Medium pays writers on reading time from members, not on views. An AI-flavored article gets opened and closed inside twenty seconds, the read-time pool stays small, and the boost system has no reason to push it further. This guide walks the five-step workflow: draft the hook by hand, scan the draft with TextSight, identify the Medium-specific AI tells, run the Balanced AI rewriter on flagged sentences, then publish and watch the read-time metric. Built for Partner Program publishers, boost-eligible writers, and anyone whose Medium income depends on readers staying past paragraph two.
Medium readers are not casual web visitors. They are paying members reading inside an app that ranks by how long they stay, and the Partner Program pays on that same time. The system rewards articles readers finish; it does not reward articles that get opened and closed.
Earnings on Medium are tied to reading time from members. A 5,000-word article that nobody finishes pays worse than a 1,500-word article that everyone reads through. AI-flavored writing tends to skew the ratio the wrong way because the opening lines do not pull readers in, and once the reader closes the tab the reading-time pool stops growing.
Medium's boost system is editor-curated, and editors are looking for articles members will stay inside. AI-flavored articles rarely pass that bar because the patterns that signal generic content also signal low retention. A boosted article reaches more members, which feeds more reading time back into the Partner Program payout. Rewriting the prose is upstream of both signals.
Most popular Medium tags carry hundreds of new articles a day. If your piece on productivity reads like every other piece on productivity, it loses on attention because the reader has nothing to stay for. Voice is the thing that wins in a crowded tag; structure and topic are commodities, and a rewritten voice is the differentiator readers actually feel.
A reader who follows you on Medium is committing to a voice. AI-flavored articles read like a feed, not a writer, and they slow the rate at which a follower turns into a regular. Build a reading audience by being consistently yourself; the followers stay because the voice stays, and the read time compounds across your back catalogue rather than collapsing on each new piece.
Plan on thirty minutes start to finish for a typical Medium article. The point of sentence-level highlights is that you do not have to rewrite the whole piece; you only rewrite the lines that read AI, and you do it before Medium's read-time metric tells the truth on you.
Ten minutes, no model open. Write the first three or four lines of the article in your own voice. The top of a Medium article is the only part most readers see; if those lines do not sound like a person, the article is closed before paragraph two and the read-time damage is already done. Open with one specific image, one honest admission, or one line that commits to an angle a model would not have picked. Polish later. The hook is the contract with the reader, and you cannot delegate it to a model and expect the reader to keep reading.
Paste the full article into TextSight and read the sentence-level highlights, not just the headline score. A baseline scan tells you exactly which paragraphs are carrying the AI signal Medium readers feel inside the first thirty seconds. Save the score so you can measure the delta after the AI rewriter pass. The headline number is less useful than the per-sentence view, which points at the lines you copied from a model and forgot to edit back to voice.
Mark the patterns that fail in Medium-shaped ways before you touch the AI rewriter. Look for uniform listicle structure, generic insights that could apply to any topic, three-line openers that say nothing, formulaic "I've been writing about" introductions, and tidy closing summaries that restate the article. These are the lines a regular Medium reader closes the tab on, and they are the lines the Balanced pass is going to do the most work on.
Balanced is the default for Medium articles, and it is the one we recommend over Light and Maximum for this format. Light is closer to a proofread and tends to leave assistant-register sentences intact. Maximum is aggressive enough that it can flatten the voice you have built, especially in opinion pieces where the angle is the whole point. Balanced shifts cadence and vocabulary enough to clear the patterns Medium readers feel without erasing what makes the article yours. Run it on the highlighted sentences, then re-scan to verify the score has moved into the upper two bands.
Publish the cleaned draft and open the Medium stats page after the first three days. Read time is the only ground truth the platform gives you, and it does not lie. A high view count with a low read time means readers opened the article and closed it; that pattern is the clearest signal an article still reads AI-flavored. Compare read time across your articles, find the ones where readers stayed, and study what those have in common. Change the workflow when read time slips; it tells you whether the previous four steps actually held under real readers.
