How-to guide
To turn a newsletter into posts that still sound like you, start from your own words and your own past posts, not a blank prompt. Give the tool 10 to 20 posts you already published, tell it your habits, and cut the habits chatbots add: abstract nouns, pep talks, slogan endings and “studies show” with no study. When I tested these against 162 real newsletter editions, each turned up four to six times more often in chatbot-style text. Read the first posts before anything goes out. I build Quibr, which writes against your own posts.
Disclosure: I am Vicente Évora, the founder of Quibr, which writes posts from newsletters, and of the free Roast my newsletter, whose test data this page uses. The chatbot-style editions in that test were written for the test, not taken from real newsletters, so treat the gaps as a sign, not a measurement.
Because the model fills gaps with the most common way to say things. Ask for “a LinkedIn post about this edition” and you get the average LinkedIn post: a big claim, three tidy points, a motivational last line.
Your edition already has what the average lacks: your examples, your numbers, your opinions and the way you phrase them. A good process keeps those and stops the model adding its own habits on top.
I measured this while calibrating Roast my newsletter. I put 29 habits to 162 editions from 15 real newsletters and to 17 editions written in a chatbot style, and kept the habits that separated them. These four separated them best:
| Habit | What it looks like | The fix |
|---|---|---|
| Abstract noun soup | Framework, landscape, journey, ecosystem, mindset, with nothing you could picture | Name the actual thing: the tool, the number, the person |
| Unrequested pep talk | “You’ve got this.” “Start small and stay consistent.” | End on the idea, not on encouragement |
| Poster-slogan ending | A last line that sounds like a motto | Stop when the point is made |
| “Studies show” with no study | “Research suggests” with no study, expert or number | Name the source, or cut the claim |
Some habits that people call AI tells are just common writing. Sweeping lines about what “most people” do, and short sentences repeated like a chant, each showed up in more than half of the real editions, so I dropped them. You can check your own editions for all 29 habits that stayed with Roast my newsletter.
No. Wikipedia’s guide to the signs of AI writing says so directly.
“Not all text featuring these indicators is AI-generated.”
Tools that claim to detect AI text get it wrong too. A Stanford study by Liang and others found that GPT detectors consistently misclassify writing by non-native English speakers as AI-generated. Stanford HAI reported that they flagged 61.22% of TOEFL essays by non-native students, while being near-perfect on essays by US-born eighth-graders. The point of cutting these habits is not to pass a detector. It is to sound like a person your readers want to hear from.
Pick 10 to 20 posts you published and liked, on the platform you are writing for. Only yours: reposts and other people’s posts teach the wrong voice.
How long your posts run, how you open, whether you use lists and how you mark them, whether you start sentences in lower case.
Give the tool the edition itself, so the ideas, examples and numbers are yours.
Tell the tool to avoid them, and check for them when you read.
Approve the first week of posts by hand. Fix what sounds wrong, and tell the tool why.
For the rest of the routine, read how to promote a newsletter on social media.
Closer than a blank prompt, if it learns from posts you already published and starts from your own edition. Read the first posts before you let any tool publish on its own.
Abstract nouns such as framework, landscape, journey, ecosystem and mindset, used where a concrete thing would do. In my test they showed up in 16% of real newsletter editions and 76% of chatbot-style ones.
No. Heavy em dash use is one of the signs Wikipedia lists, but people use em dashes too. Quibr replaces them in every post it writes, as a house style.
Not reliable enough to judge a writer. A Stanford study found that GPT detectors consistently misclassify non-native English writing as AI-generated.
Ten to twenty of your best posts on that platform is a good start. They should be yours alone, not reposts or replies to others.
Founder of Quibr
Vicente writes Notes to 10K MRR, a weekly build-in-public newsletter on beehiiv, alongside university and a part-time job. Every edition went out on time, and the social posts that were supposed to follow it never did.
He built Quibr to write those posts for him. It started as an idea on 26 November 2024, and the first version shipped in January 2025. Quibr is built and run by one person, and Vicente answers support email himself.
Last updated