AI Tell Scanner
Paste a draft and see which generic writing patterns a reader will notice. It flags nine specific tells and shows you where two or more land in the same sentence, which is what actually reads as machine-written.
One file, 22 KB. Nothing to install.
It cannot send your data anywhere, because it never talks to a server
Drafts of your own résumé and cover letters are not something to hand to a third party for scoring. This tool has nowhere to send them. The text you paste is analysed in the page and is gone the moment you close the tab.
The file loads no scripts, stylesheets, fonts, or images from anywhere, and
contains no fetch,
XMLHttpRequest, or
WebSocket calls. There is no server
side to it. The only web addresses in it are the three links in the header and
footer, which go nowhere unless you click them. You do not have to take any of
that on faith:
- Open it with your browser’s network tab showing and watch nothing happen.
- Download it, disconnect from wi-fi, and use it exactly as normal.
- Open the file in a text editor and read it. It is one file, and it is legible.
The copy served on this page and the copy you download are the same file.
SHA-256: fead7ea8e763a89c9745b3c4f890f774a48bb0e913ba52100b8d623fbf3533b2
What it does
You paste in a draft. It scores it and gives you back three things, in this order.
High-risk passages come first: every sentence where two or more tells overlap. That ordering is the whole point. One hedge in a paragraph is nothing. A hedge, an inflated verb, and a participle-phrase tail in the same sentence is what makes a reader stop trusting the page, and it is the thing a raw count buries.
Category totals come second, with a count for each of the nine patterns and a one-line fix for each.
Document-level checks come last, and only when there is enough text to justify them: whether every list you wrote happens to have exactly three items, and whether every sentence is the same length. Both are strong tells and neither shows up in a per-sentence scan.
The nine tells
| Pattern | What it looks like |
|---|---|
| Negate then assert | “It’s not about X, it’s about Y” |
| Hedged phrasing | “may potentially help to somewhat improve” |
| Overused transitions | “Moreover”, “Furthermore”, “Additionally”, stacked |
| Punctuation overuse | Em dashes doing the work commas should |
| Decorative rhetorical questions | “So what does this mean for your search?” |
| Inflated vocabulary | “leverage”, “utilize”, “spearheaded”, “robust” |
| Participle-phrase filler | “…, highlighting the importance of collaboration” |
| Uniform three-item lists | Every list, always exactly three items |
| Uniform sentence rhythm | Every sentence the same length, no variation |
What it is honest about
It is a style-pattern checker. The patterns it flags read as generic regardless of who or what produced them, and that is the actual measurable thing. It makes no claim about authorship, and the page says so directly rather than in a footnote.
This matters practically, not just philosophically. If you are worried a hiring manager will think you used AI, the fix is not proving you did not. The fix is writing something specific enough that the question does not come up.
Questions about this tool
No, and be wary of anything claiming to be one. No tool can reliably determine whether a particular passage was written by a person or a model, including the commercial products that say otherwise. What this measures is different and defensible: whether your writing contains patterns that read as generic, whoever produced them.
Nine patterns. Seven are per-sentence: negate-then-assert construction, hedged phrasing, overused transitions, punctuation overuse, decorative rhetorical questions, inflated vocabulary, and participle-phrase filler. Two are document-level: whether your lists are reflexively three items long, and whether every sentence is the same length.
The two document-level checks need a reasonable sample before they say anything: eight or more sentences for rhythm, three or more detected lists for list uniformity. Below that they abstain rather than guess, because firing on a five-sentence sample produced false positives in testing.
No. There is no server. Everything happens in the page, and nothing is stored between sessions.
Not necessarily. This is a checker, not a judge. A single tell in a sentence you chose deliberately is fine. What is worth looking at is the high-risk section, where two or more land in the same sentence.
The course this came from
The tool is free either way. If you want the thinking behind it, that lives in the community.
The tools are free. So is the community.
Ten courses, 109 lessons, and people going through the same thing.
JOIN THE COMMUNITYMembership is free. Members who are able may optionally contribute $5 per month. Both options receive the same access.