Regi, Chief Verification OfficerAbout Real Press
Why I built this, where we are, and where we're going.
Why Real Press Exists
I started building Real Press for one reason — I want to know when I'm looking at content that is AI generated or not. I was honestly tired of getting served recipes that were generated instead of developed by a human, or team of humans. You can't stick ingredients together and hope they work — the AI generated images were also very annoying to me. That's how we ended up with Real Press. My hope is that you find Real Press useful too. I have tried to curate it to show you information you care about, and transparently show why it is categorized as such.
How It Works
Real Press is a search engine where every result comes with a score. It is modeled after an old school newspaper — from the design to the operations. We want to verify what we “publish,” and we are not afraid to correct our mistakes. Real Press runs content through multiple detection systems — text classifiers, writing-pattern analysis, image forensics, and we're working on video classification next — to combine their signals into a single rating from Human to AI. There is a badge on every result so you can trust that it's been verified.
Transparency
If you're curious to learn more you can dig into the details. Paid users see exactly which detection models contributed, where they agreed and where they didn't. If we change how scores are calculated, we document it and show you the history. If a result is wrong, we want you to flag it, and we'll rescore it so that it's more accurate. Transparency is our goal — we strive to give this back to you so you can continue to trust us with your searches.
Independence
Real Press is self-funded. That's on purpose. It means nobody can tell us to go easy on a company, change a score, or look the other way. Our ratings are honest because there's nobody to answer to except you. We will never change a rating because of a deal, a bribe, or a partnership. We will never sell your data to advertisers or data brokers — we make money from subscriptions, not surveillance. The big AI companies can't be trusted to check their own work — that's like asking the fox to guard the henhouse. Real Press is independent because it has to be.
Our Methodology
Text Analysis
Multiple classifiers examine writing patterns, statistical properties, and linguistic features independently. We use GPTZero as the primary paid signal and supplement it with custom heuristics that measure things like sentence-length variation, hedging language, and transition density. No single classifier decides the result.
Image Analysis
Forensic models check for AI-generated visual artifacts — unnatural textures, impossible geometry, statistical anomalies in pixel distributions. We also check for C2PA content credentials, a signed provenance standard that some cameras and AI tools embed at generation time.
Combined Scoring
Signals from all active models are weighted by confidence and merged into a single Human → AI rating. Text carries 50% of the composite weight, images 35%, and video 15%. When providers disagree, we show the disagreement rather than hide it — Pro users see the full breakdown.
Score Classifications
| Badge | What It Means |
|---|---|
| Human | Strong human signals across all models |
| Likely Human | Mostly human signals with minor ambiguity |
| Unsure | Mixed or insufficient signals — could be either |
| Likely AI | Mostly AI signals |
| AI | Strong AI signals across all models |
Known Limitations
- Short text is less reliable. Content under roughly 250 words gives our classifiers less signal to work with. Scores on short posts and snippets carry lower confidence and should be treated as rough guides.
- Non-English content has lower accuracy. Our primary text classifiers are trained predominantly on English data. Scores on content written in other languages are less reliable.
- Adversarial content can fool detectors. Text specifically engineered to evade AI detectors — through paraphrasing, synonymization, or prompt engineering — can reduce accuracy. This is an active research area, not a solved problem.
- Long-form text is most reliable. Confidence scales with content length and richness. News articles and essays score more reliably than headlines, product descriptions, or social posts.
- Scores may change over time. As detection models improve, we re-score existing content. A score from six months ago may differ from today's score on the same piece. We preserve score history so you can see how it changed.
How Corrections Work
If you believe a score is wrong, flag it. Flagged content is re-scored with the latest models. We preserve the full score history — previous scores are never deleted. If our detection improves and a classification changes, you'll see the full trail. Corrections submitted by users have already helped us identify weak spots in the pipeline.
What We Are / What We're Not
We Are
- A search engine for authentic content
- An independent verification tool
- A transparency layer, not a gatekeeper
- Pro-informed choice — we label, you decide
We're Not
- A plagiarism checker
- A content policing service
- An AI company
- Anti-AI — undisclosed AI content is the problem
Where We Are
We're very early in our journey, and I am so grateful that you're here. It's me and a handful of developers against this impossibly large problem. We have indexed about 1,000 articles so far and hope to scale very quickly to most of the useful web (~22M websites). I also recognize that my models aren't perfect yet, and they likely will never be fully perfect. I'm not going to pretend that we've solved the issue, but we're doing everything we can to bring you transparency at scale.
Thank you for joining us, and helping to make the internet a more transparent and authentic place to be.
You can also install our browser extension for Chrome or Firefox to check any page while you browse.
Read our full privacy policy.