AI writing is no longer just a tool for drafting emails or experimenting with chatbots. A large analysis by researchers from Stanford University, the University of Washington, and Emory University shows that generative AI has become part of professional writing across several public-facing sectors.
The study examined more than 1.5 million texts produced between January 2022 and September 2024. Its core finding is straightforward: AI-assisted writing has moved quickly from novelty to routine support in the areas the researchers studied.
Press releases show the clearest shift
The most visible adoption appeared in press releases. The researchers reviewed over 537,000 press releases from Newswire, PRNewswire, and PRWeb, finding that up to 24 percent of content now comes from generative AI systems or shows significant AI modification.
ChatGPT's launch in November 2022 marked the major turning point in the data. After a 3-4 month adjustment period, AI use rose sharply across the press release platforms.
Newswire reached peak adoption at 24.3 percent in December 2023, then stabilized around 23.8 percent. PRNewswire followed a similar path, peaking at 16.4 percent, while PRWeb showed comparable numbers.
The study also found that technology and business sectors lead in adoption rates. That makes press releases an important signal for understanding how AI writing is entering corporate communications, because these documents are designed for public distribution and often sit between marketing, investor communication, and media outreach.
AI writing is appearing outside corporate messaging
The same pattern is not limited to company announcements. The researchers also identified AI adoption in consumer complaints submitted to the U.S. Consumer Financial Protection Bureau (CFPB), where around 18% of complaints were AI-generated by late 2024.
That finding matters because consumer complaints are different from press releases. They are not polished promotional documents. They are individual submissions, often written by people trying to explain a problem clearly enough for an institution to understand it.
In that setting, AI writing may help users organize a complaint or express a case in more formal language. At the same time, it raises the same concerns the researchers note elsewhere: professional writing may become easier to produce, but it may also become more standardized and less authentic.
Company age affects job ad adoption
Job advertisements showed another important pattern. Small and recently founded companies are leading adoption, with clear differences based on when companies were established.
- Firms founded after 2015 use AI-generated texts for certain job postings at rates between 10% and 15%.
- Companies founded between 2000 and 2015 show lower adoption, at 5% to 10%.
- Companies established before 1980 rarely use AI assistance, with adoption rates below 5%.
This suggests that AI writing is not spreading evenly. Newer firms appear more willing to use generative AI in hiring communications, while older organizations are moving more slowly, at least in the job postings studied.
The pattern is useful because job advertisements combine operational language with public branding. They must describe roles, expectations, and company needs, while also presenting the employer to potential applicants. AI assistance in that area can change both the speed of posting and the tone of recruitment language.
United Nations communications also reflect the trend
The researchers found similar figures in UN press releases, where AI-generated texts accounted for slightly under 14%. Usage was not uniform across teams.
UN teams in Latin America and the Caribbean showed the highest usage at around 20 percent. This points to a broader pattern: AI writing assistance is not confined to businesses or consumers, but is also appearing in institutional communications.
The study frames this growth as potentially democratizing professional writing. In plain terms, generative AI can help more people and organizations produce formal text that fits established professional expectations.
But the researchers also warn about possible drawbacks. If many organizations use similar AI systems to produce similar kinds of documents, the result may be more uniform language, less distinctive voice, and a weaker sense that a text reflects a specific person or institution.
The measured numbers may be too low
The researchers suspect the real level of AI adoption is higher than the study shows. Their analysis may miss heavily human-edited AI content, as well as text from advanced AI models that closely mimic human writing.
The study also did not examine every possible use case for AI writing. Social media content creation, for example, was outside the scope of the analysis.
To estimate AI content, the researchers used a specialized statistical method based on maximum likelihood estimation (MLE). The method compares known human and AI-generated reference texts, then estimates AI content proportions across larger collections of writing.
Rather than labeling each individual document as human or AI-written, the approach looks at statistical patterns in aggregate. According to the researchers, this makes it more robust, accurate, and efficient than traditional detection approaches for large-scale text collections.
The same research team had earlier found increasing AI presence in academic research and scientific publications. Taken together with the newer findings, the picture is clear: AI-assisted writing is becoming part of professional communication across very different fields, from press releases to complaints, job ads, UN communications, and research writing.