Workplace writing is where most people first seriously try AI tools — and also where many first run into trouble. The appeal is obvious: you have an email to write, a report to draft, a set of meeting notes to turn into action items. You ask ChatGPT or Claude for help, and something comes back immediately. It looks plausible. Sometimes it's genuinely useful. Sometimes it sounds nothing like you and goes straight in the bin.
After spending considerable time testing these tools on actual work tasks — not toy examples, but the kinds of things that show up in a real working day — a clearer picture has emerged of when AI writing assistance is worth the effort and when it quietly makes things worse. The answer is more nuanced than either "AI is going to write everything for you" or "AI writing is fake and hollow."
The Tasks Where AI Genuinely Speeds Things Up
There are specific writing tasks where AI assistance is consistently useful, and they share a common characteristic: the structure of the output is fairly predictable, but the actual drafting is tedious or slow.
Meeting summaries and action item lists are the clearest example. If you paste a rough set of notes into Claude or ChatGPT and ask it to pull out the key decisions and next steps, the result is usually solid on the first attempt. The AI is not making judgment calls — it is reorganising information you already have. You still need to review it, because AI tools can occasionally merge two separate action items or attribute something to the wrong person. But you're reviewing a draft rather than starting from scratch, and that saves real time.
Routine professional emails follow the same pattern. Acknowledging a request, confirming details, following up on a delayed item, declining something politely — these are tasks where most of the mental work is knowing what you want to say, not how to say it. AI is good at the "how to say it" part. In my own work, I've found that asking Claude to draft a polite but firm follow-up email, then editing it down to match my actual tone, takes about a third of the time it would take to write from scratch.
The same applies to first drafts of structured documents — project updates, status reports, proposal outlines. If you give the AI a clear brief (what the document is for, who reads it, the key points to include), it will produce something you can edit into shape. That first draft, even if imperfect, is valuable because the blank page is gone.
Where Things Start to Go Wrong
The problems tend to cluster around two patterns: asking AI to write things that require your specific judgment or context, and using AI output without editing it to sound like you.
The judgment problem shows up most visibly with sensitive communications. If you're writing to a colleague about a performance issue, or to a client who is frustrated about something, or to your manager about a situation that has history and nuance — AI doesn't have that history. It will produce something that sounds professionally reasonable in a generic sense, but misses the specific register the situation requires. You know things about the relationship, the stakes, and the right tone that no AI tool can infer from a brief prompt. Using an AI draft in these situations often produces something technically fine but somehow off — and the recipient usually notices.
The voice problem is subtler but just as real. AI writing tools have a characteristic style: thorough, well-organized, slightly formal, occasionally fond of phrases like "it is worth noting" or "this underscores the importance of." If you send that style to people who know how you write, they will notice something is different. That's not a catastrophe, but it can create a low-level sense that your communications have become impersonal — and over time, that erodes something.
The goal isn't to have AI write for you. It's to have AI do the tedious parts while you do the parts that require your actual judgment and voice.
How to Use AI Writing Help Without Losing Your Voice
The most effective approach I've found is to treat AI as a drafting assistant rather than a ghostwriter. The distinction matters in practice.
When you ask AI to write something from scratch, you get AI's version of what that thing should sound like. When you ask AI to help you improve something you've already sketched — even roughly — you get something that starts from your thinking. The output is easier to edit and sounds more like you.
A practical workflow: write two or three rough sentences capturing what you actually want to say. They don't need to be good. Then paste them into ChatGPT or Claude with a prompt like: "Expand this into a professional email, keeping the direct tone." Read what comes back, cut anything that doesn't sound like you, and keep what does. This takes about the same time as writing from scratch, but with less of the staring-at-the-screen phase.
For longer documents, a different approach works better. Use AI to build the structure first. Ask it: "What sections should a project status report include for a technical audience?" Review the outline, adjust it to match what you actually need to cover, then write each section yourself — or ask AI to draft individual sections with specific prompts. The AI-generated outline gives you scaffolding; you fill it with the actual content and judgment.
A Few Specific Things Worth Trying
If you want to start somewhere concrete, these are tasks that tend to work well with minimal frustration:
Editing for clarity: Paste a paragraph you've written and ask Claude or ChatGPT to "make this clearer without changing the meaning." This is low-risk — you wrote the content, AI is just improving the phrasing — and the results are often genuinely better.
Tone adjustment: If you've written something that came out harsher than you intended, paste it and ask to "soften the tone while keeping the substance." This is one of the things AI handles well, and it's useful for the kind of email you write when you're frustrated and know you shouldn't send it as-is.
Summarising long documents: Paste a long report or email chain into Claude (which handles long documents well) and ask for a three-sentence summary. Claude tends to be more accurate on this than ChatGPT in my experience, though both are useful. Always verify the summary against the source before acting on it — AI summaries occasionally miss key nuance.
Generating options: If you're stuck on how to phrase something sensitive, ask AI for three different ways to say it — formal, neutral, and direct. You're not going to use any of them verbatim, but seeing three options often unsticks your own thinking.
The Thing That Doesn't Change
AI writing tools are genuinely useful for the mechanical parts of workplace writing. They are not useful for the parts that require knowing your specific context, your relationships, your professional reputation, and your judgment about what a particular situation calls for. Those parts are still yours.
The risk of over-relying on AI for workplace writing is not that your writing becomes wrong — it's that it becomes generic. Generic writing doesn't usually get you in trouble. It just quietly reduces the impression of you that comes through in your communications. Over time, that matters.
The people I've seen use AI writing tools most effectively treat them like a capable but impersonal assistant: good for getting a draft on paper, not a replacement for the editing pass where you make it actually sound like you. That pass doesn't take long. But skipping it is where most of the problems come from.
Pick one writing task this week that you find tedious — meeting notes, a routine update, an acknowledgement email — and try the workflow above. The goal isn't to automate your writing. It's to spend less time on the parts that don't require your judgment, so you can spend more time on the parts that do.