How to Label Research Notes as Primary Sources, Commentary, or AI Output
Summary
- Label every note as one of three types: Primary Source, Commentary, or AI Output to prevent accidental misattribution.
- Separate original text (verbatim) from paraphrase and interpretation, and record which one you wrote.
- Track verification status (e.g., Unchecked, Checked, Conflicted) so you know what is safe to cite or publish.
- Use a reusable record format with fields for source details, quote boundaries, your claim, and what still needs checking.
- A worked example shows how to turn messy research notes into clean, auditable entries without inventing quotes or sources.
If you have a pile of research notes, it is easy to lose track of what came directly from a source, what was your own interpretation, and what was generated by an AI tool. That confusion can lead to incorrect citations, overstated claims, or accidentally treating AI text as evidence.
This guide gives you a practical labeling system (Primary Source, Commentary, AI Output), a simple verification-status scheme, and a reusable record format you can copy into your notes app. It also includes a worked example that demonstrates how to keep verbatim text, paraphrases, and interpretations clearly separated.
What the three labels mean (and what they do not)
1) Primary Source (evidence)
Use this label for: material that is directly attributable to a specific source you can point to again (a document, webpage, interview transcript, dataset, email, meeting recording, etc.).
Primary Source notes can include:
- Verbatim excerpts (exact words copied from the source) with clear quote boundaries.
- Factual observations about the source itself (e.g., “This table lists X and Y,” “The author defines term Z as ...”).
- Metadata needed to re-find the source (title, author, date, URL, page number, file name, timestamp).
Primary Source does not mean “true.” It means “traceable.” You still need verification status to indicate whether you checked it and whether it conflicts with other sources.
2) Commentary (your thinking)
Use this label for: your analysis, interpretation, synthesis, and planning. Commentary is where you connect dots, propose explanations, outline arguments, and decide what to do next.
Commentary can include:
- Your paraphrase of a source (in your own words).
- Your interpretation of what a quote implies (clearly marked as interpretation).
- Questions, hypotheses, and next steps (“Need to verify X,” “Find a counterexample”).
Commentary is not evidence. It can be correct and valuable, but it should not be cited as if it came from the source.
3) AI Output (machine-generated text)
Use this label for: any text produced by an AI system (summaries, rewrites, outlines, suggested claims, drafted paragraphs, extracted bullet points, etc.).
AI Output is not a source. It may be helpful for drafting or brainstorming, but it should not be treated as proof. If an AI output includes a claim you want to use, convert it into a Commentary task: “Verify this claim,” then locate and record a real Primary Source (or decide not to use it).
Distinguish original text, paraphrase, and interpretation
Even inside a single label, you should mark what kind of writing you captured. This prevents “quote drift,” where a paraphrase slowly turns into something that looks like a quote.
Verbatim (original text)
- Copy the text exactly.
- Put it in quotation marks or a dedicated “Verbatim excerpt” field.
- Record where it came from (URL/page/timestamp).
Paraphrase (your wording of what the source says)
- Write in your own words.
- Keep it close to the source meaning; do not add new claims.
- Link it to the specific excerpt or location you paraphrased.
Interpretation (your conclusion about what it means)
- State that it is interpretation (use an “Interpretation” field or prefix).
- Explain the reasoning that connects the source to your conclusion.
- Note assumptions and what would falsify the interpretation.
Add verification status so you know what is safe to use
Labels tell you what kind of note it is. Verification status tells you how confident you are that a claim is supported and accurately represented.
Use a small set of statuses you can apply consistently:
- Unchecked: captured but not reviewed for accuracy, context, or conflicts.
- Checked: you re-opened the source and confirmed the excerpt/summary matches the source.
- Conflicted: you found another source that disagrees, or the source itself is ambiguous.
- Needs context: the excerpt is real, but you need surrounding text/data to interpret it safely.
- Not usable: you cannot re-find the source, the excerpt is incomplete, or the claim is too uncertain for your purpose.
Verification status is especially important when you are working quickly, collaborating, or drafting content that will be published.
