How We’re Really Using AI: Lessons From 1.1 Million Conversations

When generative AI exploded into the mainstream, predictions flew in every direction. Would it replace search engines? Destroy creative industries? Become the universal productivity tool we didn’t know we needed?

Now, looking at real-world usage data, we have something better than speculation: evidence. OpenAI’s latest system card for GPT-5 included a detailed analysis of more than 1.1 million conversations sampled between May 2024 and June 2025. The data offers a look into how people are actually using AI. And the results will challenge assumptions about its role in our digital lives.

The Top-Line Story: Writing and Practical Help Dominate

If you’ve ever opened a chat window with an AI, chances are you weren’t trying to debug a neural network or draft a doctoral thesis. More often, you were probably asking for help writing something or for guidance on a problem.

That instinct shows up clearly in the numbers. Writing-related tasks account for 28.1% of all usage, while Practical Guidance makes up another 28.3%. Together, they represent more than half of all conversations.

• Under Writing, the heaviest hitters are Edit or Critique Provided Text (10.6%) and Personal Writing or Communication (8.0%). AI is less about cranking out novels (fiction writing clocks in at just 1.4%) and more about smoothing drafts, sharpening tone, and helping people communicate clearly.
• Under Practical Guidance, Tutoring or Teaching (10.2%) leads, followed by How-To Advice (8.5%). Here, AI is less like a search engine and more like a patient explainer that adapts to your pace and context.

If the early years of the internet were about finding information, the AI era so far seems to be about making that information usable — rewriting, restructuring, and repackaging it to fit human needs.

Seeking Information: AI as a Research Partner

The second-largest category, Seeking Information (21.3%), highlights another familiar behavior: curiosity. Users turn to AI for answers to very specific questions — what the report calls Specific Info (18.3%).

What’s striking here is how much of this overlaps with traditional search engines. But instead of clicking through links, people are starting to expect a synthesized, conversational response. That shift may explain why Google and other incumbents are racing to reimagine search around AI.

Interestingly, Purchasable Products (2.1%) and Cooking and Recipes (0.9%) represent only slivers of usage. Despite fears that AI might devour e-commerce and lifestyle publishing, the data suggests users aren’t yet outsourcing shopping or meal planning at scale.

Technical Help: The Quiet Backbone

One might expect programming and math to dominate, given AI’s reputation among developers. But Technical Help makes up just 7.5% of conversations.

• Computer Programming contributes 4.2%.
• Mathematical Calculation accounts for 3.0%.
• Data Analysis barely registers at 0.4%.

This doesn’t mean developers aren’t using AI — it may simply reflect how specialized technical tasks are compared to the broader base of general users. Still, when you zoom out, it’s clear that AI isn’t only a coder’s assistant. It’s increasingly everyone’s assistant.

Self-Expression and Human Connection

Not everything is utilitarian. About 4.3% of conversations fall under Self-Expression, which includes Relationships and Personal Reflection (1.9%), Greetings and Chitchat (2.0%), and a small slice of Games and Role Play (0.4%).

These aren’t trivial. They suggest that people sometimes treat AI as a sounding board, a practice buddy, or even a low-stakes social outlet. That may raise eyebrows, but it reflects a deeper truth: AI can be as much about companionship and expression as about productivity.

Multimedia: Less Than You’d Think

Despite splashy demos of AI-generated art, music, and video, Multimedia tasks represent just 6.0% of the sample. Of that, Create an Image makes up 4.2%, while Generate or Retrieve Other Media is only 1.1%.

It turns out that while image generation captures headlines, it doesn’t yet dominate day-to-day use. Text remains king. (See “sidebar” below)

The Long Tail: Niche Uses, Big Potential

Some smaller categories hint at future growth areas:

• Health, Fitness, Beauty, or Self Care accounts for 5.7%.
• Creative Ideation shows up at 3.9%.
• Translation represents 4.5%.

These may look modest, but they represent millions of interactions. As AI tools become more specialized and trustworthy, these niches could expand rapidly. Imagine AI nutritionists, career coaches, or creative partners built on top of today’s general models.

What This Says About Us

The distribution tells a story not just about technology, but about people.

We are not using AI primarily to entertain ourselves or escape into fantasy. Nor are we using it mainly to shop or outsource decision-making. Instead, we’re leaning on it to clarify, explain, and refine.

In other words, AI today functions less like an oracle and more like a writing coach, tutor, and thought partner. The tasks may look small — editing an email, explaining a math step, brainstorming phrasing — but in aggregate, they reveal a profound shift in how we interact with knowledge.

The Next Phase

As the technology matures, these usage patterns may change. We may see more multimodal adoption as image, audio, and video become as seamless as text. We may see an uptick in decision-making tasks as trust in AI deepens. Or we may see usage fragment into specialized domains, each with its own tailored assistant.

But for now, the message is clear: AI is embedding itself into the most human of activities — writing, learning, and asking questions. It’s not replacing us. It’s augmenting the way we think and communicate.

And that might be its most transformative role of all.

SIDEBAR: Why Multimedia Lags Behind

If AI image and video tools dominate headlines, why don’t they dominate usage? The data shows only 6% of conversations involve multimedia. That gap comes down to four factors:

• Friction. It takes more effort to describe and refine a visual than to ask for a quick edit to text.
• Need. Most of us write daily; few of us need new artwork or video clips on a regular basis.
• Trust. People are confident tweaking AI text, but feel less certain about using AI visuals in professional or personal contexts.
• Workflow. Text drops seamlessly into documents, emails, and chats. Multimedia often requires extra tools and steps.

In short, multimedia AI is still in the “wow” phase, impressive demos and viral post, while text-based AI has already entered the “workflow” phase. As integrations deepen and friction drops, those numbers are going to swing sharply.

I share AI insights while building the world’s first AI-native Corporate Intelligence and Investigation Agency.