This page explains what our tools actually do, where they use AI and where they deliberately do not, and what they cannot tell you. It will grow as each tool is released.
Where we use AI, and where we don’t
A lot of research tools describe everything they do as “AI-powered.” We would rather tell you which parts are and which parts aren’t, because it changes how much you should trust each answer.
- Matching your citations against retraction records
- Looking up a DOI, journal, or publication record
- Checking whether a journal is currently indexed
- Gathering job listings from public sources
These are lookups against real databases. The answer is either in the record or it isn’t. Nothing is generated, so nothing can be invented.
- Reading a messy reference list and working out where each citation starts and ends
- Summarising a job posting in two lines
- Interpreting what a reviewer comment is asking for
- Drafting narrative text for you to rewrite
These involve judgement, and judgement can be wrong. Every one of them is marked in the interface, and every one is yours to check.
You will always be able to tell which is which
Anywhere a language model wrote something, it carries a green rule down the left-hand side and a small label. Like this:
A two-line summary of a job posting, a draft paragraph, or an interpretation of a reviewer’s comment would appear in a block like this one.
Text without that marking came from a database record or from us. We use the marking consistently, including in places where it would be more flattering not to.
Where our data comes from
- Retraction Watch Database, via Crossref Retractions, corrections, and expressions of concern. Distributed openly by Crossref and updated every working day.
- Crossref DOI records and publication metadata for most of the scholarly literature.
- PubMed and Europe PMC Biomedical and life sciences literature, including full text where it is openly available.
- OpenAlex and DOAJ Broader publication metadata and open access journal records.
- Public job listings Institutional and research-council feeds. We summarise and link to the original posting; we do not reproduce it.
We do not scrape Google Scholar, and we do not present paywalled full text we have no right to. Where a source has limits — and they all do — those limits become our limits, and we say so on the relevant tool.
What we never send to a language model
Your work is not training material. We do not use anything you give us to train models, ours or anyone else’s, and we do not sell it. Where a task genuinely needs a language model, only the text required for that task is sent, and it is not retained for training afterwards.
If you are peer reviewing someone else’s manuscript, please don’t upload it here or to any other AI system — the confidentiality you agreed to when you accepted the review covers that.
Three things our tools cannot do
- They cannot prove something is fine. A clean retraction check means we found no recorded problem — not that none exists. Absence of a record is not evidence of quality.
- They cannot replace your judgement. Screening decisions, extracted numbers, and draft text are a first pass for you to review. The responsibility for what appears in your paper stays with you.
- They cannot see what isn’t indexed. Grey literature, non-indexed journals, very recent publications, and work in languages our sources don’t cover will not appear.
This page will get more specific
As each tool is released, we will add a section here describing exactly what it does, which sources it uses, and where it is known to fail. If you find a failure we haven’t documented, tell us and we will add it.