Backed by Y CombinatorTwo documents and two sheets that rate every source against each of your research questions separately, never filed once by topic.
Free download · No account needed
What you get
Pack of 4 files: 2 documents, 2 spreadsheets
Annotated Bibliography Template by QuestionSource Register
| Method | Topic tag | RQ1: [Question 1] | RQ2: [Question 2] | RQ3: [Question 3] |
|---|---|---|---|---|
| [RCT / survey / case study] | [for browsing only] | [Central / Background / Not relevant] | [Central / Background / Not relevant] | [Central / Background / Not relevant] |
One row per source, one rating per research question, each with a one-phrase reason. The topic tag is for browsing and never substitutes for a question-specific rating.
Every annotated bibliography guide describes the same shape: a formatted citation, then one annotation paragraph that closes with a sentence on how the source fits your project. University library guidance names exactly three ways to organize the list that sentence sits inside: alphabetically, by date, or by topic. All three file a source once. A real research project rarely runs one question, and a source that answers your hardest question well can say nothing at all about the other two, which one relevance sentence and one topic tag both hide.
The pack ships filled in for a project on microlearning, short spaced lessons instead of one long session, and workplace retention. Three questions drive it: does microlearning beat a single session, what spacing interval works best, and does the gain show up on the job rather than only on a test. Nine sources, twenty-seven source-question ratings, and only nine of them, one in three, rate Central. Nine sources sort into eight different topic tags, and the two Central for the hardest question, transfer to the job, sit under tags naming an industry, not a method.
This space does not write the review itself; literature review synthesis picks up once your questions are answered and argues something from what you found. It does not extract structured fields from a study either, which the evidence extraction table does deeper. A relevance rating is not a quality rating, either: the critical appraisal pack rates a study's risk of bias once you have decided it belongs in your reading at all. If your library needs deduplicating first, start there.
Source Register4 of the 9 sources, chosen to show the pattern. RQ1 Format, RQ2 Spacing, RQ3 Transfer.
| Source | Topic tag | RQ1 | RQ2 | RQ3 |
|---|---|---|---|---|
| Renner 2022 | Microlearning design | Central | Background | Not relevant |
| Ferraro 2021 | Spaced repetition | Background | Central | Not relevant |
| Kowalski 2020 | Safety training case study | Central | Background | Central |
| Nakamura 2020 | Onboarding training | Central | Background | Central |
Both sources Central for RQ3 sit under topic tags, Safety training case study and Onboarding training, that share no word with microlearning or spaced repetition.
Citation ExportMatched to Source Register by source ID, in the citation style the project needs.
| Source ID | In-text citation | Used in draft |
|---|---|---|
| Renner 2022 | (Renner, 2022) | No |
| Kowalski 2020 | (Kowalski, 2020) | No |
| Nakamura 2020 | (Nakamura & Silva, 2020) | No |
A source rated Central is not the same as a source already cited. This column tracks the second fact, checked against the actual draft, not assumed from the rating.
The full three-part annotation per source: a summary specific enough to skip rereading the abstract, an evaluation naming what kind of evidence it is, and a topic tag for browsing.
One row per source, one Central, Background or Not relevant rating per research question, each with the one-phrase reason a rating alone cannot carry.
Reference-ready citations matched to Source Register by source ID, with a column for whether a source has actually made it into the current draft yet.
The 120-180 word annotation format, and the rule that a relevance rating is never written as one blended sentence at the end of it.
The standing space rule every prompt reads first, which is why Source Register keeps one column per research question instead of one relevance sentence per source.
The two or three specific research questions this bibliography needs to answer, not the general topic, plus the sources or citation library you are starting from.
Summary, evaluation and a topic tag in Reading Notes, from the full text rather than the abstract, before any relevance judgment gets made.
Central, Background or Not relevant against every research question separately in Source Register, with the one-phrase reason for each rating.
Citation Export generates the reference-ready form and tracks which sources have actually made it into the current draft, not just which ones are rated Central.
Yes. Download the whole pack as Word documents and CSV sheets with no signup and no credit card. Edit with AI is a separate, optional path that has the agent write the annotations and ratings with you. The template library holds the rest of the packs.
Word documents for the two write-ups and CSV for the two sheets, in one zip. They open natively in Word, Pages, Google Docs, Excel, Numbers and Sheets with nothing to convert. Inside River the same content opens as live Docs and Sheets.
A normal one files each source once, by topic, date or author, and closes the annotation with one sentence on how it fits your project. This pack keeps the annotation and adds a rating per research question instead of one blended sentence, so a source central to your hardest question does not disappear under an unrelated topic tag.
It drafts them from the source, but a summary or evaluation it writes should be checked against the actual text, the same as a draft you'd write yourself. What only you can set are the research questions: River rates each source against whatever questions you give it, and a question you never stated cannot be rated against.
Then Source Register needs only one relevance column. It still beats a single relevance sentence, because Central, Background and Not relevant are checkable ratings with a stated reason, not a sentence someone has to reread the source to verify. Add a second question later and every source gets rated against it without redoing the first column.
Yes, and it is normal to. Add the column to Source Register, then go back through sources already read and rate each one against the new question specifically, rather than assuming an old rating for a different question still applies. A source read for Format can turn out Central for a Spacing question nobody had written down yet.
Written by
Co-founder & CEO, River
Chandler has spent the last six years working in AI, building tools for millions of users and Fortune 100 companies. In 2025, he co-founded River to build a new kind of operating system for AI agents.
Send your two or three research questions and the sources you are starting from. Each one gets annotated, then rated against every question separately.
Edit with AI