How Scholarly Checks Its Own Work
Most AI tools hand you the first draft. Scholarly does not: almost everything it makes goes through a review pass before it reaches you, checked against the sources it was built from.
This is why a creation takes a few minutes rather than a few seconds. It is also the main reason the output is worth reading.
This page explains what that review actually does, so you know what you can rely on and what still needs your eyes.
What Gets Reviewed
| What you made | What the review does |
|---|---|
| Video lectures and podcasts | Edits the script before it is voiced: clearer sentences, smoother transitions, less filler, same facts, same length |
| AI Slides | Edits every headline, label, and table on screen so each slide states its point |
| Flashcards | Checks each card against your source, tightens answers, improves hints, and adds cards for topics the first pass missed |
| Quizzes and exams | Checks every answer key against your source |
| Summaries | Rewrites any section that is too thin or misses a topic your source covers |
| Essays and pages | Fact-checks against your sources and tightens the writing |
| Story books | Keeps characters and art style consistent across every page |
| Infographics, mind maps, flowcharts | Checks for clipped, crowded, or overlapping panels before delivery |
| Standard video scenes | Previews the rendered slides and corrects visual problems before the video is encoded |
You do not switch any of this on. It is part of making the thing.
What It Actually Catches
The review is good at the failures that come from a first draft being written in one pass:
- A claim your source does not support. The most valuable catch, and the reason it runs against your sources rather than in the abstract.
- A topic that got skipped. Summaries plan a section per topic in your source and the review rewrites sections that came back thin.
- A wrong answer key. A quiz is worse than useless if it marks a right answer wrong.
- A slide that does not say anything. "Key Considerations" becomes the actual consideration.
- Narration that reads like writing. Written sentences sound wrong out loud; the edit fixes the rhythm before it is voiced.
- A visual that got cut off. Crowded panels and clipped text are caught before you see them.
What It Does Not Do
Worth being straight about:
- It cannot make a weak source strong. If your PDF is a scan with no readable text, or the wrong 40 pages, the review cannot invent what is not there. Getting the source right is still the highest-leverage thing you do. See Source Quality.
- It is not a substitute for reading the output. For anything that matters, something you will submit, present, or train people on, read it. The review raises the floor; it does not remove your judgment.
- It does not change what you explicitly asked for. If you set a length, a style, or supplied your own script, the review works within that rather than overriding it.
- It is not a second opinion on your material. It checks the creation against your sources, not your sources against the world.
Why Creations Take a Few Minutes
The honest answer is that the review is most of the time. A first draft is quick; checking it against your source, finding the gaps, and rewriting them is not.
While it runs, the progress page shows which stage it is on, reading, planning, narration, rendering, final check, so a long wait is legible rather than a spinner. You can close the tab; it runs on our servers and emails you when it is done. See Background Tasks.
Better Models, Better Review
Picking a stronger model improves the review as well as the draft. On the strongest options, the plan and the final edit are handled by the better model while the draft is written by a fast one, which is why a stronger model often shows up as better structure and fewer errors rather than flashier prose.
See Choosing an AI Model for what each model costs and where it helps most.
If Something Still Comes Out Wrong
It happens, and it is worth telling us, a thumbs-down on a chat answer, or feedback on a creation, goes to a person who looks into it.
- For a specific creation, use the feedback control on the item and say what was wrong.
- A Too short or not good enough verdict on a lecture offers a remake.
- A quiz answer key you think is wrong can be challenged from the answer review.
See Feedback and Bugs and When a Creation Fails.
Related
- Source Quality: what makes a source readable, and how to fix one that isn't.
- Choosing an AI Model: what each model costs and when to override Auto.
- Sources and Citations in Chat: checking where an answer came from.
- Creation Progress: what each stage means while you wait.
- When a Creation Fails: why a run failed and how your credit comes back.