How to Annotate a Source to Verify Its Claims

Verification used to be something editors and librarians did. Now every reader needs a working version of it, because well-formed, authoritative-sounding text costs almost nothing to produce, and plenty of it is wrong. The good news is that the method is simple and the tools are the ones you already use to highlight. This guide shows how to annotate any article, report or web page so that every claim ends up either traced to evidence or clearly marked as unsupported.

The method

1. 1

   Read the whole piece once without marking

   Get the argument first. Marking on the first pass produces highlights on everything that sounds important, which is not the same as everything that needs checking.

2. 2

   Mark every checkable claim in yellow

   A checkable claim is a statement of fact that could be true or false: a number, a date, a quotation, a finding attributed to a study, a statement about what someone said or did. Opinions and predictions are not checkable in this sense; leave them unmarked.

3. 3

   Mark the evidence the text offers in green

   Next to each yellow claim, look for what supports it inside the text: a link, a citation, a named source, a figure. If there is none, write a small question mark in the margin or add a note saying "no source given".

4. 4

   Open each source and find the exact passage

   Follow every link and citation. Find the sentence in the source that the claim depends on, and check that it says what the text says it says. Make a highlight link to that sentence so you can return to it, and paste it into a note beside the claim.

5. 5

   Read laterally for anything unsupported

   For claims with no source, leave the page and search for the claim elsewhere: who else reports it, and where did they get it? This is lateral reading, and it is how professional fact checkers work.

6. 6

   Summarise in one line per claim

   At the end, each yellow claim should carry one of three notes: supported (with a link to the evidence), contradicted (with a link), or unverified. The summary is what you keep.

Lateral reading in practice: a worked example

Here is an invented claim walked through the method with a real primary source. Suppose a newsletter you are reading says: "The UK's population passed 70 million in 2023, according to official figures." It is a checkable claim (a number, a year and an implied source), so it gets a yellow mark. The text offers no link, so the margin gets a question mark.

Now leave the page. A search for the official mid-year population estimate leads to the Office for National Statistics bulletin "Population estimates for the UK, England, Wales, Scotland and Northern Ireland: mid-2023", released on 8 October 2024. Its first main point reads: "The UK population at mid-year 2023 was estimated to be 68.3 million, an increase of 1.0% since mid-2022." The claim in the newsletter is contradicted, and by a wide margin.

Two details on the ONS page belong in your note as well. The bulletin says it is not the latest release and points to a mid-2024 estimate, and it says the figures are subject to revision as international migration data improve. So the one-line verdict reads: contradicted; ONS mid-2023 estimate 68.3 million; released 8 October 2024; a later release exists. Paste the highlight link to the sentence beside it, and that claim is finished in a few minutes.

A colour code for verification

This is a variant of the general three-colour system in our study colour code, adapted for checking rather than learning. On paper, use two highlighters and a pencil. In a PDF, use the viewer's colours and notes (see highlighting in a PDF). On a web page, the bookmarklet gives you the yellow marks, and a notes file holds the links.

Why a highlight link is the best way to record evidence

A citation tells you where to look; a highlight link takes you there. When you record the evidence for a claim as a link that opens the source scrolled to the exact sentence, anyone reading your notes, including you in six months, can check your judgement in one click. The link stores nothing and needs no account, and if the source is later edited so the sentence changes, the link stops highlighting, which is itself a useful signal that the evidence moved. Our guide to linking to highlighted text covers how to build these links in every browser.

Red flags while annotating

• Citations that lead to a page that exists but does not contain the claimed passage.
• Quotations with no named speaker, date or place, or with a speaker who could not plausibly have said them. Our guide on tracing a quote to its source covers these.
• Precise-looking numbers ("73 percent of students") with no survey, year or sample named.
• Studies described by their conclusion but never by author, year or journal.
• Sources that all trace back to one original, so that five links are really one piece of evidence.

Expert tips from people who check claims for a living

• Check the number's unit before its size. Percent of what, per year or per month, and in which currency. A figure with the wrong denominator can be exactly right and still mislead.
• Give each claim a time limit. If you have not reached a primary source when the time is up, mark the claim unverified and move on. A stalled search is not a verdict, and a note saying "could not reach the source" is honest.
• Prefer the dull source. A statistics bulletin, a court record or a standards document beats a news summary of the same thing, because the summary is one more step away from the evidence.
• Write the note for a stranger. Assume someone will read your annotation without you in the room. Publisher, date, the sentence, the verdict.
• Keep the unsupported claims visible. A source with three question marks and no crosses is a source you can still use carefully, and the marks tell you which parts to leave out.

Three pitfalls undo the method. Marking whole paragraphs, when the mark should cover only the claim. Trusting a citation because it has a page number, when the page may not say what is claimed. Counting several outlets that copied one press release as several sources, when they are one. Each of these feels like checking and is not.

Text produced by AI tools

Everything above applies with extra force to text produced by chat assistants and writing tools, which can produce citations, quotations and statistics that look real and are not. The method does not change, but expect more question marks, and never treat a citation as checked until you have opened it and found the passage. Our walkthrough for annotating a chatbot answer applies the same steps to that case.

Your annotations as a checkable record

When you finish, the marked page is more than a set of verdicts. It is a record of what you read, which passages you relied on and what each one said on the day you checked it. Four things make that record checkable by someone else: the highlighted passage, the date you read it, a link to the exact line in the source, and your one-line note. Keep the four together, in the document or in a notes file, and the reading you did becomes something a teacher, an editor or a colleague can follow in minutes.

This is the idea behind keeping highlights as a record of what you read, and the same four fields are the columns of a newsroom-style verification log, which extends this guide with screenshots, archived copies and a rule for what to do when the source disagrees with the claim.

This guide sits in our collection of highlighting guides alongside the practical pieces it relies on: the PDF highlighting walkthrough for reports and papers, and marking passages on web pages for everything online.