A reusable record format (copy/paste template)
Use one record per idea, excerpt, or claim. Keeping records small makes them easier to verify and reuse.
| Field | What to write |
|---|---|
| Record ID | A unique ID (e.g., RN-2026-10-17-001) |
| Label | Primary Source / Commentary / AI Output |
| Verification status | Unchecked / Checked / Conflicted / Needs context / Not usable |
| Source pointer | URL + access date, or document name + page, or recording + timestamp (leave blank for pure Commentary) |
| Content type | Verbatim / Paraphrase / Interpretation |
| Captured text | The excerpt (if verbatim) or your paraphrase (if paraphrase). Keep it short. |
| Your claim (optional) | The specific claim you want to make using this record (separate from the excerpt) |
| Context notes | What surrounds this excerpt, definitions used, scope limits, caveats |
| Open questions / next step | What you need to verify, find, or reconcile |
| Do not do | Any warning like “Do not quote as-is,” “Do not treat as evidence,” “Needs page number” |
Worked example: turning messy notes into clean labeled records
Below is a worked example using placeholder text to demonstrate the method without inventing real quotes or sources. Replace the placeholders with your actual source details and excerpts.
Step 1: Start with the messy note (what many people actually have)
- “The policy says users can request deletion within 30 days. Also it sounds like they keep backups longer. Need to mention this in the article.”
- “AI summary: They delete everything immediately and never keep backups.”
The problem: the first line mixes a claim, an interpretation (“sounds like”), and an action (“need to mention”). The second line is AI output that contradicts the first line and is stated too strongly.
Step 2: Create a Primary Source record for the exact excerpt (verbatim)
- Record ID: RN-2026-10-17-001
- Label: Primary Source
- Verification status: Unchecked
- Source pointer: [Document/Webpage Title], [URL or file name], section “[Section Name]”, accessed [YYYY-MM-DD]
- Content type: Verbatim
- Captured text: “[PASTE EXACT SENTENCE(S) HERE]”
- Context notes: “This appears under the ‘Deletion Requests’ section; check if it applies to all users or only certain accounts.”
- Open questions / next step: “Re-open source and confirm timeframe and any backup retention language nearby.”
- Do not do: “Do not paraphrase into stronger language until verified.”
Step 3: Create a Commentary record for your paraphrase (separate from the quote)
- Record ID: RN-2026-10-17-002
- Label: Commentary
- Verification status: Unchecked
- Source pointer: RN-2026-10-17-001
- Content type: Paraphrase
- Captured text: “The document describes a deletion-request window and separately references retention/backup handling that may extend beyond the request window.”
- Your claim (optional): “Deletion timelines and backup retention can be different; avoid implying immediate deletion.”
- Context notes: “Need exact wording to avoid overstating.”
- Open questions / next step: “Find the backup/retention section and capture verbatim lines.”
- Do not do: “Do not present as a direct quote.”
Step 4: Create an AI Output record (and downgrade it to a verification task)
- Record ID: RN-2026-10-17-003
- Label: AI Output
- Verification status: Unchecked
- Source pointer: (none)
- Content type: Interpretation
- Captured text: “AI-generated summary claims immediate deletion and no backups.”
- Your claim (optional): “This conflicts with RN-2026-10-17-002; treat as untrusted until verified.”
- Context notes: “AI output may be compressing nuance or inventing certainty.”
- Open questions / next step: “Verify against Primary Source excerpts; if not supported, discard.”
- Do not do: “Do not cite or publish this as factual.”
Step 5: Update verification status after checking
When you re-open the source and confirm the wording, update RN-2026-10-17-001 to Checked. If you find contradictory language elsewhere, mark the relevant records Conflicted and add a new Primary Source record for the conflicting excerpt.
Practical rules that prevent mislabeling
- One record, one job: keep each record focused on a single excerpt or claim.
- Never “clean up” a quote: if you fix grammar or remove words, it is no longer verbatim. Use paraphrase instead.
- Always store a source pointer with Primary Source records: if you cannot re-find it, you cannot responsibly rely on it.
- Keep AI Output quarantined: do not merge it into Commentary without explicitly marking what you changed and what still needs verification.
- Write interpretations as conditional: “This may imply...” plus what would confirm or refute it.
How to handle common edge cases
When you only have a screenshot or photo of text
Label the record as Primary Source if the screenshot is your captured evidence and you can re-open it. Your source pointer can be the file name plus where it came from (e.g., “Screenshot from [site/app], taken [date/time]”). If you later find the original page/document, add a new Primary Source record with the canonical pointer.
