Claims about AI and productivity tend to be dramatic. Writers are told they can produce ten times the content, teams are told they can cut editing time in half, and conference slides promise a future in which drafting costs nothing. Careful measurement tells a more modest, and more useful, story.
A 2025 working paper from the National Bureau of Economic Research by Anders Humlum and Emilie Vestergaard studied how workers in Denmark used chatbots and what changed in their working hours. The average measured time savings were around 2.8 percent of work hours, roughly one hour a week. That is a real gain. It is also far smaller than the marketing headlines suggest.
What 2.8 Percent Really Means
One hour a week is not nothing. Across a year, it adds up to more than fifty hours, which is more than a full work week. For a small team, that is time that can go toward research, client calls or deeper editing.
But the figure also tells a writer what not to expect. The tools do not turn a slow writer into a fast one overnight, and they do not eliminate the need for review. The savings come mostly from specific tasks, such as summarizing, reformatting and producing first drafts of routine text, not from the creative and analytical core of writing.
Why Savings Are Smaller Than Expected
Review takes time
A draft that arrives in thirty seconds still needs checking. Facts, names, claims and tone all need a human eye. For many tasks, review absorbs a large share of the minutes saved on drafting.
Rewriting takes time
Raw model output often needs structural adjustment. Writers describe cutting generic openings, replacing stock phrases and breaking up evenly sized paragraphs. If that work is done entirely by hand, it eats into the gain.
Not every task benefits
Original reporting, expert analysis and sensitive client communications resist automation. A tool that helps with a newsletter blurb is little use in a negotiation memo.
Closing the Gap Between Draft and Finished
The largest controllable source of lost time is the distance between a model’s draft and a finished piece. Narrowing that distance is where refinement tools can help. A tool that adjusts rhythm, phrasing and structure reduces the manual editing needed to make a draft sound natural.
Writers looking to shorten that step often use a tool designed to humanize AI text. Phrasly AI Humanizer provides three modes on a free account, Easy, Medium and Aggressive, so the strength of the change can fit the draft. Paid plans extend the limit to 5,000 words per request, which suits long documents. The tool does not remove the need for a human read. It reduces the amount of line editing that read requires.
Looking at the Evidence in One Place
Readers who want a broader look at how refined AI text is evaluated can consult this 2026 AI humanization study guide, which collects the material in a single reference.
Reading primary sources, including the NBER paper above, remains the best way to separate measured results from marketing. A sentence beginning “studies show” is only as strong as the study behind it, and a good writer checks.
Practical Ways to Capture the Gain
-
Apply the tools to repeatable tasks first: summaries, reformatting, outlines, routine updates.
-
Keep a short log of how long tasks take with and without assistance, for two weeks.
-
Standardize prompts and settings that work, so improvements carry across projects.
-
Protect time for the human steps: fact-checking, voice and final reading.
-
Review the log monthly and drop uses that do not pay for themselves.
A log is more persuasive than intuition. People often feel faster with a tool even when measured results are flat, and the reverse is also true. Numbers settle the question.
Put plainly, the honest message of the research is neither hype nor dismissal. It is an invitation to measure, adjust and keep the parts that work.
Choosing Where to Look for Gains
Teams often start with the most visible task, usually blog drafting, and assume that is where the biggest savings sit. A better approach is to list every recurring writing task, estimate its weekly time cost, and rank them. Meeting summaries, status updates, product descriptions and internal documentation often turn out to be the heaviest and the most automatable. Moving AI help to those tasks frees more hours than speeding up a single article.
It is also worth noting where a tool adds friction. Writers who spend more time correcting a model than they would have spent drafting alone should stop using the tool for that task. The goal is not to maximize usage. It is to maximize finished, accurate work per week.
Setting Honest Targets
A target of ten percent faster, set per task and checked after a month, is more useful than a vague goal of working smarter. Small, specific targets can be tested, and a test either shows a gain or does not. When a gain appears, the team can document the prompt, the setting and the review steps that produced it, so the next person starts from a working method instead of reinventing it.
This kind of discipline is unglamorous, but it is how measured gains turn into habits. The NBER finding is a useful anchor: if a large study of real workers found savings near one hour a week, a team that finds two or three hours a week on its own tasks has done well and can say so credibly.
Quality Gains Count Too
Time saved is not the only measure. A tool that helps a writer produce clearer, more varied prose, or that catches a stiff paragraph before a client sees it, improves quality even when the clock shows no change. For some teams, the more valuable result is consistency across a large body of content, not speed.
What Teams Often Overlook
Onboarding costs are easy to ignore. Staff need time to learn prompts, settings and review habits, and the first weeks often show lower output than before. Teams that expect an immediate jump tend to abandon useful tools too early. A realistic plan allots a month of adjustment and judges results after that.
Another overlooked factor is variance between people. In most teams, a few members gain a lot from the tools and a few gain little. Sharing the methods of the first group, through short demos or written notes, usually raises the average faster than any mandate does. Over a quarter, that sharing can matter more than any single tool choice.
Lastly, remember that the measured average hides a wide range. Some tasks and some workers will show zero gain, while others show several hours a week. The sensible response is to look at the distribution in your own team, not to assume the average applies to everyone.
Conclusion
The evidence supports a balanced view. AI writing tools produce real but modest time savings on average, with larger gains on routine tasks and smaller ones on work that demands judgment. Review and refinement are part of the job, not obstacles to it, and measuring results beats trusting promises. Writers and teams who set realistic expectations, apply the tools where they fit and keep a human in charge of facts and voice will capture the available benefit without being disappointed by the hype.
