Free tool
Roast my newsletter reads the latest editions of any public newsletter and checks each one for 29 slop tells: a greeting before the point, stock chatbot words, "it's not X, it's Y", advice that fits any newsletter, and more. Jev, a small decision model from TypeSafe AI, answers yes or no to each tell. The more yeses an edition collects, the more it looks like slop. You get a score from 0 to 100, a meme face and a card to share. It counts habits. It does not know who or what wrote the edition. It is free and needs no signup.
Hire this writer0 to 7
Actually good8 to 15
Beige16 to 24
Slop adjacent25 to 39
Content mill40 to 100Newsletters whose writers chose to be listed, cleanest score first. Each one was roasted with the same tool.
Pictures and subscriber numbers are added by the writers. Quibr does not check the numbers.
Loading the ranking.
You paste a newsletter address. Our server reads that one public feed and takes the newest editions that carry their full text, up to 12. Each edition goes to Jev with a list of yes or no questions, one for each slop tell. Your browser then turns the answers into a score and draws the card.
Jev is a decision model from TypeSafe AI. It does not write anything. It reads a piece of text and answers each typed question with the probability that the answer is yes. A tell counts as found when that probability is 30% or more.
The newest editions with at least 150 words of text, up to 12. Podcast episodes, teasers, short notes and the previews of paid editions are skipped, and the page says how many.
Each edition is checked against 29 tells. Jev answers 28 of them. The last one, an em dash habit, is a plain count, because a model cannot count characters.
An edition's score is the share of the tells found in it. The newsletter's score is the average over its editions.
The score picks one of five faces. The roast lines are written from the real counts, such as "7 of 12 editions open by clearing their throat before the point."
Only the start and the end of a long edition are sent to Jev, because the opening and the ending carry several of the tells. At least 5 readable editions are needed. With fewer, the tool says so and shows no number.
A tell is a habit that shows up often in text written by chatbots, and also in text written by people in a hurry. Wikipedia's field guide to signs of AI writing lists many of them, and says that not all text featuring these signs is AI-generated. I chose 29 and sorted them into six families.
| Family | Tells | Number |
|---|---|---|
| Openings and endings | Throat-clearing opening, Announces what is coming, Summing-up ending, Begs for a reply, Poster-slogan ending, Unrequested pep talk | 6 |
| Stock words and phrases | Corporate buzzwords, Chatbot vocabulary, Worth-noting filler, Sweeping claim about the modern world, Abstract noun soup, Performed honesty | 6 |
| Structure tics | Not X, but Y, Everything in threes, Here's-the-thing reveal, One-line paragraphs, Bold-label bullets, Generic headings, Em dash habit | 7 |
| Not much in them | No specifics, Generic advice, Nothing personal, No position, Fits any newsletter, 'Studies show' with no study | 6 |
| Tone | Breathless hype, No rough edges, Fake intimacy | 3 |
| Titles | Template title | 1 |
Every tell is a question about one edition, and yes always means the tell is there. Here is the full list, exactly as it is put to Jev.
| Tell | The question |
|---|---|
| Throat-clearing opening | Does the text open with an apology, a remark about the week or the time of year, or a long preamble before it gets to its first real point? |
| Announces what is coming | Does the text announce what is coming with phrases such as 'in this issue', 'let's dive in', 'let's unpack this' or 'here's what you need to know'? |
| Summing-up ending | Does the text end by summarising itself, with a line such as 'in conclusion', 'the bottom line', 'the key takeaway' or 'remember'? |
| Begs for a reply | Does the text end by asking the reader to reply, comment, share or answer a question? |
| Corporate buzzwords | Does the text lean on corporate buzzwords such as 'leverage', 'unlock', 'elevate', 'seamless', 'robust', 'game-changer' or 'ecosystem'? |
| Chatbot vocabulary | Does the text use words that chatbots overuse, such as 'delve', 'tapestry', 'testament', 'realm', 'pivotal', 'crucial' or 'navigate the landscape'? |
| Worth-noting filler | Does the text use filler such as 'it's important to note', 'it's worth noting' or 'it goes without saying'? |
| Sweeping claim about the modern world | Does the text make sweeping statements about the modern world, such as 'in today's fast-paced world' or 'in an ever-changing landscape'? |
| Not X, but Y | Does the text contain a flat denial followed by a correction, such as 'it's not X, it's Y' or 'this isn't about X, it's about Y', at least twice? |
