Inside 13,252 ChatGPT Conversations: What AI Use Really Looks Like

🧠 What We Found

• Chats start with briefs, not queries: Avg. prompt = 348 words (70× longer than a Google search query).
• AI takes orders: 1 in 5 chats start with a direct command, not a question.
• One-and-done conversations rule: Despite long prompts, most chats end after just 1.7 messages.
• Creativity leads the way: Content generation (13.8%) and basic explanations (14.3%) top all use cases.

Three years after ChatGPT’s launch, the headline numbers are almost too big to comprehend. Over 800 million weekly active users. 5.4 billion monthly website visits. $3.7 billion in revenue for 2024. The adoption story is remarkable, and no consumer technology has ever scaled this quickly.

But outside the eye-popping stats, there is a lot we don’t know, mainly how exactly people are using ChatGPT?

OpenAI doesn’t publish user analytics (yet). There’s no dashboard showing conversation patterns, prompt lengths, or interaction behaviors. We know people like making cartoon images, using it for homework, and asking it to do plenty of things that are…kind of strange. And plenty of people will sell you a course all about prompting!

But so much remains unknown. Are people becoming AI power users, or are they still just experimenting? How long are chats? What types of questions are people asking?

I decided to find out. Over the past few weeks, I collected and analyzed 13,252 publicly shared ChatGPT conversations. The findings reveal what information discovery is likely to look like in the years ahead.

ChatGPT conversations are detailed but brief

The first surprise came from the conversation length. The average ChatGPT user types 348 words per conversation — that’s nearly 70 times longer than a typical Google search query of around 5 words. Power users are often bringing detailed context, specific goals, and nuanced requirements to their AI interactions. The longest conversation in our data set was 149 messages.

Most interesting to me: The opening messages average 103 words. We’ve talked about the growth of conversational search, and this is another example of that.

Gone are the days of typing “lower back pain” into Google. A ChatGPT chat might say, “I spent all day yesterday walking around Six Flags and now my back hurts on the left side. I rode a lot of rides and first noticed the pain this morning when I woke up. I have a 5k race coming up next week, and I’m not sure if I’ll feel better by then…”

Users understand intuitively that AI needs context to be useful, even if they haven’t read prompt engineering guides and are starting conversations with that frame of reference.

But here’s the paradox: Despite writing much longer prompts, users send an average of just 1.7 messages per conversation. Most sessions are brief but purposeful. Multi-turn conversations happen, but extended back-and-forth remains surprisingly uncommon.

A lot of this is powered by chats with explanation intent: “Explain to me how a v8 engine works and how it’s different from a v6.” ChatGPT provides a detailed, effective answer, and then the chat ends.

This pattern suggests something important about how people currently view ChatGPT:

While power users engage in long, detailed chats, most users are treating it more like an ultrasophisticated search engine than a totally new technology. They are bringing longer, personalized conversational queries, getting an answer, and then leaving.

This tracks with typical patterns of new technologies: The easiest way for mass adoption is to view it as a slightly different way of doing something that is already familiar…in this case, traditional search.

Questions rule but commands are growing

While questions remain popular (32.9% of conversations), a surprising 19.3% of ChatGPT interactions are command-focused: Users giving instructions and delegating tasks rather than asking for general information. This represents a fundamentally new type of search behavior that didn’t exist before generative AI. People are not even searching so much as they are commanding…asking ChatGPT to deliver a particular output.

Command-style chats that are most popular include:

• Copywriting (“Write me a blog post that…”)
• Building reports (“Give me a detailed analysis of…”)
• Instructions (“I want step by step…”)

The real usage categories tell the story

Aside from questions and commands, there are a lot of specific use cases that were most popular in the data we analyzed. Some of these use cases are well-defined and others are more unique ways of using ChatGPT.

Let’s unpack this a bit more: First, there is a very long tail of ChatGPT usage that doesn’t fall into any of the categories above. Lots of these long-tail use cases are very random and one-off, not unexpected for an emerging technology that people are still figuring out how to use.

For those that are easily classified, there are two overarching themes that emerge for how ChatGPT is being used currently.

