HomeAI Detector › Score Your Article for AI

Score your article for AI — a pre-publish credibility check.

Paste the draft. Get a 0-100 Authenticity Score in roughly thirty seconds, plus a sentence-level colour map that shows which lines read AI to an editor and which read in your reporting voice. Built for journalists, feature writers, magazine staff, and bylined columnists checking news pieces, features, opinion columns, and longform before the desk sees them. Editorial register, attribution preserved, no signup for the first scan.

Score my article free See pricing
0-100 score in 30 seconds Sentence-level highlights Editorial register Newsroom CMS ready
The 2026 newsroom reality

Why filing an article without a pre-flight check is now a real risk.

In 2023 an article that read slightly AI was a stylistic concern. In 2026 it is a desk-rejection concern at most national outlets and a trust signal that the audience itself has learned to read. There are three overlapping risks for working writers; pre-file scoring addresses the prose-pattern layer of all three before the desk sees the draft.

Newsroom AI policies and editorial review

Major outlets including the New York Times, Reuters, the Guardian, and the Associated Press updated their AI guidelines in 2024 and 2025. The current line is that AI assistance is acceptable for research and structural drafting but that the filed copy must read in the writer's voice. Editors are running AI-prose checks on flagged drafts and asking writers to rewrite formulaic passages. A pre-file score catches what the desk would catch and lets the writer fix it before the conversation happens.

Reader trust on bylined work

Subscribers to a magazine or a columnist sense AI prose before they can name it. The unsubscribe is not triggered by a single piece; it is the slow drift across four or five articles that all read a little flatter than the writer's archive. The score is the early warning that the drift has started, and the highlights show which sentences are pulling the register toward the generic.

Wire pickup and syndication checks

Aggregators and wire services run AI checks on submissions in 2026. A piece that scores low on a syndicate's internal check gets quietly dropped from the pickup queue, often without explanation back to the originating writer. Pre-file scoring catches the patterns that flag the submission so the piece clears the gate it would otherwise stall at.

The pre-file workflow

Four steps from draft to desk-ready copy.

This is the routine working journalists use before filing. For a 1,200-word piece, plan on 12 to 20 minutes start to finish. The point of sentence-level highlights is that you do not rewrite the whole article; you rewrite the lines that pull the score down.

Step 1: Paste the article and read the score

Drop the full draft into the scoring panel. In about thirty seconds you get a 0-100 Authenticity Score plus a section-by-section breakdown. Save the baseline if it falls below your filing target so you can measure the delta after the rewrite. For pieces longer than 800 words on the free tier, paste the lead and nut graf first, then the middle, then the kicker; the per-section score is more useful than the overall number anyway.

Step 2: Review the sentence-level highlights

Green lines passed every signal. Yellow lines tripped one or two and are usually acceptable on their own. Red lines tripped three or more and are the highest-leverage rewrites. The colour map matters more than the headline score; a 65 with four red sentences clusters in the lead is easier to fix than a 78 spread evenly across the middle.

Step 3: Rewrite the flagged lines in your reporting voice

Work the red sentences one at a time. Ask whether you would actually speak each line on a call with the editor. If not, rewrite it. Add a concrete detail from your reporting: a specific quote, a number from the document, an observation from the scene. These are the cues AI prose strips out. Replace formulaic transitions ("furthermore," "moreover," "in conclusion") with the natural connectives a reporter would actually use in narration.

Step 4: Re-scan, check attribution, file

Paste the revised draft back in. Target 80 for news and features, 85 for opinion columns and personal essays. Confirm every attribution and direct quote is still in position after the rewrite (AI-assisted edits sometimes drift attribution; verify manually). Then paste into the CMS or send to the desk as usual. The free Chrome extension also lets you re-scan inline against most newsroom CMSes for a final spot check before submission.

Score bands

What each range means for editorial copy.

The bands below are calibrated against published editorial content across news, feature, opinion, and longform. Numbers shift slightly by format (wire copy runs a few points lower at the same craft level because the register is tighter) but the bands below are the working defaults to file by.

85-100: Clears any desk

Reads as a working writer's voice. Editorial AI-checks see no meaningful patterns. Subscribers will read it the way they read your archive. This is the floor for opinion columns, bylined essays, and any piece where voice is the product. Senior columnists with established voices typically sit here on a clean first draft.

