The web got loud. PRIMA makes it quiet.
A free, 100% local reader's lens that draws a clean red line through AI slop and SEO filler - and leaves the real writing standing. The paid detectors upload your reading to a server. PRIMA never sends a byte, because there is no server.
Live on the Chrome Web Store - one click, free, and it auto-updates. Works in Chrome, Edge, Brave, Opera, Vivaldi, and Arc.
In today's fast-paced world, it's important to note that artificial intelligence has become a truly pivotal cornerstone of modern business. OpenAI reported 400 million weekly users in February 2025. Moreover, leveraging these robust, cutting-edge tools can unlock the potential of any organization.
Pick a sample. PRIMA reads it locally and strikes the slop. Human journalism and plain ESL stay clean - that is the fairness paid detectors keep failing.
Finds the noise
A local, deterministic heuristic scores each block for the templated, low-signal, likely-AI sentences - AI-typical vocabulary, formulaic transitions, filler and SEO phrases, conclusion scaffolding, em-dash overuse.
Draws the line, shows its work
It strikes the slop and labels the block (Likely AI-generated / SEO-spun / Templated) with a signal-strength read and the exact markers it matched. An estimate, not a verdict - you stay the judge.
Fully reversible
Nothing is ever deleted or rewritten. One click restores the original page byte-for-byte, and turning the lens off removes every mark.
Take the clean copy
One click copies the block's substance to your clipboard with the slop removed - exactly the sentences left standing. Skim a bloated page, keep only what is real. The copy happens on your device; nothing is uploaded.
It refuses to punish plain writing.
The documented trap for AI detectors is a roughly 61% false-positive rate on non-native English writers - their uniform sentence length and smaller vocabulary read as "machine" to lazy classifiers.
PRIMA refuses to flag text on uniform length and low diversity alone. Flip ESL-safe in the demo above and watch the borderline marks disappear. Most detectors only claim fairness. PRIMA shows it.
No account. No upload. No server can read what you read.
Because there is no server. Every page is analyzed on your own device and the text is discarded immediately. This is not a promise you have to trust - it is how the thing is built.
third-party requests loaded by this page.
Self-hosted fonts. No CDN. No analytics.
Open DevTools › Network and watch.
| PRIMA | Paid cloud detectors | |
|---|---|---|
| Price | Free | Subscription |
| Where it runs | Your device | Their servers |
| Your reading | Never leaves the page | Uploaded to be scored |
| Account | None | Required |
| Source code | Open (MIT) | Closed |
| What you get | A signal + the reasons | A verdict / a score |
| Plain & ESL writing | Protected by design | Often penalised |
Same honest stance everywhere: a signal, not a verdict. No accuracy percentage, ever.
PRIMA asks for exactly one permission, and it is in the manifest for anyone to read:
No tabs. No host permissions beyond running the content script. No scripting, no activeTab. It cannot read your history, your other tabs, or your accounts. The whole thing is open source under MIT - read every line, or build it yourself.
PRIMA's markers are deterministic and research-backed - excess-vocabulary work (Kobak et al.), the patterns GPTZero and Pangram document, and Wikipedia's "Signs of AI writing". It is honest about its limits: heuristics flag patterns, not truth, so expect occasional misses on terse prose and well-edited marketing. There is no accuracy percentage here, and there never will be - that would be a verdict, and PRIMA only gives a signal.
Is PRIMA really free?
Yes. Free, open source (MIT), no account, no trial, no upsell. There is nothing to pay for and nothing to sign up for.
Does it send my reading anywhere?
No. Every page is analyzed on your own device and the text is discarded immediately. There are no servers and no network requests - open DevTools › Network and watch the page itself load zero third-party requests.
Will it slow my browser down?
No. Blocks are analyzed lazily, only as they approach the viewport, with a cheap deterministic heuristic - no model, no network, no background work. Heavy and infinite-scroll pages stay responsive.
What if it flags my own writing?
It is a signal, not a verdict - and it is fully reversible. ESL-safe mode protects plain and non-native writing from the signals that unfairly target it, you can dismiss any single flag, and turning the lens off restores the page exactly.
How is it different from the paid AI detectors?
They are paid, cloud-based, closed, and upload your text to score it - then hand you a verdict. PRIMA is free, 100% local, open source, shows you the exact reasons instead of a verdict, and is built to be fair to plain and ESL writing. See the comparison above.
Does it work on social media?
Yes. It detects the generic containers that feeds use (LinkedIn, X, Facebook, Reddit) and descends into open Shadow DOM, so it works well beyond classic articles - including infinite-scroll feeds.
Can it be wrong?
Of course - heuristics flag patterns, not truth. Expect occasional misses on terse prose and well-edited marketing. That is exactly why there is no accuracy percentage and why you stay the judge. A signal, not a verdict.
Get PRIMA.
Live on the Chrome Web Store - one click and it auto-updates. Prefer to load it yourself, or run it on Firefox? The source build is right here too.
One click installs it and keeps it updated forever - no re-downloading. Works in Chrome, Edge, Brave, Opera, Vivaldi, and Arc.
- Download the .zip below and unzip it
- Open chrome://extensions (or about:debugging on Firefox) and turn on Developer mode
- Click Load unpacked and pick the unzipped prima folder
Every Chromium browser: Chrome, Edge, Brave, Opera, Vivaldi, Arc. Safari needs a Mac build; DuckDuckGo has no extensions.