Learning & Understanding

The most common way ChatGPT is utilized is for learning and understanding: explanations, how-to’s, definitions, and deep analysis of subject matter. 31% of conversations fall into this overarching category.

Despite all the advanced capabilities, foundational understanding remains the top use case (again: positioning ChatGPT as a new/better way to search is a great flag to plant for early adoption). People are using AI to learn and comprehend, treating it like an infinitely patient tutor.

Learning & understanding examples from the data:

• “Can you explain what machine learning is and how it differs from traditional programming? I’m trying to understand this for a project at work.”
• “What is the difference between B2B and B2C marketing strategies? I need to present this to my team next week.”
• “Help me understand what SEO actually means and why it matters for small businesses like mine.”
• “Walk me through the complete process of setting up Google Analytics 4 for an e-commerce website. Include what events I should track and how to set up conversion goals.”
• “Give me a step-by-step checklist for onboarding a new remote employee, from equipment setup to their first week training schedule.”
• “How do I create a content calendar for social media? Break it down into actionable steps and tell me what tools I’ll need.”
• “Analyze the pros and cons of implementing a freemium pricing model for a SaaS startup. Consider customer acquisition costs, lifetime value, and competitive positioning.”
• “What are the underlying economic factors driving the current labor shortage in healthcare, and how might this impact hospital operations over the next 5 years?”
• “Break down why some companies succeed with remote work while others struggle. What are the key organizational and cultural differences?”

Creating & Producing

The next bucket of activity involves creating and producing things…users frequently prompt ChatGPT to generate copy, product descriptions, stories, names, and ideas. This reveals how quickly AI has become a content production tool, especially for businesses trying to scale their marketing efforts. There is also a ton of usage here from developers and financial analysts, who rely heavily on ChatGPT for code assistance and financial projections.

Creating & Producing examples from the data:

• “Write a compelling product description for a noise-canceling headphone targeting remote workers. Focus on productivity benefits and comfort during long calls.”
• “Create 5 different subject lines for an email campaign about our new project management software. Make them attention-grabbing but professional.”
• “Help me brainstorm 10 creative names for a dog walking service that sounds friendly
• “I have a CSV file with sales data by region and month. Can you help me write a Python script to calculate year-over-year growth rates and identify the top performing regions?”
• “Create a Google Sheets formula that will automatically categorize expenses as ‘High,’ ‘Medium,’ or ‘Low’ priority based on dollar amounts and department codes.”
• “Help me write SQL query to find customers who made purchases in Q1 but haven’t bought anything since. I need their contact info for a re-engagement campaign.”
• “That email draft is too formal for our company culture. Can you rewrite it with a more casual, friendly tone?”
• “Make that product description shorter and focus more on the time-saving benefits rather than technical features.”
• “The strategy outline is good, but can you break down the implementation section into more specific, actionable steps?”

Overall, one of my other takeaways is that the AI power users you read about online or watch on YouTube are very much still the exception, not the rule. Only 2.4% of conversations analyzed involved somebody giving the AI a persona (“You are an experienced financial analyst…”) or uploading data into ChatGPT.

The future of human-AI interaction

What emerges from this data is a picture of AI adoption that’s broad but still shallow. Millions of people use ChatGPT regularly, but most haven’t yet developed sophisticated interaction patterns. They’re treating it more like an advanced search engine than a collaborative partner.

The conversation data reveals we’re still in the early chapters of AI adoption. Lots will change in the next few years, including how we all use ChatGPT and related platforms. And hopefully soon there is a bit more visibility for all of us directly from OpenAI on how we’re using ChatGPT.

Ready to build AI experiences that match how your customers actually work? Let’s talk about creating marketing strategies that anticipate the AI-native future rather than react to it.

Methodology Notes

We analyzed over 20,000 publicly shared ChatGPT conversations. We removed any from the analysis that were gibberish or repeated the same chat over and over. Since all of these conversations were publicly available, it does limit our data set to conversations that folks chose to share. This could skew our findings, as the way people interact in private conversations vs. shared conversations may differ.