Research AI should show the method, not hide it.

This page explains how SixSentences_ approaches literature discovery, screening, evidence use and human oversight. It also states what the product cannot establish for you.

Explore literature search Request access

The method stays visible

Questions, search routes, criteria, source records and material decisions remain available for inspection instead of disappearing behind a generated answer.

Evidence stays addressable

A claim should lead back to a real work and, where full text is available, to the passage that supports or challenges it.

Researchers keep authority

AI can expand, rank, extract and draft. Researchers remain responsible for coverage, eligibility, interpretation, citation and publication.

The documented workflow

From question to inspectable output.

01
Frame the research task
A search begins with the research question and the constraints that make it answerable: concepts, population or domain, time range, languages, source types, known works and the intended review method.
For a structured review, eligibility criteria and reporting expectations should be recorded before screening. The agent can help clarify them, but it does not silently choose the method.
02
Retrieve through documented routes
SixSentences_ can search OpenAlex scholarly metadata and the live web, follow citation relationships, start from known papers and ingest records or full texts supplied by the researcher. Each candidate keeps its route and source information.
No single index is complete. Discipline-specific databases, institutional subscriptions, hand searching and librarian-designed strategies may still be necessary. The workspace can preserve imported results, but it does not claim universal coverage.
03
Resolve records before judging studies
Metadata from different routes can describe the same work differently. Candidate records are normalized and duplicates can be resolved while source provenance remains available for checking.
A title, abstract or generated summary is not the study itself. Where a decision requires full text, the paper must be obtained lawfully and reviewed at that level.
04
Screen with reasons, not only labels
AI can rank candidates and propose an inclusion or exclusion decision against stated criteria. The useful output is not a confidence badge alone: it is the record, criterion and reason a researcher can inspect and overturn.
Final study selection remains a human responsibility. Independent reviewers, conflict resolution or blinded screening may be required by the protocol and cannot be replaced by confident wording from a model.
05
Extract and synthesise against the source
Structured extraction should preserve the requested field, source work and supporting passage. Unavailable, ambiguous and conflicting evidence should remain explicit rather than being converted into a polished certainty.
Generated summaries and manuscript text are drafts. Authors verify every material claim, number, quotation and citation against the original source before using it in a research output.
06
Report what was actually done
The project record can support PRISMA-oriented reporting by retaining search routes, flow counts, criteria and decisions. It does not make a review automatically compliant with PRISMA, a journal policy or a disciplinary standard.
Researchers should disclose material AI assistance where their institution, journal, funder or field requires it and preserve enough information for another person to understand the workflow.

Boundaries

What the software does not promise.

These limits are part of the product method. A transparent boundary is more useful than a confident but unsupported claim.

Search coverage varies by field, index, language, date and access route.
Full text is attached only when openly available, supplied by the researcher or otherwise lawfully accessible.
A model can misclassify a study, misread a passage or produce an unsupported statement.
Citation metadata can refer to the wrong version, a correction or a retracted work and must be checked.
AI output is not a substitute for subject expertise, research ethics approval or statistical and methodological review.
Product features support documentation; they do not guarantee publication, completeness or a particular research outcome.

Method references