Rewriting AI content is not the same task as score reduction on a detector. The detector measures patterns; your reader measures voice. A real rewrite has to handle both, and the way to do that is to stop thinking about word swaps and start thinking about the four structural patterns every AI draft carries. Inside: a five-step workflow built on the detector and AI rewriter, the four patterns that reward a manual pass (tripled adjectives, uniform sentence rhythm, corporate vocabulary, transition phrase clustering), a decision tree for choosing between hand-editing and the AI rewriter, the three AI rewriter modes and when each one fits, and a short list of the specifics only you can add. By the end you should know what to fix yourself, what to delegate, and what the detector is actually scoring as you go.
Rewriting is about replacing whole passages, not nudging a word here and there. Detect first so you can see which passages are AI all the way through. Decide which ones to regenerate from scratch versus salvage. Run those passages through the AI rewriter at the intensity the salvage needs. Re-detect to confirm the new prose landed. Then re-own it by putting back the facts the regeneration could not know.
Paste the draft into TextSight at app.textsight.ai. You get an overall 0 to 100 AI score, a per-sentence highlight map, and a bundled Plagiarism Risk score in the same scan. Capture both numbers and screenshot the highlight map; this is your before-and-after baseline. A 95 percent score is normal for an untouched ChatGPT or Claude draft. The map matters more for rewriting than for light editing, because it shows you where the red is solid rather than scattered. A paragraph that is red end to end is a regenerate candidate. A paragraph with one or two flagged sentences inside otherwise clean prose is a salvage candidate, and you would handle that differently.
This is the decision that separates rewriting from editing. For each block of solid red, ask whether the passage carries any of your own thinking or whether it is pure generated filler restating the prompt. If it is filler, mark it for full regeneration; you will replace the prose wholesale rather than picking at individual words. If it carries an argument you actually want to keep, mark it for salvage and note the specific patterns to address. Most heavy AI drafts split roughly into one third worth regenerating from scratch and two thirds worth salvaging, and writing that split down before you touch a sentence is what stops you from spending an hour polishing a paragraph that should have been deleted.
Run each marked passage through the TextSight AI rewriter at the intensity its block needs, working passage by passage rather than dumping the whole document in at once. A solid-red filler block goes through Maximum, which replaces structure and phrasing while holding the ideas. A salvage block where you want to keep the shape goes through Standard. Passage-level regeneration keeps the rewriter focused: it has less context to drift across, so a 2,000-word draft handled in five or six chunks comes back tighter than the same draft run as one pass. After each chunk comes back, read it as a fresh draft, keep what reads as yours, and send anything that drifted back through at a lower intensity.
Re-scan the reassembled draft and compare it against the starting screenshot. Rewriting earns a second detect that light editing does not always need, because regenerated prose can carry its own AI signature even after the original one is gone; a Maximum pass that swaps one uniform rhythm for another uniform rhythm will still cluster red. A genuine rewrite usually brings 95 percent down into the 20 to 45 percent band with scattered residual highlights rather than solid blocks. If a regenerated passage still reads as a wall of red, the regeneration replaced AI prose with more AI prose, so run that block again at a different intensity or rewrite its opening two sentences by hand to break the new pattern.
A regeneration replaces phrasing, but it cannot invent the specifics that make a passage yours, and it can quietly drop the ones the original draft had. So this last pass is non-optional after rewriting in a way it is not after light editing. Read the regenerated draft against your notes and put back every concrete fact: the names, the dates, the figures, the lived examples. Where a passage was regenerated whole, confirm the new version did not paraphrase a real number into a vague one. This pass barely moves the detector score, but it is the difference between a clean rewrite and a passage that reads like competent generic prose owned by no one.
When you regenerate a whole passage rather than fix individual words, the question changes: does the rewriter actually clear each pattern, or does it just trade one version of the pattern for another? These four are the ones a careless Maximum pass leaves intact, so they are the ones to check first when you re-detect a regenerated block.
"A robust, comprehensive, multifaceted approach." Three adjectives in front of one noun is one of the cleanest AI signatures, and it is also one a regeneration often reproduces because the model that wrote the draft and the model rewriting it share the same habit. A Maximum pass can hand you back "a powerful, thorough, wide-ranging approach," which is the identical pattern with fresh words. When you re-detect a regenerated passage and it still clusters, scan for new adjective stacks first. The real fix is to replace the stack with the specific thing it gestured at: "an approach that catches both the obvious cases and the edge cases." That sentence has nowhere for a triple to hide.
If every sentence in a regenerated passage still lands between 16 and 22 words, the rewriter smoothed the vocabulary but kept the metronome, and the block reads AI even though no single word is wrong. This is the pattern most likely to survive a regeneration, because intensity controls word choice more than it controls cadence. The fix on a regenerated block is to merge two adjacent 18-word sentences into one 30-word sentence and follow it with a five-word line that lands the point, then leave the next two short sentences alone. You are introducing the burstiness the rewriter would not, and one hand-broken paragraph often resets the rhythm for the whole passage.
Frontier models have favourite words in 2026: delve, robust, leverage as a verb, navigate used metaphorically, underscore, showcase, myriad, tapestry, multifaceted, foster, holistic. A regeneration usually clears the obvious offenders but reaches for adjacent ones, so "delve into" comes back as "explore in depth" and "leverage" comes back as "harness." Read the rewriter output specifically hunting for the replacement vocabulary rather than the original. The durable fix is the plainest possible verb: delve and explore both become look at or examine, tapestry becomes pattern or mix, leverage and harness both become use, underscore becomes show. Plain words have no fancier synonym for the rewriter to climb back up to.