Articles fail in article-shaped ways. Below are the patterns Medium members feel inside the first half-minute, even when the prose lower down reads passable. Mark each one before the AI rewriter pass and they become the lines Balanced does the most work on.
Five numbered points, each one a paragraph, each paragraph closing on a tidy clause. Numbered lists are the default shape a model reaches for, and Medium members scan-and-bounce on them. Use question-based subheads instead and let the body breathe between them. One or two list items longer than the others, a parallel opener broken on the third item, and the structural symmetry that gave the article away starts to read like a person again.
"Consistency is key" or "success requires dedication" are model fillers, not insights. A real writer trades the universal observation for the specific moment: the tool they actually used, the number they actually measured, the conversation they actually had. Generic insights signal a model warming up because they cost the writer nothing to type. Specific signals cost something, which is exactly why they read human.
"As someone who has been writing about productivity for years" is a model trying to borrow authority it does not have. A human writer opens mid-thought: a specific moment, an admission, a question they cannot answer. The opening is the only part most readers actually read; a formulaic credential opener collapses the read-time metric inside the first paragraph and the rest of the article never gets a chance.
"In conclusion, the strategies we have discussed are essential" is a model trying to sound conclusive. Real writers end with a question, an unresolved thought, or a one-line image the reader carries with them. The last line is what readers remember between issues; do not waste it on a recap of the article they just finished. Closing summaries are also the easiest tell for the boost-system editors to spot.
When every paragraph runs four neat sentences and the transitions are all "moreover" and "furthermore," readers feel a machine rhythm. Vary the lengths the way your speaking voice does. Drop a one-line paragraph between two long ones. Skip a transition. The unevenness is what reads like a person, and the AI rewriter's Balanced pass varies cadence in exactly this way when you run it on the flagged sentences.
Medium's economics for writers run on two loops, and a rewritten article feeds both. Knowing how the loops connect is the difference between writing for the read-time metric and writing against it.
The boost system is editor-curated. A small team of editors reads submissions from boost nominators and selects articles they believe members will stay inside. Boosted articles get distributed more widely across the platform, which means more eyeballs and more reading time. AI-flavored articles rarely get boosted because the patterns that signal generic content also signal low retention, and editors are reading for retention. A rewritten article reads the way editors are looking for.
The Partner Program pays based on reading time from Medium members, not on views. Each second a member spends inside your article contributes to a shared pool that gets distributed monthly. A 1,500-word piece that members finish pays more than a 5,000-word piece members close after the first paragraph. The AI-flavored draft loses on both ends: lower read-through inside the article and a smaller pool of members reaching it without the boost.
Rewriting the prose is upstream of both loops. It does not write the article for you and it does not insert the angle you owe the reader. What it does is remove the patterns Medium readers feel as borrowed, so the article you wrote gets a fair chance at the read-time metric. The TextSight scan tells you whether the writing should hold; the Partner Program payout tells you whether it did.
Balanced is the right default for Partner Program writers because it shifts cadence and vocabulary enough to clear AI signals without flattening the voice members are paying for. Light leaves too much of the assistant register in place. Maximum can erase the voice you built across previous issues, which is the asset members are subscribing to. Balanced on the flagged sentences, then a re-scan, is the workflow that holds across a publication cadence.
All three modes available on every paid plan. Pro is $19.99 monthly or $14.99 effective on yearly. Business adds API access for editorial workflows. 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 →
The newsletter-specific workflow, paid-subscriber tells, and how the read-aloud test catches voice failures the scanner cannot see.
Open Substack guide →The blog post AI rewriter page, with the Google E-E-A-T context and the workflow for WordPress, Ghost, and Webflow publishers.
Open blog guide →The standalone AI rewriter tool. Three modes, sentence-level highlights, closed-loop calibration, free quota, no signup.
Open AI rewriter →How the 0 to 100 score is computed, what the five bands mean, and what threshold to aim for before you publish on Medium.
Read the guide →Free to try, no card. Sentence-level highlights, three modes, Balanced default, Partner Program-safe.