Quick Start Guide: Evidence Table Builder

September 15, 2025•6 min read•ByGeorge Burchell•View publications on PubMed•ORCID
Laptop on desk next to notebook and coffee showing modern research workspace

Evidence Table Builder turns full-text research PDFs into an Excel evidence table you can check against the paper. Sign in and open Builder. Upload the PDFs, choose a starter template or add up to 10 extraction fields, press Extract Data, then press Download Excel. The Results sheet is one row per file. The Audit Trail sheet stores up to two verbatim quotes for each field, with the page and a location label. Review answers opens that page in the PDF and highlights the quoted words when the page has selectable text.

What the download contains

Download Excel saves results.xlsx with three sheets.

Results is the table you compare across studies. One row is one uploaded file. One column is one extraction field. The cell is the answer text.

Audit Trail is the record behind that cell. One row is one quote for one field on one file, and the sheet keeps at most two quotes per field. The columns include File Name, Section, Field, Question, AI Answer, Final Answer, Evidence #, Verbatim, Location Type, Location Label, Page, Confidence, not_found, ambiguous, needs_human_review, Verification status, Reviewer, and Reviewed at. The same record is described in Audit trails.

Verification notes explains the reviewer statuses in the workbook:

  • Unverified means a reviewer has not checked the answer against the PDF.
  • Verified means a reviewer confirmed the answer, and the final answer matches the original extraction.
  • Edited & verified means a reviewer corrected the answer after checking the PDF. Use the final answer.

A yellow highlight locates the quoted passage on the PDF. Verification is a separate status a reviewer sets with Verify answer, or with Save & verify after an edit. Download proof PDF, inside Review answers, exports a copy of the source with those quotes highlighted.

Evidence Table Builder is aligned with RAISE guidance: the draft speeds the first pass, and a person decides what stands. That commitment is on the responsible AI page.

Steps on the Builder page

How it works shows this sequence beside the separate quality-assessment workflow.

1. Upload PDFs

The upload area on Builder accepts PDF files. Drag them in or click to select. Each file must be 50MB or smaller, and one run accepts up to 20 PDFs. A file that is not a PDF, or a file over 50MB, is rejected. Builder also warns when the selection is larger than the credits you have left, because in Builder 1 credit processes 1 PDF. Allowances are on pricing.

A searchable PDF, where you can select the text, lets the viewer highlight the verbatim quote on the cited page. On a scan, the viewer can still open that page. When the page has no text layer, the viewer says so and the highlight can be unavailable. Matching a quote on a scan can take longer, and the viewer says when it could not highlight that quote.

2. Choose the fields

Builder asks how you want to set up extraction.

Use a template opens Starter Templates and My Templates. Starters include PICO Extraction, Study Characteristics, RCT Study Extraction, Observational Study Extraction, Outcomes and Results, Safety and Adverse Events, Diagnostic Test Accuracy, and Qualitative Study Extraction. A starter is read-only. The first edit saves a personal copy in My Templates. Extraction templates covers duplicate and archive.

Start from scratch lets you name the setup and write your own fields. The setup saves to My Templates once it has a field.

A run accepts at most 10 active fields. PICO Extraction already uses all 10. A wider schema needs another pass, or you turn fields off until 10 remain active. Each question can be up to 500 characters. Ask for one variable in each field, such as the population, the timepoint, or the arm. How to write those questions for a systematic literature review goes further. Copy the column shape from the evidence table example and template. Which columns earn a place is covered in what columns a systematic review table should include.

PICO Extraction already asks "What population was studied?", "What condition, disease, or health state was studied?", and "What intervention or exposure was evaluated?"

3. Extract and open the page

Extract Data runs the active fields on the uploaded files. In the preview, each file has Review answers. Each stored quote is its own control, labelled with the page and the location (a section or table name when the paper has one) plus the quote text. Choosing it moves the PDF to that page.

When the paper does not report the field, the preview shows "Not reported in paper". The flags ambiguous and needs_human_review mark answers that still need a person.

4. Download the workbook

Download Excel is the evidence table you keep. Results is the grid. Audit Trail is the quote, the page, and who verified the cell.

Practice on five published papers

Download the sample zip (five PDFs). The files are published papers on cognitive behavioral therapy for insomnia:

Upload them, then try Study Characteristics or Outcomes and Results, or add the Observational Study Extraction question "What sample size was included in the analysis?" Read sample sizes and outcomes from the PDF through Review answers.

Credits for a first table

The Free plan includes 3 credits to get started, and no credit card is required. In Builder, 1 credit processes 1 PDF, so those 3 credits cover 3 PDFs. Pro, Power, and top-up packs are on pricing.

Quality assessment is a different run, with its own credit estimate. How it works shows that workflow. How to choose a quality assessment tool compares the profiles. Planning the table itself starts in Getting started with evidence tables and Best practices for data extraction.

When something blocks a run

For account or run questions, contact via the contact page. Common setup questions are also on the FAQ.

Open Builder and run the first PDF on the Free allowance.

Tags:

quick starttutorialdata extractionsystematic reviewsguide
George Burchell

About the Author

Connect on LinkedIn

George Burchell

George Burchell is a specialist in systematic literature reviews and scientific evidence synthesis with significant expertise in integrating advanced AI technologies and automation tools into the research process. With over four years of consulting and practical experience, he has developed and led multiple projects focused on accelerating and refining the workflow for systematic reviews within medical and scientific research.

Systematic ReviewsEvidence SynthesisAI Research ToolsResearch Automation