AI that earns its place in professional work.

A mid-sized review means two people reading four thousand abstracts against the same criteria for weeks. Consistency drifts, disagreement rates climb, and the audit trail is a spreadsheet of initials.

Feed it your protocol and the records; it returns an IN / OUT / UNCLEAR verdict for each with the passage that justifies the call, so a second reviewer checks reasoning instead of re-reading everything. Runs on Anthropic, OpenAI, or Google models — never an unauditable black box.

Full-text eligibility is where reviews quietly break: two hundred PDFs, a tiring reviewer, and a criterion applied differently on day one and day nine. Six months later nobody can reconstruct why a study was excluded.

Retrieval-augmented appraisal that reads past the abstract and attaches the supporting text to every eligibility decision. The exclusion log becomes evidence rather than recollection — which is what a journal editor or a replication attempt will actually ask for.

Six databases return three thousand records with heavy overlap. Manual deduplication misses title variants and quietly inflates the numbers in your PRISMA diagram — a figure reviewers do check.

Merges every export, matches on DOI, title, and author–year, and produces a clean set plus a log of exactly what was removed and why. The screening count you report becomes reproducible instead of approximate. Web and desktop builds.

An early-career researcher gets a flattering invitation from a journal they have never heard of. Judging whether it is legitimate means cross-checking several lists — and the cost of getting it wrong lands on their record permanently.

Search by name or ISSN for a risk profile assembled from publication growth, authorship concentration, and title warning lists. It surfaces the signals worth a second look in seconds, and stops short of a verdict it has no standing to give.

A Scholar results page gives you a hundred titles and citation counts, and nothing about the venue. Checking each journal’s quartile by hand costs minutes per paper — on papers you may decide not to read at all.

Renders Scopus quartiles, SCImago indicators, and Web of Science indexing badges directly on the results page. The quality signal arrives while you are triaging, not three tabs later. Free on the Chrome Web Store.

Business and management academics are assessed against the ABDC list, but Scholar does not show it. Finding which results sit in an A* venue means a separate manual lookup for every single hit.

Overlays the A*, A, B, or C rating from the ABDC Journal Quality List straight onto Google Scholar, with a count of each tier at the top. Reading priorities and submission targets resolve in one glance.

AI-assisted work is entering the literature with no standard way to declare how AI was used, while “peer reviewed” is asked to carry more weight than one label can bear. Readers cannot tell an approved record from a scrutinised one.

A disclosure-first publishing platform where approval and community verification are deliberately separate. Every work gets a persistent identifier; verification is attributed and carries the verifier’s expertise, basis, and declared conflicts. Self-hosted, so no vendor owns the record.

Teachers and researchers meeting generative AI in their own classrooms get vendor webinars on one side and institutional prohibition on the other — very little written by someone who reads the research and has nothing to sell them.

An independent, scholar-led publication and toolmaker, founded and run solo. It turns research on generative AI in education and applied linguistics into explainers, critical commentary, and the free utilities listed on this page.

Small programmes and independent educators still need enrolment, attendance, and assessment — but commercial LMS licensing and its administrative overhead are priced and scoped for institutions ten times their size.

Courses and lessons, enrolment by instructor code, attendance check-ins, and assessments that auto-score multiple choice while keeping open-ended responses under human marking. Runs on ordinary shared hosting.

Timed protocols — classroom observations, think-alouds, structured writing blocks — need a countdown you can read at a glance, not a productivity app that builds a profile of you and asks for an account first.

An analogue countdown whose red sector grows as the session is spent, with an independent stopwatch for timing tasks and observations. Deliberately unquantified: it shows you the time and keeps no record of you.

A researcher’s task list routinely holds unpublished study titles, participant details, and embargoed dates — material that has no business sitting on a task-management vendor’s servers under someone else’s privacy policy.

Dated tasks with due dates, priorities, and progress, held entirely in your own browser’s storage. Nothing is transmitted, no account is offered, and none is possible. Proof that a useful tool need not become a data relationship.

You need one quick figure mid-analysis, and the fastest route is a search box that logs the query or an application three clicks and a loading screen away.

Full keyboard support and a running tape of previous results, so you can see how you got to the number. Not everything needs to be intelligent — some things just need to work.