When you translated text
Keep two records:
- Primary Source (Verbatim): the original-language excerpt.
- Commentary (Paraphrase): your translation/paraphrase, with a note that it is a translation and may require review.
When you summarize a long section
Create:
- One Primary Source record with a short verbatim excerpt that anchors the section.
- One Commentary record with your summary, linked to the anchor record and any additional pointers (page range, section headings).
When a source is secondary (e.g., an article quoting another document)
You can still label your excerpt as Primary Source in the sense that it is traceable to the article you read. But add a context note: “This is a quote-of-a-quote; locate the original document before relying on it.” If you later find the original, create a new Primary Source record for the original and update your verification status.
Quick checklist before you cite or publish
- Is the supporting material labeled Primary Source (not Commentary or AI Output)?
- Is the excerpt verbatim if you plan to quote it?
- Does it have a source pointer that someone else (or future you) can follow?
- Is the verification status Checked (or clearly marked otherwise)?
- Have you separated what the source says from what you conclude?
Frequently Asked Questions
FAQ 1: What is the simplest way to label research notes without overcomplicating my system?
Answer: Use three labels only: Primary Source (traceable evidence), Commentary (your paraphrase/analysis), and AI Output (machine-generated text). Add a single verification status field so you can see what has been checked.
Takeaway: Three labels plus one status field is enough to prevent most attribution mistakes.
FAQ 2: How do I mark the difference between a paraphrase and an interpretation?
Answer: A paraphrase restates what the source says in your own words without adding new meaning. An interpretation adds your conclusion about what it means or implies. In your record, set “Content type” to Paraphrase or Interpretation and write interpretations as conditional statements with assumptions and next checks.
Takeaway: Paraphrase stays faithful to the source; interpretation adds your reasoning.
FAQ 3: Should AI-generated summaries ever be labeled as Primary Source?
Answer: No. AI-generated text should be labeled AI Output because it is not directly attributable to a specific source excerpt unless you separately capture and cite the underlying source material. If an AI summary contains a useful claim, treat it as a prompt to find and record a real Primary Source excerpt.
Takeaway: AI output can guide your search, but it is not evidence by itself.
FAQ 4: What verification statuses should I use if I want a minimal set?
Answer: A minimal set is: Unchecked (captured, not reviewed), Checked (confirmed against the source), and Conflicted (disagrees with another source or is ambiguous). You can add “Needs context” if you frequently capture short excerpts that require surrounding text.
Takeaway: Keep statuses few so you actually use them consistently.
FAQ 5: How do I handle notes when I cannot find the original source again?
Answer: Mark the verification status as Not usable (or keep it Unchecked with a prominent warning) and avoid citing it. If you still want to keep the idea, convert it into Commentary and add a next step to locate a new, traceable Primary Source that supports the claim.
Takeaway: If you cannot re-find it, treat it as a lead, not as citable evidence.
FAQ 6: How should I label a note that mixes a quote and my thoughts?
Answer: Split it into two records: a Primary Source record containing only the verbatim excerpt plus the source pointer, and a Commentary record containing your paraphrase/interpretation linked back to the Primary Source record ID. This keeps quote boundaries clean and prevents accidental misquoting.
Takeaway: Separate evidence from analysis by splitting mixed notes into two records.
FAQ 7: How do I label conflicting information from two different sources?
Answer: Create separate Primary Source records for each source excerpt and mark the relevant records Conflicted. Then add a Commentary record that explains the conflict, lists possible reasons (scope differences, dates, definitions), and sets a next step (find a more authoritative document, check publication dates, capture more context).
Takeaway: Keep each source intact, then document the conflict in Commentary.
FAQ 8: How do I keep my team consistent when multiple people take notes?
Answer: Agree on (1) the three labels, (2) a small verification-status list, and (3) a shared record template with required fields (at minimum: Label, Source pointer for Primary Source, Content type, Captured text, Verification status). Encourage people to split mixed notes into separate records and to avoid pasting AI output into Primary Source sections.
Takeaway: Consistency comes from a shared template and a small, enforced vocabulary.