Paste the post into the checker below, press check, and you get a 0 to 100 score plus one of three verdicts: reads as AI-generated, borderline, or reads as human-written. It is free, 10 credits a day per network, and a check costs 1 credit whatever the length, so a 400-word news item and a 2,500-word guide cost the same single credit. The minimum is 65 words, and there is no account, no login, no card and no email address. Then read the number for what it is. It is not a search penalty waiting to happen, and it is not a verdict on who wrote the post. It is a reading of how closely your draft resembles the average of everything already written on your topic, and that resemblance is the thing worth fixing before you publish, because it is what readers notice first. This page covers what the score does and does not mean, which lines in a draft push it up, and how to check a post properly before it goes live. The short version of the whole thing is that a high reading on a draft you are about to publish is worth acting on, and the action is usually to add something only you could know rather than to run the text through another scanner.
The reading describes how densely certain patterns appear in the text you pasted. It is an editorial signal about the writing. It is not a prediction of what a search engine will do with your page, and it is not a prediction of what a different detector will report on the same paragraph. No tool can tell you that from the outside.
Search engines publish their own guidance on AI-generated content, and the short version of what they say is that helpful content is rewarded regardless of how it was produced, while pages mass-produced to game search are treated as spam. Read it at the source rather than acting on anyone's summary, including this one. The split it draws is the part that matters to you: a page built to occupy a keyword is one thing, and a useful post that happened to start as a model's draft is another.
What the score does track well is sameness. Generated drafts converge on the middle of what has already been written about a topic, because that is what the model learned from. A reader who already has three tabs open on the same query recognises the fourth quickly and leaves, and that behaviour costs you more than the score ever will.
Ask a model for a post on a topic and it returns the consensus of everything written on that topic before, with the specifics filed off. Every claim is one that was safe to make in general, which means every claim is one your competitors have already made.
The missing ingredient is first-hand material: the number from your own dashboard, the price you actually paid, the thing that broke when you tried it, the sentence a customer said on a call. A model has none of that, so it fills the space with reasoning about the topic instead of information from inside it.
The second cause is the skeleton. Generated posts follow one outline. Define the term, explain why it matters, list three benefits, give numbered steps, list common mistakes, close by restating the headings. That shape is acceptable on a definition page and deadly on any query where you have a view worth reading.
These are the patterns that push a score up, and they are also the lines a good editor deletes first. The registry of 44 named AI writing tells explains each one with the reasoning behind it, and it runs in your browser with no daily limit, so you can paste a working draft into it as often as you like while you edit.
Read through the list once with your own draft open. Most posts that score high are carrying three or four of these, usually in the introduction and the conclusion, and fixing those two paragraphs moves the reading more than any other single edit.
Do these in order, because the first one changes what the rest of the post is able to say. Adding information beats rearranging sentences every time, and it is the only edit that also helps the page compete.
The credit arithmetic page works through what a full rewrite costs, but for most posts you will not need one once these edits are in. The score tends to fall as a side effect of the writing getting better.
A check costs 1 credit whatever you paste, so length never costs you anything. What spends credits is splitting the post up. The whole thing in one go is one credit, and the same post checked section by section is one credit per section.
Ten credits a day covers ten checks, which is enough to map a long post and still re-check the parts you rewrote. The free AI checker page covers the flat pricing in more detail, and the verdict band to watch is the middle one: borderline means the reading is unresolved, so treat it as neither result rather than as a pass.
The humanizer rewrites every sentence in one click. A run takes 12,000 characters, roughly 2,000 words, with a 25-word minimum, and it costs 1 credit per 50 words. A 1,500-word post is therefore 30 credits, which is three free days, so the arithmetic is worth doing before you paste.
The better move is to spend the credits where the tells are. Rewrite the introduction and the conclusion, usually 300 to 500 words together and six to ten credits, then fix the body by hand with the free tools. Body sections built from material you supplied are rarely where the problem sits.
One thing to get right about the order: add your first-hand facts before you rewrite, not after. A rewrite works only from the text you hand it, so a draft with nothing specific in it comes back as a rewritten draft with nothing specific in it. The sentences change and the reason to read the page never appears.
Detectors measure style, not authorship. Clean, orderly, correct prose scores high whoever produced it, which is exactly what a professional writing to a house style produces. It falls hardest on people writing in a second language. Liang and colleagues at Stanford HAI found in 2023 that 61% of TOEFL essays by non-native English speakers were falsely flagged as AI-generated across seven detectors. Every one of those essays had a human author.
For evidence, the signed drafting record captures the shape of a writing session: when you worked, how much you cut and rewrote, and whether anything large arrived in a single paste. It signs that record with an ECDSA P-256 key and commits it to your exact text with a hash, so anyone can check the signature and the match on the verify page without contacting us. It records forward, so start it on the next post rather than the one already delivered. The originality check answers the separate question of whether your text overlaps published sources.
On the rules, briefly. Rewriting a draft you were permitted to generate is ordinary editing. Where a client contract or a publication's guidelines say AI-generated text is not permitted, that covers the draft however it is edited afterwards. Read the agreement you are publishing under.
Search engines publish their own guidance on AI-generated content, and the short version is that helpful content is rewarded regardless of how it was produced, while pages mass-produced to game search are treated as spam. That is as far as we will go, because the rest would be guesswork. The practical version: a page that exists only to occupy a keyword is the exposure. A page carrying something a reader cannot get anywhere else is not, whatever wrote the first draft.
The score is not a search signal and it does not travel anywhere. It measures how densely certain patterns appear in the text you pasted, and it cannot tell you what a search engine or any other detector will conclude about your page. Use it as an editorial read on whether the draft sounds like everything else already ranking for the query.
One credit. Checking is flat-priced whatever the length, so a 2,000-word guide and a 200-word introduction cost the same. The free allowance is 10 credits a day per network with no account, no card and nothing to verify, and the checker needs at least 65 words.
Not in a single day. One run holds 12,000 characters, about 2,000 words, so a 1,500-word post fits comfortably in one run, but rewriting costs 1 credit per 50 words and 1,500 words is 30 credits against an allowance of 10 a day. Either run it across three days, or spend the credits on the introduction and conclusion where the tells cluster and fix the rest by hand with the free tools.
Start with what you already have: document version history, dated drafts in email, research notes, interview recordings, the screenshots you took while working. Then add a signed drafting record for the work you do from here, which gives timestamps and revision behaviour that a third party can verify independently on the verify page. Nothing proves authorship on its own, but specifics answer an accusation far better than a denial does.