Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe short version: OpenAI’s study found that consumer ChatGPT use was mostly about everyday guidance, information, and writing—not workplace coding or emotional companionship. By July 2025, about 70% of analyzed consumer messages were unrelated to work, while practical guidance, information-seeking, and writing made up roughly 77–78% of messages.
Those figures describe messages, not users, and the study covered consumer ChatGPT activity through July 2025—not live usage in September 2026. That distinction matters when considering what AI may mean for household finances, work, education, and productivity.
As an Amazon Associate I earn from qualifying purchases.
What OpenAI actually studied
OpenAI analyzed consumer ChatGPT usage across several related datasets, including roughly 1.1 million conversations for broad message classification, about one million de-identified logged-in consumer messages collected between May 2024 and June 2025, later samples extending through July 2025, and a smaller sample of approximately 40,000 users for occupation and work-activity analysis.
The company’s promotional summary refers to roughly 1.5 million conversations. The technical paper describes several component samples with different purposes and dates, so those figures should not be treated as one perfectly uniform survey.
#1 Best Overall
The research did not cover ChatGPT Enterprise conversations, API traffic, third-party products using OpenAI models, other AI assistants, most logged-out usage, users who opted out of message sharing for model training, or users who identified themselves as under 18.
OpenAI says human researchers did not read the raw messages used in the analysis. Instead, personally identifying information was removed, automated classifiers categorized messages, and researchers examined aggregate classifications and metadata. Human-coded samples were used to validate some of the classifiers.
That privacy design is significant, but it is not the same as saying no systems processed user content. Automated models still processed the text, and classification errors or sampling choices can affect the results. The paper reports relatively strong agreement for its work-related classifier, with Cohen’s kappa of about 0.83, and for its Asking/Doing/Expressing classifier, at about 0.74. Agreement was much weaker—about 0.14—for interaction-quality judgments such as whether an answer appeared successful or satisfying.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Read the full working paper and OpenAI’s summary for the methodology and source data.
The biggest finding: ChatGPT is mostly a personal-use tool
By July 2025, approximately 70% of consumer ChatGPT messages were classified as non-work-related and 30% as work-related. Non-work use was growing faster than work use.
That does not mean personal use is frivolous. “Non-work” can include tutoring, health information, travel planning, household organization, financial education, job searching, and other valuable unpaid work. For a household, saving time on research or producing a clearer budget draft can have real value even if it does not appear as income or show up in GDP.
Rank #2
It also does not mean 70% of people used ChatGPT personally. Heavy users generate more messages, so message shares cannot be converted directly into user shares, time shares, or economic-value shares.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsWhat people use ChatGPT for
Three broad categories dominated the study:
- Practical guidance: advice and help with everyday decisions, planning, and problems.
- Seeking information: explanations, facts, research assistance, and questions requiring judgment about information.
- Writing: drafting, editing, translation, summaries, personal communication, and related tasks.
Together, these categories accounted for roughly 77–78% of messages. Practical guidance was the largest category at approximately 29% during the paper’s reported period.
The writing result is more nuanced than “people ask AI to write everything for them.” About two-thirds of writing messages involved modifying text supplied by the user, such as editing, critiquing, summarizing, translating, or improving an existing draft. In personal finance, that might mean making a dispute letter clearer, explaining a lease clause in plain English, or organizing questions for a tax professional—not automatically producing a trustworthy financial decision.
People ask for help more often than they delegate a complete task
OpenAI grouped messages by apparent objective:
- Asking — about 49%: requests for information, guidance, advice, or help with judgment.
- Doing — about 40%: requests for an output usable in a process, such as a draft, plan, or program.
- Expressing — about 11%: reflection, exploration, play, or other interactions outside information-seeking and task completion.
These are message shares, not shares of users or fully automated work. The categories can also overlap in practice. Someone might ask ChatGPT to compare two insurance options and then draft an email to an employer; that conversation contains both advice and output creation.
The broader implication is that ChatGPT often acts as an adviser or decision-support system. It may help someone identify options, understand unfamiliar terms, or prepare for a conversation while leaving the final judgment to a person. That is particularly relevant to money decisions, where a fluent answer is not a substitute for checking rates, fees, eligibility rules, deadlines, or professional advice.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Writing is more common than coding in the analyzed work use
Writing accounted for about 42% of work-related messages. Computer programming represented approximately 4.2% of all messages.
Rank #3
This does not show that coding is unimportant to the AI economy. The study excluded enterprise and API traffic, integrated development environments, and dedicated coding tools. OpenAI’s later Signals reporting also notes that its workplace analysis excluded Codex usage. Coding can therefore be concentrated in products and accounts outside this consumer sample.
The narrower conclusion is that, among the consumer ChatGPT messages OpenAI analyzed, writing was a much more common work-related use than coding.
