What Research Brainie will and will not do

At every pass, Research Brainie asks its question and waits. Brainie’s analysis stays hidden until you have written your own answer in your own words. There is no button that skips this, because skipping it would remove the only thing the product is for.

When your answer arrives, the feedback tells you what you got right, what you missed, and what you misunderstood, and ends with a further question. It does not hand you a better answer to copy.

• ×We do not write your literature review, your argument, or your dissertation.
• ×We do not generate text for you to submit as your own.
• ×We do not produce one-click summaries that let you skip reading the paper.
• ×We do not tell you a paper is good or bad, we ask you what you think, and why.

Fabricated quotations, page numbers and citations are the most damaging failure mode of AI in academic work. Research Brainie is instructed never to produce them, and classifies every analytical claim it makes so you can see where it came from:

Directly supported

Stated explicitly in the paper.

Inferred

A reasonable interpretation, not a quotation.

External knowledge

From outside your document, always labelled, never presented as the paper’s own content.

Uncertain

Not enough evidence in the text to say.

Where the paper does not state something, Research Brainie says so rather than guessing.

Limitations the authors themselves acknowledge are shown separately from limitations Brainie has noticed. The second kind are labelled as analytical observations, not facts about the paper, because an AI noticing something is not the same as a weakness existing.

A limitation is also not automatically a flaw. Methodological quality is judged relative to the research question, and non-empirical papers are not penalised for lacking a sample.

“I have not found research about this” is not the same as “no research exists about this”. Research Brainie reports potential gaps based on the papers you have uploaded, marked with a confidence level and a standing warning that a gap must be verified through a broader literature search before you build a project on it.

An uploaded PDF is read for its text and deleted in the same request, it is never written to our project storage. Once the analysis is complete, the extracted text is discarded too. What remains is the structured analysis and the notes you wrote yourself.

This means re-analysing a paper requires uploading it again. That is a deliberate trade: we would rather ask you for the file twice than hold publishers’ content indefinitely.

Research Brainie uses a large language model, and language models make mistakes. It can misread a table, miss a caveat in a footnote, or misjudge a method. Long documents are read only as far as one pass allows, and you are told when that happens.

Nothing here replaces reading the paper, and nothing here replaces your supervisor. Research Brainie is a training partner that argues with you, it is not an authority, and it should never be cited as one.

Students can generate a read-only link showing what they have read, the answers they wrote before seeing any AI output, and how their judgement has changed over time. The record is evidence of engagement, not of output. Students control the link and can revoke it at any time.

If your institution needs something specific in this statement before approving use, we would rather hear it than guess.