How to cite a source in an article

Citing sources means naming the document a sentence depends on, close enough to that sentence that a reader can see the dependence. Outbound links are how they open it. Google’s notes on links that Google can crawl, updated 10 December 2025, say linking to other sites is not something to be afraid of, and that citing sources is one way a link helps a reader trust the page. The same note says to use nofollow when you do not trust the destination, not on every external link. A quota of outbound links is not the practice those pages describe.

What citing sources means

A blog citation is a pointer a person can follow. It is not an academic bibliography, and it is not a decoration that makes a paragraph look researched. The reader should be able to answer three questions without leaving the paragraph: what you are claiming, who said the part you did not establish yourself, and where to read that part.

Some sentences need no citation because they are yours. “On our team, the standup is three sentences in a thread” is a report of your own practice. A citation there would be costume. Some sentences need one because they borrow a rule, a date, a definition, or a measurement. “Google treats a redirect as a strong canonical signal” is that kind of sentence. The signal is Google’s description, on a page you can link. Your next sentence, about what your site should do with an old slug, is the part you add.

Helpful-content guidance asks whether a page that draws on other sources avoids simply copying or rewriting them, and whether it provides something of its own. It also says clear sourcing is one of the ways a page can present information so that a person is willing to trust it, alongside a byline and a way to learn who published the site. Sourcing and originality are both required when you use someone else’s work. The author page is who will stand behind the result. The citation is which document the result depends on.

Cite the document that states the fact

The primary document is the one that establishes the fact: the vendor’s own documentation, the standard, the dataset, the study, the law, or your own measurement. A recap is someone else’s paragraph about that document. Recaps are useful when you are learning. They are a weak thing to cite, because you are now depending on their paraphrase.

Search writing has a particular version of this trap. A newsletter says a tool blog said an expert said Google confirmed a ranking factor. By the time you cite the newsletter, the sentence may not match the documentation. If an answer engine quotes you, it quotes the chain. Name Google Search Central, or whichever body actually published the rule, and link that page. If you disagree with a recap, you can mention the recap and explain the difference, and you still link the document you are comparing it with.

When the fact is yours, you are the primary source. A support thread you handled, a price on your pricing page, a limit in your product, a count from a report you ran: say that it is yours and say how you know. “We looked at the last 40 standups in the Northwind workspace” is a method. “Studies show” with no study is the sentence the brief tells you not to invent. Do not cite a roundup to launder a number you do not have.

Prefer the stable URL of the document, not a copy on a scraper and not a social post that screenshotted it. If the document is a PDF, link the PDF. If the publisher offers an HTML version and a PDF with the same text, link the one a reader can open without a download unless the PDF is the version of record. Use the publisher’s preferred URL if they state one. Sending someone to a tracking-parameter copy of their docs is the same mess described in the canonical tag guide, only now it is their URL.

Put the citation beside the claim

A citation works when it sits in the passage that needs it. A list of URLs at the bottom of a 2,000-word article does not tell anyone which sentence depended on which URL. Answer engines have the same problem. They extract a passage. If the source is named inside that passage, the extracted answer can keep the attribution. If the source is a footnote marker, the extracted answer is an unsupported sentence with a number.

You do not need a citation on every sentence in a section. One well-placed reference can cover a short series of sentences that all come from the same page, if the reader can see the scope. “The rest of this paragraph is from that same note” is enough. A new claim from a different document needs its own name. Mixing two sources in one sentence without saying which part came from which is how a reader, and a model, glues them together.

Quotes stay short and stay accurate. Copy the words that carry the rule, not a page of someone else’s explanation. If you need their whole section, link it and summarize the decision you made. A long paste asks the reader to do the reading on your URL, and it is the copying problem the helpful-content questions describe. Put their words in quotation marks when they are their words. A paraphrase should still be checkable against the link, and it should not be stronger than the original. “Google requires a canonical tag on every page” is not a paraphrase of a document that says specifying a preference is generally not critical.

What the sentence should include

Four pieces are enough for a public web page. You rarely need a formatted bibliography style.

The date can be the publisher’s “last updated” line when the page has one. Write it as a date, not as “Google recently said,” because recently rots and the page may still be online with an older rule. If you checked the page on a particular day and it has no updated line, say when you read it. On your own article, when that source later changes, the refresh is how you revisit the sentence. A citation to a dead URL is a claim with the evidence removed. Update the link or remove the claim.

Anchor text should say what the reader will open. “Google’s duplicate-URL documentation” is an anchor. “Click here,” “source,” and a stuffed keyword are not. The same rule as any other internal or external link: the words are a label for the destination. If the destination is a specific section, link that section when the publisher’s URL supports it, and still name the document.

A formal reference list is optional. Use one when the article cites many documents and a reader may want to scan them, and make each item in the list correspond to a mention in the text. A list that includes pages you never used, added because they look authoritative, is the quota problem again. It also gives an answer engine extra URLs to treat as your sources.

Outbound links, nofollow, and sponsored

A normal link to a source needs no rel attribute. Google’s page on qualifying outbound links says that for links you expect Google to fetch and parse without qualifications, you add nothing. The crawlable-links guide uses citing sources as the example of a good external link, and it asks you to give the reader context about what they will find. A robots meta tag of nofollow is a different instruction: it applies to every link on the page, not to the one citation.

Three qualifications cover the cases that are not a normal source:

• rel="sponsored" for an advertisement or a paid placement. If you were paid for the link, or the placement is an ad, mark it. nofollow is still accepted for this and sponsored is the value Google prefers.
• rel="ugc" for links your visitors insert, such as comments or forum posts. You can drop it for contributors who have earned that trust over time. A citation you chose while editing is not user-generated content.
• rel="nofollow" when the other two do not fit and you do not want Google to associate your site with the destination, or to crawl it from your page. The crawlable-links example is a page you are criticizing and do not want to vouch for. It is not a default on every domain that is not yours.