| Everything in threes | Does the text keep grouping things in threes, such as three adjectives or three parallel phrases in a row, in paragraph after paragraph? |
| Here's-the-thing reveal | Does the text use dramatic lead-ins such as 'here's the thing', 'here's the kicker', 'let that sink in' or 'think about that'? |
| One-line paragraphs | Is the text mostly one-sentence paragraphs written for dramatic effect? |
| Bold-label bullets | Does the text use lists where each item starts with a bold label followed by a colon? |
| Generic headings | Does the text use generic headings that would fit any newsletter, such as 'The problem', 'The solution', 'Why it matters' or 'Key takeaways'? |
| Em dash habit | Does the text use four or more em dashes for every 1,000 words? (counted in code, not asked) |
| No specifics | Does the text avoid specific names, numbers, dates and real examples, staying general throughout? |
| Generic advice | Is most of the text advice that could appear in any newsletter on the topic, such as 'be consistent' or 'know your audience'? |
| Nothing personal | Does the text read as if written by nobody in particular, with no 'I', no experience, no opinion and no detail that only its author could know? |
| No position | Does the text avoid taking a clear position, presenting every side as equally valid? |
| Fits any newsletter | Could this text be pasted into a different newsletter on the same subject without anyone noticing? |
| Breathless hype | Is the tone breathlessly enthusiastic, with superlatives and exclamation marks about ordinary things? |
| No rough edges | Is every sentence smooth and similar in length, with no quirks, tangents, jokes or rough edges that suggest a real person? |
| Fake intimacy | Does the text claim a shared feeling, such as 'we all know that feeling' or 'I'm so excited to share this', without anything specific behind it? |
| Poster-slogan ending | Does the text end on a short, quotable line that sounds like a motto or a poster slogan? |
| Unrequested pep talk | Does the text give the reader a pep talk, such as 'you've got this', 'you deserve better' or 'start small and stay consistent'? |
| Abstract noun soup | Is the text full of abstract nouns such as framework, landscape, journey, ecosystem, mindset or strategy, with few concrete things you could picture? |
| Performed honesty | Does the text perform honesty with phrases such as 'let's be honest', 'I'll be real', 'the truth is' or 'full disclosure', without anything risky following? |
| 'Studies show' with no study | Does the text rely on 'studies show', 'experts say' or 'research suggests' without naming a study, an expert or a number? |
| Template title | Is the title a template such as 'X ways to...', 'The ultimate guide to...', 'Why X matters more than ever' or 'The secret to...'? |
Take one edition. If Jev says yes to 6 of the 29 tells, its score is 21. The newsletter's score is the average of its editions' scores, rounded. A low score means few tells. A high score means many.
The score picks one of five bands, and each band has its own face: Hire this writer (0 to 7), Actually good (8 to 15), Beige (16 to 24), Slop adjacent (25 to 39), Content mill (40 to 100).
Hire this writer0 to 7Actually good8 to 15Beige16 to 24Slop adjacent25 to 39Content mill40 to 100The band limits are my own, and I may move them as more newsletters are roasted. The result also says how many answers Jev was unsure about, because a probability near one half means it could not decide.
No. The score counts habits. It does not know who or what wrote an edition, and it does not try to find out. Plenty of people write this way, and anyone can ask a chatbot to avoid every tell.
Tools that claim to know are shaky. A Stanford study found that GPT detectors consistently misclassified writing by non-native English speakers as AI-generated (Liang and others, 2023). Wikipedia's own guide tells editors not to rely solely on AI detection tools (Wikipedia).
So read the result as a roast, not a ruling. When a tell is found in your edition, look at the sentence yourself and decide whether it is a habit worth keeping.
I tried it on 15 real newsletters and blogs, and on chatbot-style editions that I wrote for the test. The real newsletters scored between 6% and 27%, with a middle value of 18%. The chatbot-style sets scored between 37% and 90%. I tuned the tells and the bar on part of that sample, then checked five more real newsletters I had not used. They scored between 11% and 18%. Read the gap as a sign and not a guarantee.
The score has limits. A careful chatbot text that avoids the tells can score about the same as a human one: text I wrote to imitate a well prompted, humanized chatbot scored 25%, the same as several marketing blogs. Jev also gives slightly different answers from run to run. In my repeat runs the scores did not move, but a score right on the edge of a band can land on either side.
