Paste the draft, get a 0-100 score in 30 seconds, and run a voice-consistency check against your last three issues so you can see if Issue 47 still reads like Issue 1. The classifier flags where ChatGPT defaults flattened your openings, your sign-offs, and the small specifics that made paid subscribers say yes in the first place. Built for weekly thoughts, curated link roundups, deep-dive essays, news digests, and paid-only premium issues across Substack, Beehiiv, Ghost, and ConvertKit. Free, no signup, no card. The point is the 30 seconds before send, not the rewrite after.
A blog post can be skimmed by anyone. A newsletter is opened by people who already said yes to hearing from you. They handed over inbox space, which is the most protected piece of real estate they own. The 30 seconds before send are the cheapest insurance against the slow drift that breaks that contract.
Readers have other places to find information. What they cannot get anywhere else is the specific way you notice things. ChatGPT defaults flatten that exact thing. A subscriber who opens three generic-sounding sends in a row will not unsubscribe with a complaint email. They simply stop opening. By the time you see the drop in the open-rate chart, six weeks of damage are already locked in.
Gmail, Outlook, and Apple Mail use opens, replies, scrolls, and deletes-without-open as inputs into where future sends land. Newsletters deleted unopened drift from the primary inbox into promotions, then into spam. AI prose correlates with both fewer opens and more unsubscribes, so the score is an indirect predictor of deliverability across the next six to twelve sends, not just the one in your hand.
Open rate bends slowly. Click rate moves first because readers stop clicking links inside an issue that did not feel worth finishing. Paid renewal is the lagging signal. By the time it shows up on the dashboard, four to eight issues of voice drift have already compounded. Voice consistency is a forward-looking metric the dashboard does not track for you, which is why the pre-send check has to.
A blog post can be edited after publishing. A send cannot. Once the newsletter ships, every subscriber sees whatever the score reflected at the moment you clicked. That is why the 30 seconds before send are worth more than any other moment in your writing process.
The point of the score is not to chase a number. It is to catch the moments where the prose stopped sounding like you. Four steps, fifteen minutes, no rewrite tool required for most issues.
Drop the full body of the newsletter into TextSight before you do anything else. The score returns in 30 seconds with a colour-coded sentence map: red for high-risk lines, yellow for borderline, green for safe. Aim above 70 for general content and above 80 for paid-only issues. Below 60 is where unsubscribes start showing up on the next two or three sends.
Paste your last three issues from the archive into the consistency panel. The classifier compares opening rhythm, sentence length distribution, sign-off pattern, and vocabulary density between the new draft and the historical baseline. If the new issue drifts more than 20 points from your normal range, you get a voice-drift warning even if the AI score itself looks fine.
The colour map points at the 20 percent of the draft that is causing 80 percent of the problem. Scroll the highlighted view and read each red sentence aloud as if you were saying it to a subscriber in a coffee shop. If it sounds stiff or could open any newsletter on your topic, replace it. Most red sentences cluster in three places: opener, bridges between sections, and the sign-off.
Edit the red sentences in your own voice. One specific concrete noun per red sentence (a name, a place, a number that is not round) is usually enough to flip the colour. Re-scan once to confirm the score moved into the safe band, then schedule the send. Scanning every revision turns the score into a target you optimise for, which makes the prose flatter, not better.
The bands below are calibrated against paid-newsletter outcomes. The thresholds are tighter than for blog content because subscriber inboxes are a higher-trust surface than a public web page.
Reads like you wrote it between coffee sips. Sentence rhythm varies, asides appear where a human voice would put them, and the small word choices feel personal. This is the band where loyal subscribers reply to the email instead of just opening it. Open rate, click rate, and paid renewal all hold at their normal level.
A few tells, usually clustered around section transitions or the closing call to action, but subscriber-safe overall. Open rates and reply rates should sit at their normal level. Two minutes of editing the red sentences moves this into the top band if the send is going on a high-traffic day or a paid renewal cycle.
Open rate on the next send drops slightly, usually two or three points. Attentive subscribers notice the prose feels off without being able to name why. The fastest fix is to rewrite the first paragraph in your voice and strip the worst AI-tell words from the rest. Do not ship at this score for a paid list.
Unsubscribes spike on this send. Expect a measurable bump in opt-outs within 24 hours and a noticeable drop in opens on the send after this one. Most loyal readers recognise the prose is not yours. Do not ship this send under your byline. Restructure the draft or pull the schedule by a day and rewrite.
Mass unsubscribe plus spam reports. This is an unedited ChatGPT or Claude draft. Subscribers report the email as spam in numbers large enough to hurt sender reputation across your whole domain, not just this list. Throw out the draft, start from a real observation, and rewrite from scratch.
The detector was calibrated against newsletter content across the five formats below. ChatGPT defaults toward slightly different patterns in each one, and the score weights adjust so the band you land in is calibrated to the register the format actually needs.
The 400 to 1,000 word personal-essay format that anchors most independent newsletters. Voice is almost the entire value here, so this is where the score matters most. Target above 75. The classifier weights opening rhythm and sentence length variance heavily because that is what subscribers internalise across Issues 1 through 3.
