📚 PaperTrail Verify Citations  ·  From 2 Tokens per essay

The sources are real.
Did they say that?

Modern misuse rarely invents whole sources — it invents what they said. Verify Citations checks every citation in an essay against the source itself: does the bibliography match the in-text citations, does the source exist, and does it actually say what the essay claims it says. Facts and questions, never a verdict.

Grounded in live web search — never the model's memory Quoted passages searched in the actual source text What the essay says vs. what the source says, side by side A suggested question for each finding — the conversation, ready "Not located" is a question, not a finding — no verdict, ever
📚 PaperTrail Verify Citations Citation 1 of 11

Every citation walks the same ladder — quoted passages first,
then figures, then the essay's characterization of the source.

Workflow

One essay in. Every citation checked.

Four steps from a bibliography you have doubts about to a documented, per-citation comparison — inside the extension you already use, or from any browser.

Free
1

Send the essay

Fill from the open Google Doc, upload a file, or paste — bibliography included. Every citation is extracted along with the exact claims and quotes the essay attaches to it.

Parsing is free. No citations found → no charge, no report.

2

Cross-match

Before any searching, the bibliography and in-text citations are matched both directions: sources cited but never referenced, references never cited, and format observations.

You see the citation count and choose before any tokens are spent.

2 tokens
3

Check the sources

Live web search — journal databases, publisher pages, Google Books — locates each source and compares what the essay says it says against what it actually says, quoting both sides.

2 tokens covers 25 citations; +1 token per additional 25.

4

Read the report

Each citation gets a side-by-side card, a plain-language summary, and a suggested question for the student conversation. Factual counts up top — no aggregate score.

Ready in two to four minutes. Cross-match findings are yours even if a search fails.

Three Checks, One Report

Existence is table stakes. Claim-vs-source is the point.

A fabricated bibliography is rare. A real bibliography attached to invented specifics — a quote the source never contains, a figure it never reports, a conclusion it explicitly disclaims — is what actually shows up on a teacher's desk.

Check 1 · Deterministic

🔗 Cross-match

In-text citations matched against the works-cited list, both directions. Sources cited but never referenced. References that never appear in the text. Format sanity, stated as observations.

"Three works-cited entries are never cited in the text."

Check 2 · The paid value

⚖️ Claim vs. source

Does the source exist — and does it say what the essay claims? Quoted passages are searched in the reachable source text; figures are matched exactly; paraphrases are compared against the source's own words, both sides quoted.

"This quoted passage was not found in the source text."

Check 3 · Questions only

🔢 Data plausibility

Numbers and claims attributed to sources, sanity-checked for accuracy and plausibility. The most judgment-laden check, so it is phrased only as questions — never as findings.

"This 87% figure is attributed to a single-participant case study — worth asking about."

The Citations Report

Both sides quoted.
No verdict.

Every citation gets two factual reads — was the source located, and does it say what the essay claims — with the evidence basis stated on each. Paywalled and print-only sources fail lookup routinely, so the language is built to stay honest about what a search can and can't see.

✓ Located · Consistent

The source was found and the reachable text supports what the essay attributes to it. The evidence basis is stated — full text, abstract, snippet, or publisher description.

◐ Notable difference

The reachable source text differs from the essay's characterization — both sides quoted so you can judge. A quote not found in fully-reachable source text is the report's strongest finding.

◌ Access limited · Couldn't check

The source sits behind a paywall or in print, so the claim couldn't be compared. Stated as a fact about access — never as suspicion.

? Not located

Search couldn't find the source. Print-only and regional-database sources fail lookup routinely — which is why this status opens a conversation instead of closing one.

The rule the whole report is written around: "Not located" is a question, not a finding.