This is a roaster, so the bar is set low on purpose: it is better to over roast than to under roast. Most real newsletters land in the middle bands, and the drooling faces are kept for text that collects a lot of tells. Use the result as a prompt to look at your own habits, not as a ruling on your writing.
Any public newsletter with a feed. For most platforms you paste the address of the newsletter and the tool finds the feed. beehiiv is the exception.
| Platform | What to paste |
|---|---|
| Substack | Your address, such as name.substack.com, or your own domain |
| beehiiv | The RSS address from Settings, then RSS, then New RSS Feed. The page of a beehiiv newsletter does not point to it. |
| Ghost | Your site address. A site in Private Site Mode has no public feed. |
| Buttondown | The address of your archive |
| WordPress | Your site address |
The feed has to carry the whole text of each edition. Many feeds carry only a short teaser, and an edition with only a teaser cannot be judged. Those are skipped and counted. beehiiv's help page, Ghost's and Buttondown's say where each feed lives. I read them on 18 October 2026.
Roasts are English only for now, because Jev works best in English and the tells are English phrases.
The address you type is read once and not stored, unless you put the newsletter in the ranking (see below). The text of the editions is sent to TypeSafe AI, the company behind Jev, to be judged, and Quibr does not store it. TypeSafe's privacy policy says it will not train or fine tune models on that input, and that its services are hosted in the United States.
Two small things are kept to stop abuse and to see whether the tool works. One is a scrambled version of your IP address with a request count, deleted within 24 hours. The other is a daily count of how many roasts ended well or badly, with no address in it. The site also counts clicks on the buttons, with nothing from your roast attached.
The one exception is the ranking, and only if you choose it. After a roast, you can press “Put me in the ranking”. Then the newsletter’s name, its public address, its score, the number of editions read and the date are kept, and they show on the public ranking straight away.
Also kept: the platform it runs on, how often it publishes and how long its editions are (worked out from what was read), and its logo, which is copied from its feed. A picture and a subscriber number are kept only if you add them, and the number is shown as yours and is not checked.
Pictures, including the logo, are checked by OpenAI’s moderation service before they are shown. You get a removal code once. Taking your newsletter out with that code deletes the entry and its pictures at once.
Yes, if it is public. The tool reads the same public feed anyone can open. The score is a joke from a model that reads a few editions, so please be kind about it. If you write a newsletter and do not want it roasted, email vicente@quibr.so and I will block it.
The result names it and lists the tells found. Read it with those tells in mind.
A high score is usually a few habits repeated, not thirty separate faults. Cut the one that shows up most.
Download it and add it to a post on X or LinkedIn. The caption is written for you, and you can change it.
To plan the next stretch, the growth planner turns your subscriber goal into a weekly pace and a 30 day checklist. My guide to growing a newsletter fast explains why promoting each edition on social media is usually the quickest free method, and there are more guides on the Quibr blog.
Yes. It is free, needs no signup and has no limit on how many newsletters you try, beyond a fair-use cap of a few roasts in an hour.
No. It counts slop tells, which are habits found in text written by chatbots and by people. It cannot tell you who or what wrote an edition, and a score is not proof either way.
Jev is a decision model from TypeSafe AI. It does not write text. It answers a typed question about a piece of text, such as a yes or no, with a probability. Here it answers one yes or no question for each tell.
Up to 12 of the newest editions that carry at least 150 words of text. It needs at least 5. The result says how many were read and how many were skipped.
Fewer than 5 editions in the feed carry enough text. Feeds that show only a teaser, short notes, podcast episodes and the previews of paid editions are skipped, so a newsletter that is mostly paid or mostly teasers cannot be roasted.
Not yet. Jev works best in English and the tells are English phrases, so a newsletter in another language gets a message instead of a score.
The score counts habits, not quality. A link roundup has little first-hand experience in it, and a tutorial can use bold bullets for good reason. Jev can also be wrong. Look at the tells it found and decide for yourself.
Yes, if it is public. The score is a joke from a model that reads a few editions. A writer who does not want to be roasted can email vicente@quibr.so and the address is blocked.
Yes. Quibr reads each new edition from your RSS feed, writes posts in your voice and publishes them to X, LinkedIn, Threads and Facebook. It costs $29 a month, and the first seven days are free.
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