The "five things I read this week" format. The risk is parallel structure across items reading uniform. The score flags roundups where every item has the same length, the same connective tissue, and the same neutral framing. Target above 70. Roundups have lower voice density than personal essays, so the band can be slightly looser.
The 1,500 to 4,000 word issue paid subscribers usually expect at least once a month. ChatGPT structures these cleanly with template H2s and bullet-heavy bodies, and the cumulative effect is what reads AI. Score the issue section by section if it is over 5,000 characters. You do not need to rewrite the whole essay, just the sections with the highest density of red sentences. Target above 80 for paid deep-dives.
News-style summaries of the week's important stories in your beat. Summaries are inherently structured, so it is easy for the prose to read template. The score flags digests where every story is framed the same way and where editorial reaction is missing. Add a one-sentence opinion per item. That is the move that turns a digest into something a subscriber would pay for.
Subscriber-only deep dives, frameworks, interviews, and analysis. This is the format where the score matters most because the value proposition is your specific judgement. Target above 80. Add an experience anchor every 400 to 600 words: a specific number, a client name with permission, a concrete failure, an opinion the model would not commit to. The classifier flags where these anchors are missing.
Free covers a couple of issues a week. Pro is $19.99 monthly or $14.99 effective on yearly. Business adds API access for automated pre-send checks across multi-title teams. Full details on the pricing page.
Billed $89.88/year — Save $30
Billed $179.88/year — Save $60
Billed $359.88/year — Save $120
Yearly billing saves 25%. View full pricing →
Most red sentences in newsletter drafts share a small set of patterns. None is wrong in isolation. The problem is the cumulative effect when subscribers see the same patterns across two or three issues. The scorer flags them so the five minutes before send go to the right lines.
"I have been thinking about," "Here is what stood out this week," "Let me share a quick observation." None of these is bad writing in isolation, but if every issue opens with one of them, regular readers notice by Issue 3. Replace the opening with something that describes a specific moment from your actual week: a place, a name, a number that is not round.
"Thanks for reading," "Until next week," "Talk soon." The closing meta-paragraph is where ChatGPT is least varied. The scorer flags it because the sign-off is the highest-recall part of a newsletter, the line subscribers see right before they close the email. Either rotate across three or four genuine closing patterns or drop the meta-close entirely and end on the last substantive line.
"It can be argued that," "in many cases," "it is worth noting that." This is the most damaging tell for paid content. Paid subscribers paid for a specific point of view, and hedged prose reads as a researcher's summary rather than as your judgement. The scorer sharpens these lines and flags where you should commit to a position instead of presenting two balanced perspectives.
Delve, leverage, robust, navigate, underscore, showcase, myriad, tapestry. These cluster in AI-drafted newsletter prose because they were optimisation targets in marketing training data. If you cannot replace one without losing the meaning, the sentence needs a real rewrite, not a single-word swap.
In a roundup or weekly-thoughts issue, ChatGPT defaults to list items of roughly the same length, the same sentence structure, and the same emotional register. Real human roundups have one item that runs three times as long because you had more to say, one item that is two sentences because there was not more to say, and one item that breaks the parallel structure entirely. The scorer flags the symmetry.
The AI score on its own does not catch slow drift. A newsletter can score 75 and still sound like nobody in particular. The voice-consistency check compares the new draft against your last three issues so you catch the gap the score misses.
Paste your last three issues from the archive into the consistency panel alongside the new draft. The classifier extracts four features from the historical baseline: opening rhythm, sentence length distribution, sign-off pattern, and vocabulary density. The baseline is a small statistical fingerprint of what "your voice" has looked like across recent sends.
If the new draft falls more than 20 points outside the baseline range on any one feature, the panel returns a voice-drift warning. A 20-point drift is the threshold where regular readers start internalising the new issue as a different writer. Below 20 they cannot name the change. Above 20 they can. The warning is feature-specific so you know whether to rewrite the opener, the cadence, the sign-off, or the vocabulary.
After the score and the consistency check, run one last sanity check. Imagine a close friend who reads your newsletter every week and knows the way you talk. Imagine the subject line and first paragraph of this send arriving in their inbox without your name on it. Would they recognise it as yours? If no, the draft needs one more pass, no matter what number the scanner returned.
The classifier reads specificity as a strong human signal because AI defaults toward category-level abstractions. Replacing a vague claim with one concrete noun (a name, a place, a number that is not round, an exact moment) is usually enough to move a red sentence into yellow or green. One specific noun per red sentence beats ten generic-word swaps.
The sibling page for rewriting flagged lines. Three modes, open-rate aware, voice-led rewrite.
Open the AI rewriter →The sibling scorer for blog content on Substack, Medium, and Ghost. Looser bands, different signals.
Open the scorer →The detector page focused on Substack newsletters, voice drift, and the subscriber-trust framing.
Open the detector →How the score is computed and what threshold to aim for before you schedule a paid send.
Read the guide →Free to try, no card. 0-100 score in 30 seconds, voice-consistency check vs prior issues, sentence-level highlights.