Transparency
About our AI
The AI features switched on at Scholars Central, what each one does and where to find it. Text written by AI is labelled as such wherever it appears, and the text an editor has checked says so too.
The tools in use
Ask about this article
Ask a question on an article page and get an answer from that article's title, abstract and keywords only, each sentence pointing to where it comes from.
Not checked by an editor: labelled wherever it appears.
Explain in plain words
Select words in an article's title or abstract, or choose a sentence, to have it explained in everyday language.
Not checked by an editor: labelled wherever it appears.
Compare these studies
Tick 2 to 6 saved articles to see their populations, designs, interventions, outcomes, findings and limitations side by side, each cell taken from one sentence of that article's abstract.
Not checked by an editor: labelled wherever it appears.
Suggested order in the journal finder
When you ask for it, the journals that match your text by words are put in order of topic, with one sentence on the shared topic. Every match stays listed.
Not checked by an editor: labelled wherever it appears.
Check your abstract
Checks of length, structure and wording on a draft abstract, with written advice when you choose it. Advice only: it never predicts acceptance.
Not checked by an editor: labelled wherever it appears.
Wording signals in Check your abstract
Part of Check your abstract: rules on our own server point to wording that research-integrity checks look at. No AI model is used, and nothing is sent or kept.
No model is used: the rules run on our own server.
Ask a research question
A written answer to a research question from our abstracts, every sentence citing the abstract it comes from.
Not checked by an editor: labelled wherever it appears.
Article summaries
One-line and plain-language summaries of articles, drafted by AI and published only after an editor checks them.
Checked by an editor before it appears.
Translated abstracts
Abstracts translated into Hindi, Tamil, Telugu, Bengali, Marathi, Spanish, French, Chinese (Simplified) and Arabic, drafted by AI and published only after an editor who reads the language checks them. The English record stays authoritative.
Checked by an editor before it appears.
Session summaries
Summaries of past conference sessions, made by AI from the talks' abstracts (never from recordings) and published only after an editor checks them.
Checked by an editor before it appears.
Order of your recommended conferences
On your own conferences page, the conferences chosen for you from what you saved and follow are put in order by AI, which may also write each row's one-line description. The rule that chose a row is always printed as well, and you can switch the whole section off.
Not checked by an editor: labelled wherever it appears.
What each tool sends and keeps
Our provider is OpenAI, through its API. Nothing is sent until you ask for it, and what you type is not used to train a model.
| Tool | Sends to OpenAI | Kept |
|---|---|---|
| Ask about this article | your question and that article's title, abstract and keywords | counts and timings only |
| Explain in plain words | the words you select and that article's title and abstract | counts and timings only |
| Compare these studies | the titles and abstracts of the 2 to 6 articles you choose | counts and timings only |
| Suggested order in the journal finder | your title and abstract, and the titles and subjects of up to 15 journals | counts and timings only |
| Check your abstract | the title and abstract you paste, only when you choose AI advice | counts and timings only |
| Wording signals in Check your abstract | nothing: the rules run on our own server | nothing |
| Ask a research question | your question and up to 6 abstracts | counts and timings only |
| Article summaries | the article's title, type, journal and abstract | published summaries, with the model, the prompt version and the editor who checked them |
| Translated abstracts | the article's title, type and abstract, and the language to translate into | published translations, with the model, the prompt version and the editor who checked them |
| Session summaries | the conference's title, the session's track and day, and its talks' titles and abstracts | published session summaries, with the model, the prompt version and the editor who checked them |
| Order of your recommended conferences | the titles, subjects, city, country, month, year and series of up to 12 conferences we already chose for you | counts and timings only |
How we check them
What has been measured so far, in the words of the measurement itself. Where something has not been measured yet, it says so.
Citations point at real text
The share of cited sentences found word for word where the answer says they are, on fixed probe questions over real articles. It shows where a citation points, not that the text supports the answer.
100% of cited sentences were found word for word where the answer said (this shows where a citation points, not that it supports the answer)
Questions declined
How often fixed questions that our articles cannot answer, or that are not about research, are declined instead of answered.
100% of questions the articles cannot answer were declined
- Accuracy on questions with known answers — Accuracy against a gold set: not measured yet (needs editor-written questions).
- What editors did with AI summaries — Not measured yet
- Journal finder ordering — Not measured yet
- What editors did with AI translations — Not measured yet
- What editors did with AI session summaries — Not measured yet
- Answers graded by a person — Whether answers are supported by what they cite: not graded yet (a person grades a sample of real answers).
Limits
From one connection, each day: 60 research questions, 20 comparisons, 100 explanations and 60 questions about articles.
Two features that are not AI
Both match words. Neither writes text.
The journal finder
It compares the words of your title and abstract with the articles each journal has published. Only if you ask for a suggested order is AI used, to put the matches in order of topic.
Try the journal finderSimilar articles
Article pages list other articles on the same topic, found by the words of their titles and keywords. It is a match by topic, not a record of what other readers read.

