 2,500 words, and these typically represent the most complex, highest-value tasks: editing, coding, tutoring, and data analysis.
- Mean words: 732
- This is heavily skewed by long document drops.
- Assistant output: ~1.5x the user’s.
- Median user contribution: 16-17% of the conversation.
How people actually use AI assistants. Petrovic classified 24,259 sessions across 42 intent categories and found that most AI chats aren’t commercial – 64.6% sit outside any purchase funnel. Users write, brainstorm, plan, learn, analyze, or simply chat. Here’s the breakdown:
- Other: 25%
- Jailbreak attempts, roleplay scenarios, and highly specialized requests dominated here.
- Brainstorming: 7.7%
- Planning: 6.5%
- Conversation / emotional support: 6.2%
- Analysis: 5.7%
- Learning: 4.7%
- Transformation (summaries, translations): 4.6%
- Creation (writing, code, docs): 3.9%
35.4% of chats showed any commercial intent. And most were early funnel. Other findings:
- Awareness (10%) and consideration (8.5%) together made up 18.5%, which Petrovic highlighted as the strongest territory for product content.
- Post-purchase needs (5.1%) outranked transactional support (4.8%), discovery (4.1%), and decision support (2.8%), indicating that users turned to AI more for “how do I use or fix this?” than “should I buy this?”
Bottom line. AI assistants are used far more for creation, cognition, and conversation than for commerce.
The report. How do people use AI assistants?
Search Engine Land is owned by Semrush. We remain committed to providing high-quality coverage of marketing topics. Unless otherwise noted, this page’s content was written by either an employee or a paid contractor of Semrush Inc.
