Built because the
citations had to be real.

The problem we built to solve

Every general-purpose AI assistant — ChatGPT, Claude, Gemini, Grok, Llama, every model trained the same way — has the same defect when it comes to academic citation: it invents sources.

The mechanism is the same regardless of the brand. Large language models are statistical text generators. When you ask them for references, they don't query a database — they generate plausible-looking citation strings based on patterns in their training data. The author names sound right. The journals look correct. The DOIs are formatted properly. None of it is verified, and a meaningful fraction of it doesn't correspond to any real paper.

This isn't a fringe edge case. The 2023 federal court matter Mata v. Avianca made it famous: an attorney submitted a brief with six fake citations generated by ChatGPT, Judge Castel sanctioned him, and dozens of similar court sanctions have followed since. Universities have begun seeing the same pattern in student papers. A professor reading a bibliography that contains a paper which cannot be found in any database now reasonably suspects the paper was generated by an AI that fabricated the reference.

For students, this creates a brutal asymmetry. The same tools that draft your essay can introduce fabricated citations you have no way to spot — and your name is the one on the bibliography.

Why competing tools have the same defect

Most academic AI products are wrappers around the same general-purpose models that hallucinate sources. They prompt-engineer the model to be more careful. They post-process the output to check for obviously broken DOIs. They add disclaimers asking users to verify. None of this fixes the underlying problem, because the citation is still being generated by the model rather than retrieved from a real database.

Across the academic AI landscape, fabricated citations remain widespread. Test any of these products with a topic outside the most well-trodden academic territory and check the bibliography against CrossRef. The hit rate of real, verifiable papers is often less than half. Sometimes it is much less.

Papyra exists because that approach has a structural ceiling. You cannot prompt your way out of a generation problem. The fix has to be architectural.

What we built differently

Papyra separates citation from generation. The AI never produces author names or year numbers in your draft. It produces reference markers — placeholder tokens — and a separate verification system retrieves the actual citation from real academic databases.

Specifically, the references in every Papyra output are pulled from CrossRef (the registry of digital object identifiers covering 140+ million scholarly works) and OpenAlex (the comprehensive open catalog of scholarly publications). The DOI in your bibliography resolves on doi.org. The journal exists. The paper is real.

This makes Papyra structurally different. The same architectural constraint that makes our drafts more conservative also makes them defensible. When your professor checks one of your citations, the citation will be there.

The honest student's AI

Our positioning isn't accidental. We are explicitly building for students who want to use AI tools the right way — to draft, to structure, to overcome blank-page paralysis — without crossing into fabrication or institutional-policy violations.

We do not build features that help students cheat. We do not optimize for evading AI-detection tools. We do not produce content designed to deceive academic integrity reviewers. The features we ship are designed to make the legitimate use of AI assistance better — better citations, more substantive drafts, clearer structure — while leaving the academic responsibility where it belongs: with the student.

Read more about this on our Academic Integrity page.

About Aithena LLC

Papyra is operated by Aithena LLC, a Colorado limited liability company based in Denver. The team is small and academic in background — the design decisions you encounter in the product reflect direct experience with university coursework, citation conventions, and the institutional realities students face.

We are independent and self-funded. No venture capital, no data-monetization model, no external pressure to ship features that would compromise our positioning. The product is paid for by users who buy credits or subscriptions. The relationship is straightforward: you pay us, we deliver verified academic work, the bibliography resolves.

Our full legal framework — terms of service, refund policy, privacy policy — is on the Terms & Policies page. For any direct question, our contact page has the full set of email channels.

What you can do with Papyra

We currently ship three core features:

• Document Correction — surgical edits to a draft you already have, including citation verification and replacement of unverifiable references with real ones.
• Presentation Generator — academic presentations with the same citation guarantees that apply to our written outputs.
• Citation Guide — a free reference covering APA, MLA, Chicago, Harvard, and the verification protocol that defends against AI-generated fabrications.

The AI & Citations guide is the most direct entry point if your interest is the verification problem itself. It walks through what hallucination looks like, why it happens regardless of the AI model used, and how to defend a bibliography in a world where fabricated references have become a routine occurrence.

Where to start

If you have a draft already, run it through Document Correction. The verification pass alone — replacing any unverifiable citations with real ones from CrossRef and OpenAlex — is what most users come for.

If you have not started yet, the home page covers the full pipeline. The Citation Guide is the right place to read first if you want to understand the conventions in your field before you generate anything.