75-84: Acceptable for news and features

The standard band for working news and feature copy. Desk-side AI-prose checks pass on a first read. Aggregators pick up at this band without flagging. For opinion pieces or essays under your byline, push to 85 by editing the red sentences in the lead and the kicker, where AI patterns concentrate.

65-74: Desk-rewrite risk zone

The band where a senior editor reads the draft, hesitates on two or three passages, and asks for a rewrite. The article may still run but with the writer asked to rework the flagged sections. Editorial AI-checks at major outlets flag pieces in this band for additional review. Fixing the red sentences before filing is usually faster than the back-and-forth after.

50-64: Reads heavily formulaic

The band where the desk reads the piece, recognises it as either AI-assisted with light editing or written under heavy desk-style-guide pressure, and routes it back for a substantial rewrite. For bylined work, this is the band where the writer's editor starts asking whether the assignment was actually done. Restructure the lead and the kicker, then add three or four concrete reporting anchors throughout.

0-49: Reads as unedited AI output

Almost certainly an unedited or lightly-edited ChatGPT draft. Editorial AI-checks flag this with high confidence. Filing at this score risks a formal conversation about sourcing and attribution, not a stylistic rewrite. The fix is to throw out the draft and rewrite from notes and source documents, not to paraphrase your way out. Paraphrased AI prose carries its own fingerprint that current classifiers flag aggressively.

The editorial AI fingerprint

What gives an article away to an attentive editor.

ChatGPT writes editorial prose in a specific way that working editors learn to recognise on second read. The scorer flags each pattern below so the writer can pull them out before the desk sees the draft.

Uniform paragraph rhythm

Editorial prose has a rhythm: short paragraph, longer paragraph, short paragraph for a punch, longer for context. AI defaults to even three-to-five sentence paragraphs throughout. The scorer flags monotony in paragraph length and rewards the kind of varied rhythm a working reporter naturally produces.

Generic openings and stock leads

"In an era of," "as the world grapples with," "in today's rapidly evolving landscape." Stock lead constructions are the single highest-density AI-tell zone in editorial copy. The scorer flags the lead more aggressively than any other section and rewards leads that open on a specific person, scene, number, or observation.

Formulaic transitions between sections

"Furthermore," "moreover," "in addition," "it is worth noting that." Working writers use natural connectives or none at all (a hard scene break is fine). AI defaults to formal transitions on every paragraph junction. The scorer flags transition density and rewards prose that moves on the strength of the content rather than connective tissue.

Hedged claims and balanced register

"While some may argue that," "it could be said," "there are multiple perspectives on." AI hedges every claim because it has no stake. Editorial writing commits. The scorer rewards sentences that take a position and flags the kind of evenly-balanced framing that reads research-summary rather than reported.

Synthesis kickers and resolution endings

"In conclusion," "to summarise," "as we look toward the future," "ultimately, the path forward demands." Kicker synthesis is the second-highest AI-tell zone after the lead. The scorer flags kickers that read like wrap-ups and rewards kickers that land on a quote, a specific image, or a forward-looking detail rather than a summary.

Plans & pricing

Same scoring engine at every tier.

Free covers most solo writers filing two or three pieces a week. Starter suits busy freelancers and staff writers. Pro adds unlimited scans, longer per-scan caps, file upload, and the Chrome extension. Full details on the pricing page.

Free
$0/forever

 

Try article scoring. No card, no email.
  • 3 scans per day
  • 5,000 chars per scan
  • Sentence-level highlights
  • Plagiarism risk bundled
Start free
Starter
$7.49/month

Billed $89.88/year — Save $30

For freelancers filing 2 to 4 pieces a week.
  • 20 scans per day
  • 20,000 AI rewriter words/mo
  • Chrome extension
  • Email support
Get Starter
Business
$29.99/month

Billed $359.88/year — Save $120

For newsroom and editorial teams.
  • 150,000 AI rewriter words/mo
  • REST API + CMS hook
  • 5 team seats
  • White-label PDFs
Get Business

Yearly billing saves 25%. View full pricing →

Editorial trust

The reporting cues that AI strips out — and how to put them back.