Furthermore, Moreover, In addition, Additionally, In conclusion. ChatGPT stacks these at paragraph boundaries to signal flow, and a regeneration tends to rebuild the scaffold even after you tear it down, because the rewriter reads connective furniture as good structure. After a regeneration pass, check every paragraph opener: a fresh "Building on this" or "It is worth noting that" is the same crutch in a new coat. The fix is to delete the opener with no replacement so the sentence underneath stands on its own, and where a passage genuinely needs a hinge, write a concrete one tied to the previous point: "Those two patterns are the easy ones to catch. The third is harder because" carries the turn with no furniture phrase at all.
The hard call in rewriting is not which tool to reach for, it is whether a passage is worth saving at all. Some AI blocks carry an idea you want and only need the prose replaced; others are filler that should be regenerated from a blank prompt or your own notes. Get this call right and the rest of the work follows. Here is the rule of thumb.
Some AI passages are pure scaffolding: an introduction that restates the title, a "in today's fast-paced world" lead-in, a conclusion that summarises nothing new. There is nothing to salvage in prose that never had a point, so rewriting it word by word is wasted effort. The faster path is to delete the block entirely and regenerate it, either by writing two or three sentences from your own notes or by giving the AI rewriter a tight instruction about what the passage should actually argue. Regeneration is also the honest move when a block is filler, because polishing empty prose just produces well-dressed empty prose that the detector and the reader both still see through.
When a flagged block contains analysis you genuinely want, salvage it: run it through the AI rewriter at Standard, then read the output against the original to confirm the argument survived. The risk in salvage is the opposite of the risk in regeneration. Here the danger is that an aggressive pass paraphrases your actual point into something blander, so keep the intensity lower and the read-through closer. A 200-word section that makes one sharp claim you stand behind is worth a careful Standard pass and a line-by-line check; it is never worth a Maximum pass that might sand the claim off.
On a 2,000-word AI draft this becomes a sorting job before it becomes a rewriting job. Walk the highlight map, mark each solid-red block as regenerate or salvage, then work the two piles differently: send the filler through Maximum or rewrite it from your notes, and send the spine sections through Standard with a close read after each. A typical long draft ends up with maybe 700 words regenerated wholesale, 1,000 words salvaged through the rewriter, and 300 words you write fresh because the draft was missing them entirely. Doing the sort first is what keeps a long rewrite from turning into an hour of polishing paragraphs that should have been thrown out.
A rewrite burns rewriter words fast because you are regenerating whole passages, not nudging a few. Free covers 3 detector scans a day and a 1,500-word rewriter quota for trying the loop on one short piece. Paid tiers raise the word budget for full-document rewrites and add the Chrome extension, file upload, and REST API. Yearly billing saves 25%.
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
In a rewrite you do not pick one mode for the whole document; you pick a mode per passage, because the regenerate piles and the salvage piles need different intensities. The wrong mode on a block either leaves the AI rhythm intact or rewrites so hard the idea drifts. Match the mode to how much of the passage is worth keeping.
Light keeps the prose close to the original, so it is the wrong tool for regenerating filler and the right tool for the thinnest salvage. Reach for it on a passage you largely wrote yourself and only ran through AI for tightening; the argument and the voice are already there and you just want the residual machine cadence washed out. Light typically moves the detector score by 15 to 25 points and preserves more of the original phrasing than the other modes. On a rewrite job it usually applies to a minority of passages, the ones where regeneration would do more harm than the leftover AI signal does.
Standard rewrites more aggressively while keeping the structure intact, which makes it the workhorse for salvage blocks, the passages that carry an argument you want but were written in AI prose. It handles the four patterns (tripled adjectives, uniform rhythm, corporate vocabulary, transition clusters) without dismantling the point underneath. Standard usually moves the score by 35 to 55 points on a heavy AI passage. Start your salvage piles here: if the output drifts from your meaning drop that block to Light, and if it comes back still clustering red on re-detect, escalate that block to Maximum.
Maximum rewrites the most, replacing structure and phrasing while attempting to hold the ideas, so it is the regeneration tool, the right pick for solid-red filler where you want the topic kept and the prose gone entirely. It is built for ChatGPT or Claude blocks whose underlying point you accept but whose every sentence reads machine-made. Maximum typically moves the score by 50 to 70 points. The trade-off is sharpest here: a Maximum regeneration can paraphrase a specific figure or claim into something vaguer, so the re-detect and the read-against-notes pass are non-optional. Run it passage by passage, never across a whole draft at once.
The sibling guide focused on editing rather than rewriting. Same four patterns, different emphasis.
Open the editing guideThe tool itself. Three modes, sentence-level diff view, free quota every day with no card.
Open the AI rewriterHow the 0-to-100 metric is computed and what each tier means for graded or published work.
Read the guideThe prevention angle: how to draft from scratch so the rewrite step is unnecessary.
Open the prevention guideDetector, AI rewriter, and sentence-level highlights in one workflow. Free to try with no card. 3 detector scans and 1,500 AI rewriter words on the free tier, every day.
Related walkthroughs on detection, scoring and authentic writing.