The interrupting pair
Before
I've been using it for about a month — mostly for client emails — and the difference is noticeable.
After
I've been using it about a month, mostly for client emails. The difference is noticeable.
The em dash tell
Remove the em dashes from AI text.
Replace each em dash with a comma, a period, or a new sentence. That single edit removes the most recognisable AI fingerprint in a draft. RawReply does it on every reply automatically, along with the openers and sign-offs that give the rest of it away.
Examples
The dramatic pivot
Before
It isn't a tooling problem — it's a process problem.
After
It isn't a tooling problem. It's a process problem.
The bold label list
Before
**Setup** — took about ten minutes. **Docs** — sparse but usable. **Support** — surprisingly fast.
After
Setup took about ten minutes. The docs are sparse but usable, and support came back faster than I expected.
Questions people ask
Why does AI writing use so many em dashes?
Because the models were trained on edited prose, where the em dash is a normal tool, and because it lets a sentence add a qualifier without committing to a new one. People write that way occasionally. Models do it two or three times a paragraph, and that density is what readers notice.
Can I just find and replace them?
Mostly, yes, and it is the highest-value thirty seconds you can spend on a draft. Two cautions. Do not replace the en dash in number and date ranges like 5–10 or Tue–Thu, because those are correct. And an em dash is usually a symptom of a long clause-stacked sentence, so if you swap in a comma and the sentence still runs thirty words, split it instead.
What should I use instead?
A comma if the aside is short, a period if it is a second thought, and a colon if you are introducing something. When a sentence seems to need two em dashes, that is almost always two sentences.
Does RawReply strip em dashes automatically?
Yes, on every reply, without you asking. It also removes the openers that follow the same pattern, things like “Certainly”, “Absolutely” and “Great question”, the sign-offs like “Hope this helps” and “Let me know if you have questions”, and the bold-label bullet shape in the third example above. Number ranges are left intact.
Is this an AI detector?
No. Detectors score text and are wrong often enough to be useless. RawReply is a writing tool. The goal is a draft that reads like you wrote it, not a draft that games a classifier.
Are em dashes actually bad writing?
No, and that is the annoying part. They are good punctuation with a reputation problem. In 2026 a reader who sees two of them in a short reply assumes a model wrote it, so the cost is what people infer rather than anything about the sentence.