Evidence synthesis

SixSentences_ turns a research question into a review workspace where retrieval, screening decisions, full texts and the final evidence base remain connected.

Check availability Explore the workspace
SixSentences_· Research, on the recordIn plain termsp. 02

In plain terms

  • 01

    Search beyond one query

    Combine academic retrieval, semantic expansion, citation snowballing, web sources and known works without losing the origin of each record.

  • 02

    Screen with evidence

    Apply eligibility criteria at title, abstract and full-text level. Decisions stay attached to records instead of disappearing into a chat transcript.

  • 03

    Keep the review inspectable

    Protocols, exclusions, retrieval runs, reviewer notes and included studies form one project record that can be revisited and exported.

SixSentences_· Research, on the recordA documented workflowp. 03
A documented workflow

01 / 04

Define the review

State the question, criteria, time window, sources and review method before retrieval begins.

02 / 04

Retrieve and screen

Let the agent expand the search and assist screening while every material decision remains visible for human review.

03 / 04

Read and synthesise

Open full texts next to the conversation, verify passages and carry included evidence into claims, tables and the manuscript.

04 / 04

Export the record

Preserve the evidence base, bibliography and screening history as an auditable research output.
Demo review · run fmvx3f8j2k
Question
Search string · ready for your methods appendix
"large language model"LLM"GPT""generative pre-trained transformer""Claude""Llama"
AND"systematic review""literature screening""title and abstract screening""citation screening""evidence synthesis"
ANDreliabilityagreementsensitivity"recall""work saved over sampling"accuracy
0
Identified
0
Duplicates removed
0
Screened
0
Excluded
0
Unsure
0
Included
Fig. 1The demo review “How reliable are large language models for systematic literature screening?”, run fmvx3f8j2k. Every tile is a real record; which works the 37 duplicates copy is illustrative, their count is real. Hover a paper to read its decision.
Fig. 1The demo review “How reliable are large language models for systematic literature screening?”, run fmvx3f8j2k. Every tile is a real record; which works the 37 duplicates copy is illustrative, their count is real. Hover a paper to read its decision.
Demo workspace01 / 02
SixSentences_ systematic review project showing the review question, study records and documented workflow in a wide research workspace
Fig. 2The demo workspace. The quick answer to “Which screening tasks do language models handle reliably, and where do they still fail?” cites page 7 of Oami et al. (2025), which opens with the passage highlighted. Page 7: Oami, Okada and Nakada, Research Synthesis Methods 16 (2025), doi:10.1017/rsm.2025.10014, © The Author(s), CC BY 4.0; the highlight is the app’s.
SixSentences_· Research, on the recordMethod and boundariesp. 04

Method and boundaries

01 / 02

Automation is useful when it reduces repetitive work without hiding methodological choices. SixSentences_ separates the review setup from the agent activity so that scope, eligibility rules and retrieval bounds stay explicit.

The agent can search iteratively, inspect sources and suggest decisions. It does not turn an uncertain abstract into a silently accepted fact. Researchers can inspect the underlying record, open the source passage and change a decision.

  • Protocol-aware retrieval and screening
  • PRISMA, Kitchenham, Cochrane, JBI and Campbell-oriented workflows
  • Known-work checks, duplicate handling and full-text acquisition
  • Project-level claims linked back to evidence

02 / 02

The review is not an isolated export. Included sources remain available in the library and can be linked to a manuscript project. This reduces the hand-off between search software, a reference manager, a PDF reader and the writing environment.

Human review still matters. Database coverage, copyright access, eligibility judgement and synthesis quality depend on the question and discipline. SixSentences_ is designed to make that judgement easier to audit, not to make it disappear.

SixSentences_· Research, on the recordQuestions, answered.p. 05

Evidence synthesis

  1. 01Does the AI make final inclusion decisions?

    It can assist and explain screening decisions, but the record remains reviewable and researchers retain control over inclusion, exclusion and synthesis.

  2. 02Can I use my own seed papers or database exports?

    Yes. Known works, full texts and reference exports can be added to the project so they remain part of the documented evidence base.

  3. 03Is the output automatically publication-ready?

    The workspace documents the process and prepares evidence-linked outputs. Reporting quality and compliance with a journal or review standard still require author review.

SixSentences_· Research, on the recordContinue through the workspacep. 06

Evidence synthesis

Fig. 3The seven workflows, this one lit. Each line joins a guide to one it continues to; hover a workflow to follow its lines.
Check availability Explore the workspace

hello@sixsentences.com Open source on GitHub