The “AI companion” narrative is overstated—but not meaningless
“Relationships and Personal Reflection” represented about 1.9% of classified messages, with games and role-play also forming small categories. That does not support the claim that ChatGPT was mainly used as an emotional companion.
Recommended Free Tools
It also does not mean emotional use is unimportant. A small share of message volume could still involve many people, and broad categories may miss gradual or informal emotional reliance. The defensible conclusion is limited: relational and reflective uses existed, but they were a small minority of message volume in this dataset.
Who was using ChatGPT?
Age
Among users who reported their age, approximately 46% of messages from adults came from people aged 18–25. Work-related use generally increased with age, although users aged 66 and older were an exception: about 16% of their messages were classified as work-related. Among users under 26, about 23% of messages were work-related.
Again, these are message shares among users who reported age. Younger users may send more messages per person, so the figures are not a population census.
Gender is an estimate based on names
OpenAI did not collect gender identity for this analysis. It classified first names as typically masculine or typically feminine and excluded ambiguous or unclassifiable names.
OpenAI’s summary says the share associated with typically feminine names rose from 37% in January 2024 to 52% by July 2025 among names placed into those two groups. That is an adoption signal, not proof that exactly 52% of all users were women. Name-based classification cannot reliably capture nonbinary identity, cultural variation, or every person’s gender.
Use grew quickly in lower-income countries
OpenAI reported that, by May 2025, adoption growth in the lowest-income countries was more than four times the growth rate in the highest-income countries. This refers to relative growth, not necessarily the largest absolute user population.
That global expansion matters for personal finance because AI access may increasingly affect how people learn about banking, employment, education, benefits, and small-business administration. It also raises familiar risks: inaccurate advice, language and cultural bias, privacy exposure, and unequal access to reliable internet or paid features.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the report says about economic value
The paper argues that ChatGPT can create value through direct task assistance—such as drafting, editing, planning, translation, and programming—and through decision support, including gathering information, comparing options, and reasoning through problems.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
That second function is easy to overlook. A person may not hand over an entire job or financial decision to AI. Instead, ChatGPT may help them ask better questions, understand a document, or prepare a first draft. The benefit may be time saved, reduced confusion, or greater confidence rather than a measurable increase in paid output.
Best Value
The paper cites a separate estimate that U.S. users would need to be paid approximately $98 per month to give up generative AI, implying at least $97 billion per year in consumer surplus. This is an outside estimate cited by the paper, not a direct result of OpenAI’s usage classification.
It is important to separate three claims:
- Observed usage: what people did in the analyzed sample.
- Inferred value: what those activities may be worth to users.
- Proven productivity gains: measured improvements in real-world output or performance.
The study documents the first far more directly than it proves the third.
What the study does—and does not—prove
| The data supports | The data does not establish |
|---|---|
| Consumer use is broader than workplace productivity. | That 70% of users use ChatGPT for personal tasks. |
| Advice, information, and writing dominate message volume. | That ChatGPT has replaced search engines. |
| Writing is more common than coding in the analyzed consumer work sample. | That coding is only 4.2% of all AI usage. |
| ChatGPT often supports judgment and information-seeking. | That users followed the advice or made better decisions. |
| The name-based gender estimate broadened over time. | That women made up exactly 52% of all users. |
| Consumer usage expanded through July 2025. | That the findings represent enterprise, API, logged-out, or other AI usage. |
What changed after the report?
OpenAI later published 2026 Signals updates, but those releases use newer samples and different measures. They should be treated as subsequent context rather than merged into the 2025 report’s statistics.
Free tools Windows power users keep installed
One-click scans. No signup required.
OpenAI’s June 2026 Signals update said that, in an analyzed cohort, users six months after signup sent about 50% more messages per day and had tried twice as many distinct capabilities. It also reported broader international adoption. The 2026 Q1 update said non-English users represented more than half of active users in its June 2026 analysis and described workplace use continuing to shift toward content creation, health-related documentation, and information retrieval.
Those figures suggest deeper and more global use, but they do not revise the original 70/30 work split or any other 2025 statistic. Consult OpenAI’s 2026 adoption update and its 2026 Q1 Signals report separately.
What this means for your money
For personal-finance readers, the report’s most useful lesson is that AI is often being used as a first-pass adviser. That can be valuable for organizing information, comparing questions to ask, drafting correspondence, or learning basic concepts.
Use it cautiously for high-stakes decisions. Verify interest rates, fees, tax rules, benefit eligibility, investment claims, insurance exclusions, and legal deadlines against current primary sources. Do not paste account numbers, passwords, Social Security numbers, full tax returns, or other sensitive information into a chatbot. A polished answer can still be outdated, incomplete, or wrong.
The report also warns against assuming that workplace AI adoption can be measured from consumer chat alone. Employers evaluating AI should examine enterprise, API, security, governance, and data-handling requirements rather than relying on consumer usage statistics.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
