The problem is not only that AI can hallucinate.
Sometimes it fails by becoming too agreeable.
For academic writing, that is dangerous in a quieter way. A hallucinated source is visible once checked. A too-smooth paragraph can be harder to catch because it sounds reasonable, balanced, and helpful. It may soften the edge of a claim, flatter the user's framing, or drift toward the answer that feels socially acceptable rather than the answer the evidence can carry.
That is exactly where scientific writing needs a different interface.
A serious system should not only ask: can I write this nicely? It should ask: can this sentence carry a source? What would the counter-evidence say? Is the claim too wide for the study? Where does uncertainty belong?
Footnotes, not flourish
This is why NotaVera starts with footnotes, not flourish.
When a claim appears, NotaVera makes the evidence relation visible: source candidate, verified source, counter-evidence, weak support, boundary case, or no source found. If the sentence is too broad, it does not decorate it with a footnote. It narrows the sentence until the source can honestly support it.
The point is not to make AI less useful. The point is to make it useful for people who cannot afford smooth nonsense.
A source trail is not decoration. It is the difference between scholarship and theatre.
How this looks in the product
In NotaVera, this theme becomes visible as claim markers while writing, footnotes under the page, evidence cards in the margin, counter-evidence where relevant, boundary-case warnings, suggested narrower wording — and no fake citations, ever.