The score is a proxy for the trust signals an editor and a subscriber read on the page. AI prose passes the surface but strips the small craft cues that signal a working reporter. The scorer flags where they are missing so you can restore them before filing.

Concrete attribution over generic claims

"Experts say" reads AI. "Three staff at the Centre for Policy Research told me on Tuesday" reads reported. The scorer flags abstract claim sentences and rewards specific attribution: a name, a role, a place, a time. Even one specific attribution per section pulls a piece out of the risk band.

Scene detail and lived observation

"The meeting was tense" reads AI. "The meeting ran twenty minutes over schedule, and the chair tapped his pen against the table when the third question landed" reads reported. The scorer rewards sentences that anchor in observable detail (numbers, gestures, time markers, physical specifics) over sentences that summarise sentiment.

Honest caveats and admitted limits

AI prose removes caveats because they hurt the surface confidence of the writing. Working editorial prose includes them: a source who declined to comment, a number that does not reconcile, an open question the piece does not resolve. The scorer rewards these honesty signals and flags pieces that read suspiciously frictionless.

FAQ

Article scoring frequently asked.

Why should journalists run an AI score on their own article?
Even articles written entirely by hand can read AI-flavoured if the desk style guide is heavy on formulaic transitions and stock leads. Major outlets including the New York Times, the Guardian, and Reuters added AI-prose checks to their internal QA in 2025. A pre-file score catches the prose patterns that make a clean article look suspicious to an automated check, and the sentence highlights show exactly which lines need a rewrite to clear the desk.
What target score should an article hit before filing?
For news and feature articles where editorial trust is the product, target 80 or higher. For shorter wire-style pieces and aggregation summaries, 75 is acceptable. For opinion columns and bylined essays where voice carries the article, push to 85. Below 65 is the band where copy editors and editorial AI-checks both flag the piece, and where a desk rewrite is usually faster than defending the draft.
How is article scoring different from scoring a personal blog post?
The same five signals apply (sentence rhythm, transition vocabulary, paragraph structure, specificity density, conclusion synthesis) but the calibration is tuned for editorial register. Journalistic prose runs tighter than blog prose: shorter paragraphs, more attribution, more concrete nouns. The scorer expects this and adjusts. A piece that would score 78 as a blog post often scores 72 as an article because the tighter register concentrates AI signals into fewer sentences.
Will an editor be able to tell I used AI even if my score is 80 plus?
A score above 80 means the prose-pattern layer reads clean to automated checks and to most attentive editors on a first pass. What it does not catch is content errors AI tends to introduce: fabricated quotes, invented sources, statistical hallucinations, attribution drift. The score covers the writing fingerprint; the fact-check layer is still on the writer. A clean 80 plus score plus a clean fact-check is the standard most desks now expect.
Can I score a 2,500-word feature on the free tier?
The free tier caps each scan at 5,000 characters, which covers about 800 words. For a longer feature, split the piece into the lead and nut graf, the middle, and the kicker, and scan each section. This is diagnostic anyway because AI prose patterns concentrate in specific sections (most commonly the lead and the kicker). Pro at 19.99 USD monthly (14.99 USD effective on yearly) gives 10,000 characters per scan, which covers a typical magazine feature in one pass.
How does this fit into a CMS workflow for a newsroom?
Paste the draft into TextSight before pasting it into the CMS. Run the scan, review the sentence highlights, rewrite the flagged lines, then re-scan to confirm the score cleared. Once it does, paste into the CMS as usual. Business tier exposes a REST API some newsroom CMSes integrate so a draft going from writer to editor automatically routes through scoring first; below 75 routes back to the writer with highlights, 75 or above moves to editor review.
Does scoring help with opinion columns and bylined essays?
Yes, and this is where the scorer earns its keep for senior writers. Opinion columns live or die on voice; ChatGPT smooths voice into a generic centrist register that strips the cadence subscribers recognise. The scorer flags sentences that read suspiciously balanced or hedged and rewards sharpened claims and personal stake. For columnists who have built an audience on a specific voice, the score is the early warning that the voice has drifted.
Related

More for the working writer.

Score your article before you file. Ship in your reporting voice.

Free to try, no card. 0-100 score in thirty seconds, sentence-level highlights, editorial register, attribution preserved.

Score my article free See pricing
Editorial register · Helpful-content compliant · Pre-file, not post-file