AI Proofing Your Technical Writing Career
A key skill to pave your way to an AI-oriented technical writer
You know what I do when I feel insecure about AI taking my job?
I talk to it.
And the answers I get are…
surprisingly reassuring.
Usually, I get a response like this.
No—AI is unlikely to make technical writers obsolete. But it will change the job significantly.
ChatGPT, 14.03.2026
Instead of making it obsolete, ChatGPT claims that the job will evolve away from simpler, repetitive tasks like drafting documentation and into more complex tasks, like:
Understanding complex systems
Designing knowledge systems
Using AI effectively
In this article, I will focus on the latter two aspects by highlighting a method that even expert writers don’t fully grasp.
That method is the semantic structuring of content.
When Does AI Go Delulu?
Have you heard of semantic drift?
Semantic drift often manifests as what we call hallucinations in AI tools. One contributing factor to this phenomenon, especially in retrieval-based systems, is insufficient chunking of text.
Chunking refers to splitting content into retrievable units. These units are often based on tokens, semantics, or structure, and help the AI compile answers. If the original text does not offer a clear structure, chunking gets off track - resulting in semantic drift.
Let’s illustrate with a DITA XML topic:
<topic id="data-sync">
<title>Understanding and Performing Data Synchronization</title>
<body>
<p>Data synchronization ensures that two systems maintain consistent datasets over time.
It is typically achieved through scheduled updates or event-driven triggers.
To begin, click the "Sync Now" button in the dashboard.</p>
<p>Consistency models such as eventual consistency allow for temporary divergence between systems.
Users should navigate to Settings > Advanced Sync before configuring these models.</p>
<p>When synchronization fails, it may be due to network latency or conflicting writes.
Select "Retry Failed Jobs" to resolve most transient issues automatically.</p>
<p>In distributed systems, conflict resolution strategies like last-write-wins are common.
Enter your API key and press "Validate Connection" to ensure proper authentication.</p>
<p>Proper synchronization improves system reliability and user trust.
If errors persist, delete the sync profile and recreate it using the default template.</p>
</body>
</topic>If we now assume that an AI chunks this topic into paragraphs (<p>), the AI may mix instructions with background information. Or provide instructions in an incorrect context.
It may struggle to reliably distinguish between instructions and concepts.
Let’s put this to the test.
The user is under a lot of stress. He needs an answer, and he needs it now.
How do I configure eventual consistency in the system?
The chatbot produces an answer in lightning speed: To configure eventual consistency, go to Settings > Advanced Sync, then click Sync Now to apply the consistency model. If the configuration fails, use Retry Failed Jobs to ensure the system adopts the selected consistency strategy.
Sounds great! Let’s do it!
The user is thrilled to have a plausible answer that quickly, only to realize, after a few clicks, that it doesn’t work. And now he doesn’t know why.
Semantic Structures Reduce Hallucinations
As a respectable technical writer, you must understand the benefits of semantic structuring - now more than ever!
Calling an element <step> instead of <numberedlist> helps writers in two ways:
Ensure consistent use of tags: With layout-oriented tagging (<numberedlist>), you leave it to the writers to use the tag correctly. They need to learn when to use it, which leaves room for interpretation. And room for interpretation leads to inconsistency across different writers.
With semantic tagging (<step>), the name of the tags explicitly tells the writer when to use them. No need to remember rules. Consistency is built in.
In addition to consistency, semantic tagging grows in significance with the rise of LLMs. People aren’t the only ones benefiting from intuitive, semantic tags. LLMs can be designed with much more robust chunking strategies. This allows them to generate more stable responses.
If we go back to our example, you should make sure to structure it with proper meaning assigned to the tags:
<topic id="data-sync">
<title>Data Synchronization</title>
<concept>
<p>Data synchronization ensures that two systems maintain consistent datasets over time...</p>
<p>Consistency models such as eventual consistency allow temporary divergence...</p>
</concept>
<task id="perform-sync">
<title>Run a Manual Synchronization</title>
<steps>
<step>Open the dashboard.</step>
<step>Click "Sync Now".</step>
</steps>
</task>
</topic>Conceptual information is now split from instructions. The way DITA intended and the way chunking methods prefer.
Why Am I Telling You This?
If you are now wondering why I bring up this “old news” and pretend it is the next big thing, it is because I know too many highly trained and experienced technical writers who still sleep on this concept!
The vast majority of technical writers know how to use XML-based authoring tools. They go through the motions, almost blindly. Too blindly, in fact.
They do not understand why this matters so much and end up using the methods inconsistently. Why? Because they don’t consider it valuable. They got used to them and just applied the rules.
However, if you want to transform into a technical writer who is not only proficient in writing content but also designing documentation systems for AI, you need to solidify your know-how in semantic writing techniques.
If you use DITA, make yourself familiar with the actual meaning of the tags. Many just use the tags as they consider them correct and stick with them.
If you ever get the chance to design your own documentation structures, learn semantic structuring, such as Functional Design, and design your tools accordingly.
AI isn’t going anywhere.
Start AI-proofing your technical writing career today. If you only have a few minutes today, read the linked article on Functional Design.
Start building up your foundation or fix the cracks in it.