📚 PaperTrail Verify Citations — Citations Report
Daniel Osei (3117-B)  ·  "Does Observation Produce Reliable Knowledge?"  ·  11 citations checked: 6 located · 3 access limited · 2 not located  ·  Mar 14, 2026
Baird, J., & Norton, K. (2021). Attention and the observing mind. Journal of Cognitive Education, 44(2).
✓ Located — journal pageConsistent (abstract)
The essay's summary of the attention-filtering study matches the abstract's own description of its method and findings.
Whitfield, R. (2019). The Empirical Habit. Halloway Press.
✓ Located — full text via Google Books◐ Notable difference
What the essay says

Quotes Whitfield: "observation is the engine of all discovery, and the senses its only fuel."

What the source says

The quoted passage was not found in the source text. Ch. 2 argues the opposite: observation is "always theory-laden, never raw."

The quotation could not be found in the book, and the reachable text argues against the position the essay attributes to it.
Worth asking: where in the book did they find this quotation — page or chapter should be quick to name.
Tanaka, M. (2020). Measurement and inference in field biology. [Print monograph]
◌ Access limited — print onlyCouldn't check
The source appears in library catalogs but its text is not reachable online. The essay's 62% figure attributed to it could not be compared. Stated as a fact about access, not a finding.
This report describes how the essay's citations compare against reachable source text. It is not a determination of authorship or integrity, and it is not suitable as the sole basis for any academic decision. "Not located" is a question, not a finding. Educator judgment required.

Editable header for your records. Factual status counts up top —
no aggregate score, because the citations don't get a grade. The student conversation does.

Honest Limitations

What Verify Citations can't tell you

Like every PaperTrail tool, this is a structured, repeatable data point — not a verdict. Three things to keep in mind before you weigh one.

🔒 The web has walls

Paywalled journals, print monographs, and regional databases are routinely unreachable. A legitimate source a student found in the school library can come back "Not located" or "Access limited."

That's why those statuses are worded as facts about access — and why the report never says "fabricated." Ask the student to show you the source; a real one is easy to produce.

📄 Evidence has degrees

"Consistent with the abstract" is weaker evidence than "quote found on page 43." A paraphrase judged against a publisher's description is weaker still.

Every claim read states its evidence basis — full text, abstract, snippet, or description — so you always know how much weight it can bear.

⚖️ Difference is not deception

A "Notable difference" can be sloppy paraphrase, an honest misreading, a secondary source quoted secondhand — or something worth a longer conversation. The report can't tell you which.

Each finding comes with a suggested question. A student who genuinely read the source answers it in seconds — that answer is the real data point.
🔎

Search-grounded

Findings come from live web search with sources quoted — never from AI memory, which confidently confirms plausible fakes.

📤

Text sent only to analyze

The essay is transmitted securely for analysis (Anthropic) and is not retained after the report is generated.

🗑️

You control the report

Reports are cached on your device; the server copy can be cleared after delivery. Nothing is used to train AI models.

🛡️

Charged only on results

Parsing is free. If the search phase fails, your tokens are refunded automatically — you keep the cross-match findings.

Two tokens.
Every citation checked.

Verify Citations runs on PaperTrail Tokens — the same wallet as Verify and Oral Defense. Two tokens covers an essay with up to 25 citations; longer bibliographies add one token per additional 25, and you choose before anything is spent.

A token is $0.60 in the starter pack, less in bigger packs — a full citations check on a typical essay costs about the price of a photocopy run.

2 tokens / essay — from $1.20
Covers 25 citations  ·  +1 token per additional 25  ·  parsing free
  • Starter — 10 tokens · $6 (5 citation checks)
  • Semester — 35 tokens · $18
  • Year Supply — 100 tokens · $45
  • Shared wallet with Verify (1 token) and Oral Defense (2 tokens)
Buy PaperTrail Tokens

"Where did this quote come from?" — now with the receipts.

Install the extension and run Verify Citations from the PaperTrail panel, or paste an essay into the teacher web app from any browser.

⬇ Add PaperTrail to Chrome Open the teacher web app →
From 2 tokens per essay Grounded in live web search No verdict — educator judgment, always