Links with those attributes are generally not followed. The other site may still be crawled because someone else linked to it or because it has a sitemap. Qualifying a link does not make the page disappear, and it does not make a paid link acceptable. It tells Google the relationship. Nofollowing a source you trust, because you are afraid of “leaking” something, contradicts the reason to cite it. You wanted the reader, and Google, to connect your sentence to that document.

Open outbound links in the same tab unless you have a product-interface reason not to. target="_blank" is a usability choice, not a ranking one. If you use it, rel="noopener" is a browser safety habit. It is not a substitute for sponsored or nofollow. Do not hide links with JavaScript click handlers or with text that only looks like a link. Google’s link guidance is about real a elements with an href that a crawler and a reader can both use.

Credit is not the original work

Google’s helpful-content page describes effort as the human work that went into the content or into the systems behind it. Writing an analysis, translating carefully, or building a tool counts. Generating a large amount of text with a model and publishing it without review does not. The same section says attribution does not replace the need for original effort. The helpful content guide is the rest of that self-check. A paragraph that is only a credited rewrite of the top results is still a rewrite. The link makes you honest about the debt. It does not pay it.

What counts as the extra work is specific to the reader you have. For a product team that already read Google’s canonical documentation, the extra work is the decision: which of their URLs are the same document, which old slug should redirect, and which article must not be canonicalized into another. That decision is not in Google’s document. Restating the document with a link at the end does not add it. The answer engine guide makes the same point as information gain: a page that only rephrases sources other people have already seen gives an answer engine nothing new to quote.

Originality does not mean refusing to cite. A page with no sources, on a topic that depends on someone else’s rules, asks the reader to trust you for facts you did not establish. Cite the rule, then add the decision, the example, or the limit you learned by applying it. Accuracy is the other half. If the topic can affect someone’s health, money, or safety, the helpful-content notes ask for a higher bar: the page should match expert consensus, not a confident paraphrase. A product blog about standup meetings is not that category. A page about medicine, loans, or legal rights is. Cite more carefully there, and do not let a model invent the consensus.

There is a matching failure on the other side: citing a document you did not use, or citing a more impressive document than the one you used, so the page looks sourced. That is a false byline for a document. If a reviewer cannot open the link and find the claim, take the citation out and either find the real source or delete the claim. The brief’s do-not-invent list exists for this. Numbers, quotes, and “Google confirmed” all belong on it unless the document is in front of the drafter.

What an answer engine repeats

Answer engines assemble a reply from passages. A passage that states a rule and names the publisher is easier to use without stripping the credit. A passage that says “it is widely recommended” will be repeated as if you were the authority for a recommendation you borrowed. Write so that a one-paragraph extract still contains the attribution.

That is a reason to be dull and specific. “Google’s duplicate-URL documentation, updated 10 July 2026, lists redirects as a stronger canonical signal than a sitemap” can be quoted without inventing a speaker. “Canonicals are basically 301s” cannot. If your sentence is the one an overview repeats, you want it to be a sentence you would still sign, with the source visible.

Do not put the only copy of a citation inside an image. A diagram can show the four parts of a citation. The sentence next to the diagram has to carry the same point in HTML, for the same reason the image SEO guide tells you not to hide the answer in a picture. Alt text is not the citation.

When you publish a correction, correct the sentence and the citation together. Leaving the old claim and adding “update:” at the bottom means an extracted paragraph can still be the old claim. The refresh guide’s rule applies: change the passage first, then the date.

A citation example

Northwind is writing the standup procedure and needs one rule from Google: what to do with the old /status-meetings URL. The drafter first finds a blog post titled “Canonical tags pass full link equity.” That post does not link Google. It links another blog. Northwind does not cite it.

The sentence they publish is: “Google’s page on specifying a canonical URL, updated 10 July 2026, says a redirect is a strong signal that the target should become canonical, and that redirects are for retiring a duplicate. We redirected /status-meetings to /async-standups because the old slug was the same procedure. The print view stays on its own URL, with a canonical tag, because people still open it.” The link goes to Google’s documentation. The second and third sentences are Northwind’s decision. A reader can check the rule and can see which part is local.

They do not add three more outbound links to “authority” sites that define the word canonical. They do not nofollow Google. They do not write “according to SEO experts, 73% of duplicate pages recover within 30 days.” Nobody measured that. The brief already forbids the invented statistic. The citation habit is what you do instead: either you have the document, or you do not make the claim.

How this shows up in a BloGoose draft

A BloGoose draft is supposed to carry the sources the research step actually opened, next to the claims that use them. The activity trail on a run records which sources were cited. That trail is for you, the editor. It does not replace a sentence in the article. If the draft says “Google recommends” and neither the paragraph nor the trail names a URL, treat the sentence as unsourced and cut it or find the page.

The brief is the right place to list the documents a draft may rely on, and the claims it must not invent. When you approve a draft, check three things. Each borrowed rule names a publisher and a date. Each of those names is a link to the document, not to a recap. After the citation, there is a sentence that could only have been written for your reader: a decision, a limit, or an example from your product. If the draft is only the citations, it is not finished. Attribution did not do the work.

Paid mentions are a separate field, not a citation. If a draft is compensated to recommend a tool, that link is sponsored, and the page should say it is compensated in language a reader can see. A product mention of your own product, on your own site, is not an outbound citation. It still has to be true. The source for “BloGoose derives the canonical from the site URL and the slug” is your own publishing behavior, which you can confirm on one live post.

Questions about citing sources