Study Reveals How People Really Use ChatGPT (And What It Means for Consumer Brands)

A study from Resoneo, published in August 2025, looked at 87,725 public [ChatGPT](/blog/chatgpt-already-real-search-engine-brand-2025) conversations to answer one straightforward question: how do people actually use this thing? The dataset is a rare, behind-the-scenes glimpse into real conversational behaviour. The headline finding for consumer brands: queries are longer, conversations are deeper and ChatGPT is doing far more than search. Below are the five biggest signals from the data and what each one means for how you should think about [AI search visibility](/blog/ai-search-101-2025). ## 1. Queries are getting longer and more nuanced The average first message clocks in at 8 words. Google's average is 3.5. That's more than double and the pattern points to richer, context-laden queries. Users aren't typing keywords. They're describing situations. *"I sleep hot and have sensitive skin, recommend breathable bed sheets"* beats *"best bed sheets"* every time inside ChatGPT. **What it means for your content.** The keyword-stuffed product descriptions of 2018 don't work here. The model isn't matching tokens - it's parsing intent. Your content needs to mirror the long-tail constraint-led questions that real shoppers ask, not the head-keyword brief. ## 2. Conversational depth reigns supreme 60% of conversations involve five or more exchanges. 35% extend beyond eight messages. This is sustained dialogue, not one-shot questions. The shopper asks, the model answers, the shopper refines, the model narrows. By exchange five they've usually compared options, named constraints and arrived at a shortlist. **What it means for your content.** Your brand has multiple entry points to be cited. The first answer might not include you, but the third refinement might. The shopper's path through the conversation includes nuanced refinements that your content needs to be ready for. A bullet that addresses *"sensitive skin"* needs a partner FAQ that addresses *"sensitive skin in winter when also using retinol"*. ## 3. ChatGPT's role goes far beyond fact-finding Only 48.3% of interactions ask for straightforward factual information - the kind where ChatGPT directly competes with search engines. The other half is split across: - **Opinion or suggestion** (16.2%) - *"What should I get my partner who likes Japanese whiskey?"* - **Creative content generation** (14.7%) - *"Write me a list of dinner-party ideas for six vegetarians"* - **Code, analysis, problem-solving** - the long tail of remaining categories. **What it means for your content.** Consumer brands should care most about the opinion/suggestion bucket. That's the recommendation moment - where ChatGPT acts as a friend giving a shortlist and where being included or excluded directly translates to sales. If you're investing in [GEO](/blog/geo-seo-aeo-consumer-brands-where-focus-budget-2026), this is the bucket your strategy should target first. ## 4. Tech and dev use isn't a niche About 10% of conversations contain code snippets or programming tasks. GitHub is among the most-cited domains across languages. For consumer brands, this matters less directly - but it tells you something useful. Developers are now ChatGPT's most engaged power-users, which means the model's behaviour in technical domains is highly tuned. The same precision the model brings to code questions is what it now brings to product comparisons. Be precise. ## 5. A truly global phenomenon English dominates at 55.5%. Japanese is second at 21.5%. The top subjects are Science and Tech (30%) and Politics and Society (18%). **What it means for international brands.** If you sell into Japan, ChatGPT is already operating at meaningful scale among Japanese-speaking consumers. The localisation work that used to be a "nice to have" is now visibility-critical. ## What this means for consumer brands Three takeaways from the data. **Long-form content can be cited more easily than short content.** The model is parsing 8-word context-rich queries. Your single-sentence product blurb doesn't have enough surface area for a constraint-led match. Comprehensive guides, comparison pages and FAQ blocks that cover the full prompt vocabulary do. We've covered the [foundational moves](/blog/how-make-website-show-chatgpt-answers-2026) in a separate piece. **FAQs need to handle refinement, not just first-question.** With 35% of conversations going past eight messages, your FAQ section needs depth - not just *"is it vegan"* but *"is it vegan, sweetened with stevia and shelf-stable for camping trips"*. Model the constraint stack. **The opinion/suggestion bucket is where consumer brands win or lose.** It's the 16% that drives the recommendation moment. Optimise specifically for *"what should I buy if [constraint stack]"* prompts. The brands that learn to speak in the model's actual language - long, constraint-rich, refinement-friendly - get cited. The brands that keep writing for keywords get skipped. ## Important caveats The dataset has real limitations worth flagging. The insights come only from conversations users opted to make public. That skews towards more open, sharing-friendly users and may not fully represent private or professional use cases. The corpus also over-indexes on English, on tech-curious early adopters and on the kinds of questions people are willing to be seen asking. Still, it's the largest public-data analysis of ChatGPT behaviour to date and the patterns it surfaces match what we see in our own client work. Treat it as directional, not definitive. ## Want to dive deeper? The original analysis is from [Olivier de Segonzac at Resoneo](https://www.linkedin.com/posts/resoneo_new-exclusive-study-powered-by-resoneo-activity-7360728270062907393-mInJ/). Worth reading in full if you build content for a category where AI discovery is rising. If you want a Lmo7 read on what this means for your brand specifically - including a Share of Model audit on your top 30 buyer prompts - [get in touch](/contact). That is the shift.