GuidesPath

Student researcher

Build the source trail before you write a sentence, so every claim has a source you opened and dated.

Beginner
Lessons5
Time~85 min
Tools4 tools
Before you start
  • A topic you already have to write about
  • A free Zotero account, or any table where a source gets a checked date
0%0/5 steps

Workbench downloads

Free templates for the brief. Fill them before you trust an AI answer — no account required.

Path artifact

A one-page brief: question and exclusion rule, 6–8 dated source rows, three claims with caveats, methods note.

Finish every lesson and you should have: A one-page brief: question and exclusion rule, 6–8 dated source rows, three claims with caveats, methods note.

Frame the question

10 minPerplexity

Turn the assignment prompt into one question you could answer in a page, one boundary (years, country, population), and one claim you refuse to make. Ask Perplexity for three narrower versions and what each would exclude, then pick one yourself.

Pitfall
Opening a search box first. Ten minutes later the assignment has quietly become whatever the results page was good at.
Verification
You can reject a plausible-looking paper and name the rule it fails, in one sentence.
→ See playbook step 1 in detail

Collect ten candidates

20 minSemantic Scholar

Search Semantic Scholar for papers and reviews, then follow the reference list and the citing papers of the two best fits. Add one Perplexity pass for what papers do not cover: a new regulation, a national report, the institution running the pilot. Stop at ten rows.

Pitfall
Treating an answer engine's citation list as a reading list. Half of it is there because it ranked well, not because it fits your boundary.
Verification
Each row has a title, a DOI or URL, a source type, and a line on why it might belong.
→ See playbook step 2 in detail

Open and classify

20 minZotero

Open all ten and keep 6–8. In Zotero, give each survivor a type (primary paper, review, policy, news, vendor, forum), today's date as the checked date, and one line on what it measured and on whom. Delete the rest.

Pitfall
Counting a snippet or a cited answer as having read the source. The sample size and the limitation are never in the snippet.
Verification
You can point at any row and say what the study measured without opening the PDF again.
→ See playbook step 3 in detail

Map claims to sources

20 minNotebookLM

Write three claims the brief will make. Under each, name the strongest source, the weakest one you kept, and what the evidence does not cover. Load the saved sources into NotebookLM and make it quote the sentences where two of them disagree.

Pitfall
Letting a grounded answer settle a disagreement that the two papers never settled. Different samples do not average.
Verification
Every claim has a caveat someone can check against your source table alone.
→ See playbook step 4 in detail

Write the brief

15 minZotero

Write the page in your own words, with the source type visible next to each claim. Add a methods note naming which tool did which step and what you checked by opening the source. Then run the source table past a last completeness check.

Pitfall
Writing the confident version first and hoping the sources catch up. They rarely do, and the gap shows in the caveats.
Verification
Someone holding only the brief can rebuild the source trail and see where you stopped short.
→ See playbook step 5 in detail
Finish line

You should now have: A one-page brief: question and exclusion rule, 6–8 dated source rows, three claims with caveats, methods note.

  • You can reject a plausible-looking paper and name the rule it fails, in one sentence.
  • Each row has a title, a DOI or URL, a source type, and a line on why it might belong.
  • You can point at any row and say what the study measured without opening the PDF again.
  • Every claim has a caveat someone can check against your source table alone.
  • Someone holding only the brief can rebuild the source trail and see where you stopped short.