← Back to library

Run a systematic review with an agent and checkpoints

Mushtaq Bilal, PhD (@MushtaqBilalPhD) described a SciSpace agent that carries out a full literature review from a single request. It proposes research questions using the PICO framework plus screening rules you can change, then runs the long mechanical part after those checkpoints.

  • OutcomeLet a research agent draft the questions and screening rules, edit both yourself, then let it search, screen, extract data and produce a PRISMA-style review you audit.
  • SetupUnder an hour
  • LevelIntermediate
  • CreditMushtaq Bilal, PhD (@MushtaqBilalPhD)
View original
Works with
Custom agent

Published Oct 23, 2026Updated Oct 23, 2026AI dry run on sample input: passed Oct 23, 2026Source verified Oct 24, 2026

When should you use it?

  • You need a first-pass systematic review for a thesis, grant or policy brief
  • Your topic fits a PICO-style question
  • You can spend time checking the screening log

Not for: Reviews you will publish without redoing the screening; Topics where most key papers sit behind paywalls you cannot access

What do you need, and what do you get?

You need

  • Topic and the population, intervention, comparison and outcome you care about
  • Inclusion and exclusion rules
  • Which databases to search and which years

You get

  • Edited research questions
  • Screening criteria
  • Included-paper table with a reason for each
  • Extracted data
  • Review draft plus a PRISMA-style chart of screening counts

How does it work, step by step?

  1. Ask for the protocolStart with a one-line request on your topic and let the agent propose PICO questions.
  2. Edit the questionsTighten the population, outcome and comparison so the scope matches what you actually need.
  3. Fix the screening rulesRewrite the draft include and exclude criteria; this is where most bad reviews go wrong.
  4. Pick sourcesChoose the databases and the date window before any search runs.
  5. Audit the outputCheck a sample of included and excluded papers against your rules, and read the full text of anything central.

The prompt

Our own prompt, inspired by Mushtaq Bilal, PhD (@MushtaqBilalPhD) and written to work in Custom agent. Values in lime are placeholders.

Fill in the blanks (optional; the copied prompt uses your values)

Prompt8 variables
Topic: {topic}. Population: {population}. Intervention or exposure: {intervention}. Comparison: {comparison}. Outcomes: {outcomes}.
Step 1 only: propose 2 to 3 PICO research questions and draft inclusion and exclusion criteria. Stop and wait for my edits.
After I approve, search {databases} for {date_range}, deduplicate, screen titles and abstracts, then full text. Log every exclusion with a reason, mark papers you could not get in full, extract {data_fields}, and give an included-paper table plus a PRISMA-style count of each stage.

Inspired by Mushtaq Bilal, PhD (@MushtaqBilalPhD): Systematic literature reviews take 12-18 months to complete. Looks like AI is going to fully automate systematic.... The prompt above is written by playbooks.bot in our own words; nothing from the original is quoted.

What can go wrong?

  • Paywalled full texts can drop out silently; check which included papers were screened on abstract only.
  • A PRISMA diagram looks rigorous even when the criteria were loose; the numbers are only as good as your rules.
  • Extracted figures can be misread; verify each one you cite against the paper.

Our take

This is the right shape for an automated review: the agent proposes, you edit at the two points that decide everything, and only then does the long mechanical part run. The questions and the screening rules are where bias enters, so keep those human. Never let the agent write the conclusions without your rewrite. Choose the databases and date window yourself before any search runs, and check a sample of included and excluded papers against your rules afterwards. Paywalled full texts can drop out silently, so note which papers were screened on abstract only. A neat PRISMA diagram can hide loose criteria; its numbers are only as good as your rules. Verify every extracted figure against the paper before citing it. For a published review, redo the screening properly. Rewriting the include and exclude criteria yourself is the step that matters most, and you should write the conclusions in your own words rather than adopting the agent's draft.

Where the time goes back: The protocol drafting, deduplication and first screening pass, the slowest parts of a review, arrive in one run you can check instead of build.

FAQ

Is this a publishable review?
Treat it as a strong first pass. Journals expect documented human screening and a registered protocol.
What limits it?
The poster names paywalled papers as the main gap, since the agent cannot read what it cannot download.

Source and credit

Inspired by Mushtaq Bilal, PhD (@MushtaqBilalPhD). We describe the idea in our own words, link to the original, and wrote the prompt ourselves; nothing is quoted.

Inspired by
Systematic literature reviews take 12-18 months to complete. Looks like AI is going to fully automate systematic...
Creator
Mushtaq Bilal, PhD (@MushtaqBilalPhD)
Published on
X
Original date
Apr 12, 2026
License
Not openly licensed (the poster keeps all rights), so we quote nothing and link to the original
Originally for
Custom agent
AI dry run
Oct 23, 2026: passed. Dry-run tested by our AI test agent on sample input. PASS: Gates held on desk check; no live sends, bookings or purchases.
AI dry run on sample data; not run with live connected accounts. The playbooks.bot test agent filled the prompt with synthetic inputs, produced the output and checked it against the page's stated outputs and guardrails.
Verified
Oct 24, 2026: source link resolves to the original, the credited creator published it, our excerpt or prompt matches it, and the license allows our use.

Are you the creator? Claim or correct this page.

More in Career & Learning

All 37 Career & Learning playbooks →

Related playbooks

Career & LearningRoutine✓ AI dry run

Copy homework due dates from a school portal to a calendar

Have an agent sign in to the school portal daily, pick out each assignment and its deadline, and put them on your kid's calendar along with an advance nudge.

Credit: fnordpiglet on Hacker News

Under an hour
Career & LearningWorkflow✓ AI dry run

Dictate ideas on a walk and get a shaped Obsidian note

Capture rambling voice notes on your phone and come back to a linked note with transcripts, a cleaned-up outline and a draft to react to.

Credit: Mete Polat

An afternoon+
Career & LearningWorkflow✓ AI dry run

Fix an annoying app behavior without editing its files

Point a coding agent at a small bug in software you use, and have it build a fix around the app, such as a hook or script, rather than modifying the vendor's files.

Credit: btx on Hacker News

Under an hour
Career & LearningWorkflow✓ AI dry run

Move a favorite browser add-on from Chrome to Safari

Point an agent at an extension already installed in Chrome, have it port the code to Safari's format and install it locally for your own use.

Credit: jedberg on Hacker News

Minutes