A draft you can
take apart.
Review Studio turns one research question into a grounded draft through six steps you can watch, approve and undo. Nothing appears on screen that a real run did not produce.
Where does a review look?
In 3 open catalogs. Discovery runs against OpenAlex, records are enriched from Semantic Scholar, and Crossref is used to verify identifiers before a source is trusted. All three are open, so you can look up the same record yourself.
The search plan is yours to read and edit first. It shows the query, the concepts the engine read out of your question, the screening criteria and the call budget, and no provider is contacted until you approve it.
- OpenAlex, Semantic Scholar and Crossref, named in the open.
- You edit the query if it missed your intent.
- Screening runs after the search, so an exclusion never hides a candidate before it was recorded.
How does a draft get built?
In 6 recorded steps that are written down as they happen. Each line in the ledger is a durable record on the server, with its own attempts and timestamps. A step with no record has not started, and there is no invented percentage anywhere on the screen.
What does the audit actually check?
The sentence against its source. A figure quoted in a claim has to appear in the source behind it. The wording has to overlap the passage it points at. A claim cannot say a treatment improved something when the source says it did not.
This is aimed at one specific failure: an invented statistic wearing a real citation. When a claim cannot be supported, the review says which claim, and the export stays closed until the draft is regenerated or the sentence is fixed.
- Numbers in a claim are matched against the source text.
- Direction and negation are compared, not just keywords.
- Unsupported claims are counted and named, never quietly dropped.
How is this different from just asking?
The difference is not how well the prose reads. It is what you can check afterwards, and what the tool refuses to hand you when it cannot back a sentence up.
- A search plan you read and edit before anything is calledA Review Studio runIncludedAn open chat sessionThe answer starts before you see a plan.
- Every step kept as a durable record with its own timestampsA Review Studio runIncludedAn open chat sessionA transcript, not a run you can reopen at a step.
- Candidates screened against criteria and a year range you setA Review Studio runIncludedAn open chat sessionWhatever the reply happens to bring up.
- Each claim checked back against the source behind itA Review Studio runIncludedAn open chat sessionYou verify the citations yourself, one by one.
- Export held closed while a claim stays unsupportedA Review Studio runIncludedAn open chat sessionNothing stops an unsupported sentence from being copied out.
None of this makes an open chat useless for early exploration. It makes it the wrong place to build a section you will have to defend.
Where do you get the last word?
At four points. You approve the search plan, you decide which sources are included, shortlisted or excluded, you approve the outline, and you rewrite the text afterwards.
Changing your mind is a first class action. Changing the source set reopens the review from screening, because an evidence matrix built on a different set would be a lie. Every text edit is saved as a new revision, and any earlier revision can be restored.
- Include, shortlist or exclude any source by hand.
- Every save is a revision, and revisions stay restorable.
- Each result records what wrote it: a language model from the approved outline, a deterministic composer, or you.
What happens if you close the tab?
The run continues. A review lives on the server, not in your browser, so you can leave it in the background and open it again from your saved reviews to watch the rest.
When an external catalog is busy the job pauses at that step and says so instead of failing quietly, and a step that fails keeps everything already finished. You resume rather than start over.
- Reviews are saved and reopened from a list.
- A busy provider pauses the run instead of ending it.
- Finished steps survive a failure further down the line.
Questions people actually ask
No, and the name misleads people, so here it is plainly: it drafts. It takes your research question, finds and screens the literature, and writes an Introduction and a Literature Review from what survived. It does not read or grade a manuscript you upload.
There is a monthly quota per plan, and it resets at the start of the next month. A real search and a real draft cost real provider work, which is what the quota reflects. The pricing page shows where the plans stand today.
You are told that it found nothing, and asked to broaden the question or the year range. An empty result is reported as empty rather than padded out with sources that do not answer your question.
English. Drafting in other languages is not open yet, so switching the interface language does not change the manuscript. The result screen states this on the draft itself rather than leaving you to discover it.
No. While the audit still has claims it could not verify against their sources, the export is closed. Regenerate the draft or fix the claims, re-audit, and the download opens.
Watch a draft get built.
Bring one research question and follow it through search, screening, outline and audit, one recorded step at a time.
Open Review Studio