AI in the service of writing: the advantages of artificial intelligence in content production

Article written in May 2023 and revised in October 2026.

When I wrote the first version of this article, in the spring of 2023, tools that could produce a text from a request in natural language were a novelty, and it came naturally to list their advantages. Three years later they have become part of the daily work of anyone who writes in a company: the marketing department, the salesperson preparing a proposal, the professional sending a client bulletin. The argument I made then still seems valid to me. Artificial intelligence is a real help to anyone who has to produce content. But the list needs rereading in light of the experience gained since, because some advantages have held up well, others were overly generous promises, and there were a couple of important issues I did not mention at all.

The advantages that have held up

The first is time, and it remains the most obvious. An outline, a first draft, three versions of a headline, the summary of a long document: jobs that used to take an hour now take a few minutes. In an experiment with 453 college-educated professionals working on writing tasks typical of their occupations, those who could use ChatGPT took 40% less time on average, and their texts were judged to be better (Noy and Zhang, Science, 2023). Anyone who writes for a living knows that the hardest part is often getting started, and having a draft to correct, even a mediocre one, is very different from facing a blank page. The time saved does not simply vanish, though: much of it shifts from drafting to rereading, and it is wise to budget for that from the start.

The second concerns language. Language models write in grammatically correct Italian, spot typos, repetitions and inconsistencies, and can rephrase a technical passage in words that someone outside the field can understand. For a company that has to explain a complex product to non-specialist customers, or for a professional firm that has to translate a regulation into practical guidance, that is a valuable service.

The third is adaptation. The same content can be rewritten for different audiences, in different registers and at different lengths, and a model that is given the company’s guidelines and a few examples can keep a consistent tone from one text to the next. Catalog entries, recurring customer support replies, versions of the same announcement for different channels: these are repetitive jobs where the help is real and the risk is limited.

The fourth is having a sounding board. Used as a conversation partner rather than as an author, a model suggests angles you had not thought of, raises objections and points out where an argument does not hold together. In 2023 I spoke of “enhanced creativity”; today I would say, with less fanfare, that it is useful to have someone to show a draft to at any hour, knowing that the decisions remain with the person writing.

The promises that need scaling back

I wrote at the time that artificial intelligence can help gather information from reliable sources, and that it can generate articles at scale without compromising quality. Those are the two statements I would not write the same way today.

The first overlooked the most serious flaw of these systems: they produce true sentences and false sentences with the same confidence. The phenomenon now has a name even in official technical documents. The generative artificial intelligence risk profile published in 2024 by NIST, the US National Institute of Standards and Technology, calls it “confabulation” and describes it as the tendency to generate and confidently present erroneous or false content, including reasoning and citations constructed to justify the answer. A percentage, a date, an author’s name, a reference to a court ruling or a study can be made up out of thin air, and nothing in the form of the text gives it away. For anyone writing in a company the consequence is practical: every figure and every source that appears in a generated draft has to be opened and checked before publication.

The second promise, scale, turned out to be half true. Producing a hundred drafts in an afternoon is possible. Publishing them without anyone having read them is another matter, and the example I gave back then, a news site updated automatically with no newsroom, is precisely the case in which errors reach the public. Then there is the effect on visibility. Google’s spam policies classify as abuse the generation of many pages for the primary purpose of manipulating search rankings, and among the examples they explicitly cite using generative AI tools to generate many pages without adding value for users. The stated criterion is usefulness to the reader, however the text was produced, and quantity alone does not guarantee it.

Whoever signs is accountable

A company article carries a name: the author’s, or the company’s. If it contains a wrong figure, the reader does not blame the tool, and rightly so. Responsibility for what is published rests entirely with whoever publishes it, and it is a good idea for a company to be clear about who rereads, who verifies and who gives the go-ahead, especially when the texts deal with prices, deadlines, regulations or health.

Then there is the question of voice. Generated texts tend toward an average prose, correct and interchangeable, and people who read a lot have learned to recognize it within a few lines. An experiment on short-story writing points the same way: stories written with ideas suggested by a generative AI were judged to be better written, but turned out to be more similar to one another than those written without help (Doshi and Hauser, Science Advances, 2024). A company blog makes sense if it says something a competitor could not say in the same words: a case handled firsthand, a mistake made, an opinion that takes a stand. That is material no model possesses, and it has to be put in by the person writing. This is why the division of labor that seems most sensible to me is one in which the person brings the experience, the argument and the fact-checking, and hands the tool what it does best: organizing, rephrasing, shortening, correcting.

The European transparency rules

In 2023 the European regulation on artificial intelligence was still under discussion. Today it is in force, and its transparency obligations, set out in Article 50, have been applicable since August 2, 2026. For anyone producing content there are three points to know.

• Providers of systems generating synthetic audio, image, video or text content must ensure that the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. The obligation does not apply to systems that perform an assistive function for standard editing or that do not substantially alter the input provided by the user.
• Deployers of a system that generates or manipulates image, audio or video content that deceptively reproduces real people or events (so-called deep fakes) must disclose that the content has been artificially generated or manipulated.
• Anyone who publishes a text generated or manipulated by artificial intelligence with the purpose of informing the public on matters of public interest must disclose it. The obligation does not apply where the content has undergone a process of human review or editorial control and a natural or legal person holds editorial responsibility for the publication.

The last point is the one that bears most directly on people who write. The law draws the line where common sense does: a text that has been reread and verified, and for which someone takes responsibility, is treated differently from a text that is generated and published as is. What remains to be worked out, case by case, is what falls under “matters of public interest.” The European Commission is supporting implementation with guidelines and a voluntary code of practice, and in doubtful situations it makes sense to start there or with legal advice.

In conclusion

Three years on, I stand by the substance of what I wrote: for anyone who writes in a company, artificial intelligence saves time, improves form and helps you think. I am correcting the proportions. The full benefit comes only if fact-checking and the final say stay with a person who knows the subject and puts their own name under the text. Used to publish without reading, it produces texts that no one has really written, and sooner or later readers notice.