Writing
Writing took 40% less time, and quality rose 18%
In an experiment with 453 people doing writing tasks from their own occupations, the group using ChatGPT took 40% less time and produced work of 18% higher quality. They also enjoyed the tasks more.
- Time taken
- −40%
- Quality of the work
- +18%
Speed and quality usually pull against each other. Rush and the work gets rough; take care and it takes longer. In this experiment, both improved at once.
What the study found
The experiment was published in the journal Science in 2023. 453 college-educated professionals took part, working on writing tasks designed for their occupations, such as marketing, HR and data analysis. The tasks were close to real work: press releases, short reports, and emails that needed tact.
Half the participants, chosen at random, were given access to ChatGPT. In this study, the group using ChatGPT took 40% less time on average, and the quality of their work rose 18%. Quality was graded by experienced evaluators from the same occupations.
People who used ChatGPT also said they enjoyed the tasks more. Note that this satisfaction was about the experiment’s tasks, not their jobs as a whole.
Another result: the gap between people narrowed. Those who started out less skilled gained more from ChatGPT.
After the experiment, both worry and excitement about AI rose for a while. And people who used ChatGPT in the experiment were more likely than the others to go on using it in their real jobs.
For small teams and sole proprietors
Small companies and sole proprietors write something every day: notes to go with quotes, replies to enquiries, product descriptions for their online store, social posts, internal rules. All of it matters, but much of it repeats, and it tends to get put off.
If an AI agent prepares the first draft, nobody starts from a blank page. The person in charge checks that it sounds like the company and that the facts are right, then finishes it. People keep the thinking and deciding; the typing gets lighter.
The same goes for professionals who work alone, such as accountants, lawyers and designers. Proposals, reports and client updates get written between the actual work. Less time on drafts means more time for the work itself.
The tasks in this experiment were written for each occupation. Real work depends on things specific to each company, such as product specs and the history with a client. The quality of a draft depends on how accurate the material given to the AI is.
The satisfaction result matters too. At a small company where everyone wears many hats, less repetitive writing means more time for the real work.
For mid-sized and larger companies
The bigger the company, the more work runs on writing: reports, minutes, handover notes, procedures, customer notices. Each improvement is small, but with many people involved, it adds up.
More consistent quality matters too. In this study, people who started out less skilled gained more, and the gap between people narrowed. A shared AI agent that knows the company’s terms and style rules makes it easier to keep writing at one standard across departments.
More writing also means more checking. A draft that comes with its source material lets the reviewer focus on what matters.
The study also says something about training. People who tried ChatGPT once in the experiment were more likely to use it in their work afterwards. Giving everyone a real chance to try it looks like the fastest way to spread it across a company.
Still, the satisfaction in this experiment was about the tasks. Whether overall job satisfaction or staff retention change the same way, this study alone can’t say. After launch, keep listening to the people using it and adjust how it is used.
How we use this
Drafts are made inside the tools you already use. Our agents connect to everyday tools such as Gmail, freee and Shopify. They draft notes for quotes and invoices, replies to enquiries, product descriptions and reports from your company’s data. Information that rarely changes, such as style rules and terms of use, is always at hand through CAG, so every draft follows the same rules.
A person checks before anything is sent. Drafts arrive waiting for review, with the source material attached where it matters. Important messages pass through an approval gate (a person always approves important actions).
Each role has its own AI agent. Orion (marketing and sales) handles product descriptions and promotional copy, and Juno (accounting) handles billing messages. Vega (research) writes up research, and Lyra (design) takes care of how documents look. Every agent writes from the same company knowledge.
Team training builds the habit. We show your team how to ask the AI, how to check what it writes, and how to feed corrections back in, so the effect doesn’t fade after the first weeks.
We measure the effect. With the results dashboard (time saved, cost per task and accuracy, checked weekly), you see the change in your own company, not just in an experiment.