What ChatGPT's new graph reveals (and how it changes the decision funnel)
Sep 19, 2025
By Daniel Espejo, Founder & CEO at Omnia. 19th of September 2025
In this article:
What the ChatGPT usage graph really tells us (beyond reading percentages).
The surprises: why ‘practical guidance’ and categories such as health/fitness/beauty carry more weight than we imagined.
The compressed funnel: when learning, comparing, and choosing happen in a single response.
What this means for marketing and product teams (content, sources, organisation) without turning it into a never-ending manual.
How to measure what's essential as the map shifts.
What the OpenAI graph indicates
OpenAI has published the most comprehensive analysis to date on how people use ChatGPT, based on 1.5 million conversations, with one key point: most interactions focus on practical guidance, seeking information and writing. In other words, people come to solve everyday tasks rather than to chat or programme for hours. The study also highlights that usage is split between work and personal life, and that the value created is often decision support, something that traditional economic indicators do not measure well.
The NBER working paper accompanying the launch reinforces these ideas: ‘Practical Guidance’, ‘Seeking Information’ and ‘Writing’ account for nearly 80% of conversations; coding and self-expression are in relative minority. The message is clear: chatbots are establishing themselves as tools for productivity and judgement (not just output), and this is especially relevant in highly cognitive work.
What is truly surprising
Less ‘chit-chat’ than expected. The social/light-hearted part is small compared to result-oriented tasks. The AI is a teacher, editor, librarian and personal shopping assistant all at once.
Wellness competes with tech. Within ‘practical guidance,’ there is an unexpected spike in health/fitness/beauty: skincare, habits, evidence... categories considered ‘soft’ are appearing strongly. This is real demand for criteria and proof (not claims).
Writing/Editing is not just ‘I help you write’, but also deliver: emails, notes, proposals. In the professional sphere, writing ‘dominates’ over other tasks, precisely because the chatbot can generate digital outputs ready for immediate use.
The compressed funnel: when everything happens in a single response
If you look closely at the graph, you will see a pattern: the decision is accelerated. The user enters with a specific question and, without changing tabs, moves forward with the answer towards a plausible decision. In practice, this is how it happens:
Situation-related questions. ‘What is...?’ ‘How is it used...?’ and restrictions (budget, sensitive skin, integration, language).
Criteria within the same response. The model not only provides a definition; it also demonstrates what to look for and what to avoid when making comparisons.
Instant comparison. With these criteria, the answer already suggests well-reasoned options.
Product or action. Land on brands, products, suppliers, or next steps. All without leaving the thread.
That's a compressed funnel: discovery, consideration, and decision happening in the same place. The NBER frames it as value creation through decision support, precisely where humans are slowest: filtering, weighing, and deciding with incomplete signals.
Why it matters: if you don't exist in the specific information and how-to part, it's difficult to ‘enter’ later when the user is already looking at comparisons and options. It's not classic SEO of ‘attract traffic, then educate’; here, education and choice coexist in the same interaction.
Implications for your strategy
Design to resolve
In this context, the brands that win are not those that ‘sound nice’, but those that provide quotable evidence: tables with criteria, guides with methods (steps, limits, real cases), FAQs with numbers, screenshots and update dates. Form matters as much as substance: clear titles, compact sections, and data that is easy to copy and reference. (The OpenAI study emphasises that usage is task-focused, not general).
Wellness and consumption: real opportunity
The fact that health/fitness/beauty carries more weight than we imagined is no coincidence. These are categories with technical language and personal decisions. If you sell skincare, you should prioritise verifiable criteria (who it is suitable for, who it is not suitable for; how to evaluate an INCI; what ‘fragrance-free’ really means) and external sources to back it up. That is what a model can summarise and cite when constructing a response.
Localisation: adapting, not just translating
The usage pattern shows that translation competes with technical tasks. If you operate in multiple markets, prepare your own glossaries, local examples, and relevant metrics for each country. That way, when the user asks for ‘the same thing, in German and with a local twist’, the AI has the material it needs to adapt (not simplify).
Measuring without noise (what to look out for when the map moves)
The OpenAI paper and post make one thing clear: usage evolves and creates value both at work and outside of it; users deepen their usage over time. In this changing environment, you don't need twenty KPIs, but rather four outputs that tell you whether you are where decisions are made:
Decision prompt coverage: does your brand appear in the key situational questions in your category?
Consistency between engines: does it happen in several (not just one) and consistently?
Quality of mention: does the answer recommend you with arguments or just list you?
Time-to-inclusion: Once you publish/update evidence, how long does it take for it to start appearing?
This is the type of approach we use at Omnia: prioritising real questions, observing which engines and sources carry weight in each one, understanding the patterns and tracking their evolution.
Grounded example: from ‘what it is’ to ‘the product’ without changing screens
Put yourself in the shoes of someone looking for cosmetics:
‘anti-ageing cream for sensitive skin, fragrance-free, with clinical evidence.’
Good responses do three things:
provides context (what ‘anti-ageing’ means for sensitive skin);
provides criteria (ingredients, tests, tolerances, how to read the INCI);
proposes options that meet those criteria.
Where do you gain visibility?
That your criteria and data are openly available (and dated): tables, ‘updated on...’, links to sources.
Ensure that your description as an entity (what you are, who you are for...) is consistent across your website, documents, marketplaces, forums, and media.
That respected third parties (niche media, expert forums, YouTube) cite you as an example of these criteria.
None of this happens by ‘talking’ to a model in a private chat: the NBER and OpenAI are clear that adoption and value are explained by real tasks and decision support; what matters is what external signals the model can use to construct a reliable response.
Conclusion
The new graphic is a change of focus. ChatGPT and other engines are used to solve problems. That is why the funnel is compressed: in a single response, the user learns, compares and chooses. If your brand wants to be there, you need clarity and evidence: criteria, method, sources and consistency.
Three ideas to take away today:
Invest in specific information and how-to guides that a model can cite.
Publish structures (tables, checklists, rubrics) that can be easily integrated into responses.
Measure outputs (coverage of key questions, multi-engine consistency, mention quality, inclusion time) rather than chasing vanity metrics.
The map will continue to shift, but the fundamental principle remains stable: AI recommends what it can summarise and support. If you put that evidence out into the world and maintain it, the compressed funnel works in your advantage..