Call them documentation specialists, content strategists…even magicians. Technical writers are a special niche of wordsmiths who serve as invaluable links between product developers and varied audiences, such as engineers, sales and marketing staff, and end users. These writers distill complex information into user-friendly guides, instruction manuals, standard operating procedures, and reports.
But with the advent of artificial intelligence, are technical writers a threatened species? If your immediate response is “Yes,” don’t be so sure.
But with the advent of artificial intelligence, are technical writers a threatened species? If your immediate response is “Yes,” don’t be so sure.
What you can be sure about is that AI is transforming many professions—technical writing included. This transformation is resulting in benefits that enhance the work of a technical writer.
Yet, AI-assisted technical writing has its limitations, and many tasks still require human expertise.
The Benefits of AI in Technical Writing
Essentially, AI automation in documentation creates efficiency, enabling technical writers to focus on tasks that demand more creative and nuanced thinking. “I use AI daily to quickly get up to speed on complex topics,” says technical writer Steliana Vassileva.
Vassileva says AI also helps technical writers save time by moving the writing process from first drafts to usable copy more quickly, leaving the writer more time to tackle higher-level work, such as information architecture, building tooling and automation, developer experience design, and documentation strategy.
“Overall, like engineers, technical writers can use AI to cut down on low-level, repetitive tasks and free up time for more meaningful work,” she says.
AI in technical writing is useful in several ways, some of which include:
Research and Summaries
Technical writers often do extensive research, reviewing numerous amounts of data to inform their work. AI can complete this step in seconds vs. the hours it would take a person to do it. AI can create summaries and key takeaways of this data, enabling the writer to easily review and often better understand the information.
First Drafts and Revisions
AI can also save hours of a writer’s time by generating first drafts of technical material. Writers can also use the tool to revise drafts by having the tool suggest different phrasing or word choice, for example.
Task Automation
A certain amount of a technical writer’s work, including the formatting of documents, is repetitive and even dull. AI tools can restructure documentation to include elements, such as subheads, bullet points, and numbered lists. By handing off this “busy’ work to the machine, writers can focus on more of the “meat” of the content.
Quality Control
Any content must be polished and grammatically correct. Technical documents also undergo continuous review to be refreshed with updated data. Each time, the document must be checked for correct grammar, style, and tone. AI tools, such as Grammarly, Hemingway Editor, and QuillBot, can complete these tasks automatically.
Customization and Tone
Sometimes a document is intended for distinct audiences, such as a sales and marketing team or the end user, who will have different levels of understanding and knowledge of the technology. AI can generate versions of the same document tailored to the specific knowledge of the audience.
AI Limitations in Writing
People often believe that technology shouldn’t make mistakes, but it does. Examples of failure in AI technology include everything from self-driving cars that crash to AI tools that have created fake legal briefs and inaccurate health information.
AI, like humans, is far from perfect. “When working with developers and with AI, my general motto is trust but verify,” Vassileva says.
Sometimes, the tool just makes a “mistake,” called a hallucination.
“As an academic who studies technical writing, I’ve seen lots of hallucinations firsthand,” says Guiseppe Getto, Ph.D., an associate professor of information design at Mercer University, Macon, Ga. Getto says the source material provided in an AI answer to a query is not always illegitimate, meaning the source material doesn’t exist.
“AI is really good at doing the grunt work that you don’t want to do, but it needs help. It’s not perfect,” he says.
What Can Go Wrong with AI and Technical Writing
Often, when an AI tool gives incorrect information, the culprit is the data “fed” into it by humans.
Vassileva offers the example of a software engineer at a technology company who tried experimenting with the AI-powered Cursor code editor.
The engineer wanted to create a simple example showing how one of his company’s technologies should be used in a common web application. The engineer used natural-language prompts and connected Cursor to one of the company’s websites so the tool could pull in documentation and generate instructions.
However, while the output looked correct, it was not, Vassileva explains. “To someone unfamiliar with the product, the responses appeared legitimate, but they were not trustworthy,” she says.
The problem was twofold: documentation on the company’s website the Cursor tool was pointed to was incomplete and out of date, and the engineer didn’t realize he was pointing the AI tool to the wrong web property.
“As a result, the AI generated confident but misleading instructions based on stale and insufficient information,” she says. “Therefore, the AI is only as good as the input it receives.”
Several common issues with AI tools include:
- Inaccuracies/hallucinations. Just as humans make mistakes, AI algorithms can also make mistakes by giving information that is incorrect. The tool is not “lying” or intentionally providing false answers. Rather, the AI treats the information as accurate, even when technical limitations or flawed inputs cause the output to be wrong.
- Safety/security. AI bases its output on enormous datasets. If a user inputs proprietary information, that data is absorbed into the AI model, risking a leak of sensitive product information that could harm the company.
- Bias: The output produced by AI is derived from existing datasets developed by humans. The challenge is that humans have inherent but often unintentional bias. For instance, if an AI model was developed or “trained” using data derived from only one group of people, say of a specific ethnicity or age, that AI tool will be biased toward the culture and customs of that specific group.
Why the Human Touch Is Still Essential
Simply put, machines need humans to learn how to do what they do (for now, anyway). These AI models are built using documentation created by people. “AI models learn from and rely on high-quality sources and documentation—API references, code samples, knowledge bases, and product guides—that technical writers design, validate, and maintain,” Vassileva says.
People Are Better Subject Matter Experts
Getto says AI tools may be considered subject matter experts, to a certain degree, but not to the degree that a technical writer is.
Technical writers bring a deeper product understanding, the ability to test features, spot inconsistencies, and connect related systems in ways AI cannot, Vassileva says. Humans also organize content for improved wayfinding, catalog complex concepts, and help to ensure the reliability of documentation over time.
Human Review of AI-Generated Content Is a Must
Even when AI assists with drafting content, human oversight is critical. Humans are still needed to judge the accuracy and usefulness of the output of an AI model. A writer may ask the AI tool to use a certain tone or style, for example. The writer must then use his expertise to critique the outcome.
Writers provide the verification, context, and judgment needed to ensure every piece of content is accurate and aligned with real product behavior, Vassileva says.
Some Industries Require People Over AI
Another important consideration in using AI-assisted technical writing is compliance. In highly regulated industries—such as law, tax, and accounting, and healthcare—AI-generated content may not meet reliability or regulatory standards, requiring technical writers to perform much of the work themselves to ensure accuracy and compliance.
The Best of Both Worlds
If AI tools still need human expertise, the corollary is also true: in today’s environment, top-notch technical writers have AI literacy—they know how to use automation in documentation to enhance their work.
The key for the technical writer is understanding the AI tool’s benefits and limitations, and using personal skills to fill these gaps.
“But because technical writing, again, relies on subject matter expertise, companies are not just getting rid of their technical writers. And I don’t think that they will in the foreseeable future,” he says. “Unless, again, we get through the next innovation curve and AI becomes exactly like the human brain.” “Writers provide the critical thinking that goes on. I mean, nothing has really changed with that.”
