AI use,
without misuse.

What we believe

AI assistance in academic writing is not going away. The pretense that it will is increasingly absent from how universities, journals, and professional bodies are approaching the question. Most institutions now publish guidelines on permitted versus prohibited AI use, and most students have integrated AI tools into their workflow whether or not they discuss it openly.

We believe the productive question is no longer whether students will use AI but how they will use it. There is a meaningful and identifiable line between using AI to draft, structure, and refine your own thinking — and using AI to substitute for the academic engagement your institution expects.

Papyra is built for the first kind of use. The features we ship are designed to make legitimate AI assistance better. The features we deliberately do not ship are the ones that exist to make academic misconduct easier.

What Papyra will not do

This is the section that distinguishes us from most competing products. We have made specific, public commitments about features we refuse to build, regardless of the demand for them.

We will not fabricate citations

Every reference in a Papyra output points to a real, indexed paper. We retrieve from CrossRef and OpenAlex before any citation enters your draft. The architectural decision to separate citation from generation is the central technical commitment of the product, and it is not optional or toggleable. There is no Papyra mode in which the AI is permitted to invent sources.

We will not optimize for AI-detection evasion

Tools that exist primarily to help students fool AI detection systems are tools that exist to support academic misconduct. We do not build them. Our outputs are designed to be substantively defensible — meaning, they are written well and cite real sources — not to evade detection. If your institution's policy prohibits AI assistance entirely, Papyra is not the right tool for that course or assignment, and we say so plainly.

We will not impersonate identifiable people

Papyra does not generate content that purports to be the work or quotation of a real, named individual when it isn't. Fictional examples in tutorials and templates are clearly marked. Real public figures are referenced only in clearly factual or critical contexts.

We will not facilitate plagiarism

Outputs are generated for the user who requested them. They are not catalogued, redistributed, or made searchable as content other students might submit. The combination of generation patterns and citation verification produces outputs that are typically substantively different across users, but no AI-generated text can claim absolute uniqueness, and we say so in our terms.

We will not lie about what we are

Papyra outputs are clearly marked as AI-assisted in any context where that fact is required to be disclosed. We do not provide tools to scrub or hide the AI provenance of generated content. The user remains responsible for declaring AI use to their institution where institutional policy requires such declaration.

What Papyra will do

Real, verifiable citations

The single most important commitment we make. Every reference is retrieved from a real academic database before it appears in your bibliography. The DOI resolves. The journal exists. The paper is real. See our AI & Citations guide for the technical detail of how this works.

Drafts you can defend

The substantive content of a Papyra draft is generated to be coherent, sourced, and academically structured. The arguments are traceable to the references that support them. You should be able to defend any claim in a Papyra draft by pointing at the citation that backs it up — because the citation will be real and the source will say what we attributed to it.

Transparency in our methods

We publish, in our Citation Guide, the full conventions we follow when generating references. The verification pipeline is documented. The third-party services we use (Anthropic, OpenAlex, CrossRef, etc.) are all disclosed in our terms. There is no secret architecture.

Support for legitimate AI use cases

Papyra is built around the use cases that academic-integrity policies typically permit:

• Drafting an outline or first pass that you then revise into your own voice.
• Structuring an existing argument and identifying where evidence is needed.
• Verifying citations and replacing unverifiable references with real ones.
• Translating or refining the prose of a draft you wrote yourself.
• Generating a presentation from an essay you wrote.

How to use Papyra responsibly

Whatever the policy at your institution, the following principles apply:

1. Understand your institution's policy on AI assistance. Most universities have published guidance. Some courses have additional restrictions. Read both before assuming a Papyra-generated draft is permitted.
2. Engage with the output. Read it. Edit it. Question it. Replace passages that don't reflect your understanding. The output is a starting point, not a finished submission.
3. Verify the citations yourself. Even though our citations are retrieved from real databases, you should still spot-check that the source supports the claim attributed to it. The verification protocol in our AI & Citations guide takes thirty seconds per reference.
4. Disclose AI use where required. If your institution's policy requires students to declare AI assistance in submitted work, declare it. Papyra is not designed to hide its provenance.
5. Submit work you can defend. If you cannot speak intelligently about the content of your essay or the claims in its citations, that is a signal that you have engaged too little with the output. Engage more.

For professors and institutions

If you are a professor evaluating a paper that may have been generated with Papyra, the citation pattern is the most reliable signal of the tool's involvement. Papyra outputs typically have higher citation density than student-written drafts, and the citations universally resolve to real papers in CrossRef. The opposite pattern — bibliography heavy in citations that don't resolve — is much more typical of general-purpose AI assistants used without verification, and is a more reliable signal of fabrication.

For institutions designing AI policy, we welcome direct engagement. Our positioning is one of transparency: we do not believe AI policy is best served by tools that hide their involvement. If your institution wants to discuss legitimate use cases, restricted use cases, or how Papyra outputs would be classified under your specific policy framework, contact us through the contact page.

How we differ from competitors

The academic AI category is full of products that take the opposite stance. They market on the promise of evading AI detection. They produce reports that read as authentic by mimicking AI-detection-fooling text patterns. They generate citations without verification because the underlying model produces "plausible" output and that's good enough for casual use.

We have explicitly chosen not to compete on those features. Papyra wins on a different axis: substantively better drafts (because the citations are real), defensibility (because we don't hide our involvement), and trust (because we publish what we won't do).

This is a deliberate strategic choice. The market for AI-assisted academic misconduct is real but is not the market we want. The market for AI tools that academic institutions can engage with as legitimate participants in scholarly workflow is what we are building toward.

Our commitment

The commitments above are not aspirational. They are operational. They are reflected in our terms of service, our product architecture, and the features we have specifically declined to ship.

If you encounter a Papyra output that violates one of these commitments — a fabricated citation, an undisclosed AI marker, a feature that appears designed to deceive — contact us at [email protected]. The commitment is meaningful only if we hold ourselves to it, and that requires users telling us when we fail.