Citing your sources, and why nobody wants to
Almost nobody links out on purpose, and the reason given is always the same: why send a reader somewhere else. The answer is that a page which cites nothing is a page asking to be believed on its own account.
The short answer
Linking out to sources does not leak value in any way that shows up in results, and refusing to do it is one of the most persistent superstitions in SEO. A citation tells a reader where a claim came from and tells an engine what your page is about. The cost that is real is link rot, which needs a periodic sweep.
The fear of sending a reader away
The objection has two halves, and one of them was true twenty years ago. The first is that linking out leaks ranking value. The second is that a reader who leaves might not come back.
On the first: the idea that you preserve authority by hoarding links is a misreading of how links work, and the practice it produced, adding nofollow to every external link on the site, was explicitly identified as pointless by search engineers long ago. You are not spending a budget. A page that links to three authoritative sources is not a weaker page than one that links to none.
On the second: a reader who follows a citation is a reader who wanted the citation. Sending them to the specification, the study or the manufacturer's documentation is the page doing its job, and pages that do their job get returned to. The alternative, describing a source without linking it, means the reader either takes your word for it or opens a search in another tab. The second outcome is worse for you than the link was.
There is one genuine version of the concern, and it is about placement rather than principle: a prominent external link in the middle of a checkout or a signup flow is a distraction, and that is a conversion decision. It has nothing to do with the citations at the bottom of an article.
What a citation is actually doing
Four things, and only one of them is about search.
- It makes a claim checkable. The reason to cite is that somebody can verify what you said. This is the entire value and everything else follows from it.
- It shows the work was done. A page about a standard that links the standard was written by somebody who read it. A page about a standard that links nothing may have been assembled from four other pages about the standard.
- It places you in a subject. The set of things you link to is a description of what field you are in and who you consider authoritative, and that is legible to anything reading the page.
- It is a trust signal both raters and models read. Human quality raters are given guidance about whether claims are supported. Systems that assess sources look for the same thing. Neither of them can distinguish a well-researched page from a confident one unless the research is on the page.
What none of that licenses is a links section for its own sake. Five links to the same encyclopedia entry, dropped at the bottom under a heading nobody reads, is not a citation pattern. The link belongs at the sentence making the claim, because that is what makes it checkable.
Cite the primary source, not the article about the primary source.
If your claim comes from a specification, link the specification. If it comes from a study, link the study rather than the news write-up of the study. This is the difference between a citation and a reference to a citation, and it is the thing that separates a page that did the work from one that read a summary.
It also ages better. Secondary coverage disappears, gets rewritten and moves behind paywalls. Primary sources tend to keep their addresses, which matters for the section below.
The cost nobody budgets for
Every external link is a dependency on somebody else's decisions. They will restructure their site, retire a product, move a document behind a login, or shut down. Your page will keep pointing at whatever is there afterwards, and nothing will tell you.
The failure modes are worse than a 404, which is at least honest. A link can now resolve to a domain somebody else bought, to a parked page full of adverts, or to a redirect chain ending somewhere unrelated. Expired domains in an old article are a known way for a respectable page to end up recommending something nobody at your company would.
Three practices, in increasing order of effort:
- Check them on a schedule. A quarterly sweep for non-200 responses across your external links is a small script and it catches most of this.
- Name the source in the sentence. Write
the HTML specification saysrather thanas documented here. When the link dies, the sentence still tells the reader what to search for, and the page degrades instead of breaking. - Archive anything load-bearing. If a claim depends on a page that might not exist next year, link an archived snapshot alongside the live URL. This is standard practice in reference work and almost unknown outside it.
On attributes: use rel="noopener" when you open a link in a new tab, which is a security measure rather than an SEO one. Reserve nofollow and rel="sponsored" for links you were paid for or cannot vouch for, and leave your citations alone.
Every destination has to be resolved to be classified
Two things have to happen before this check can say anything. Each link has to be resolved to know whether it leaves your domain, and then fetched to know whether it still works. Both are properties of the destination rather than of the page you are looking at.
This one needs the full crawl
Telling an external link from an internal one means resolving each destination, and knowing whether it still works means fetching every one of them.
The instant search on this site audits a single page, so rather than show you a verdict it cannot support, this guide sends you to the place the check actually runs.
Does this affect AI search?
This is one of the checks with a stronger case in the machine setting than in the search one, and the reason is that assistants are answering questions where being wrong is costly. Systems that assemble answers from sources have an obvious interest in whether a source supports its claims, and the only way a page can demonstrate that is by pointing at something.
There is a second and more concrete mechanism. When a model is deciding which of several pages to quote on a factual claim, a page that states the claim and cites the primary document is easier to rely on than one that states it alone. Not because a link is a magic signal, but because the citation makes the claim traceable, and traceability is the property being selected for.
The dead-link problem has a sharper edge here too. A page whose citations resolve to parked domains is a page whose sourcing has quietly inverted: it now points at adverts. That is a bad look for a human reader and it is worse in an assessment of whether a site is a reliable source, which makes the quarterly sweep above cheaper than it sounds.
Why citing sources cannot rate higher than a Refinement
Because there is no mechanism to point at. Nothing indexes differently, nothing is excluded, no result format depends on it, and no threshold exists that a page falls below. What we have is a plausible correlation between well-sourced pages and pages that get trusted, which is exactly the category the bottom of the scale exists to hold honestly.
It also cannot be rated higher because the right number of outbound links is frequently zero. A product page, a pricing page, a contact page and a login screen have nothing to cite, and a check that pushed every URL toward having citations would be pushing most sites toward decoration.
What keeps it in the catalogue is that on the pages where it does apply, the ones making factual claims, the absence is diagnostic. A guide full of assertions with nothing behind them is usually a guide written from other people's guides, and that is worth knowing about your own site. Read this finding as a prompt to look at a page, not as a target to raise.
Sweeping your own citations for rot
The whole practice fits in one script, and quarterly is often enough.
DOMAIN=example.com
curl -s "https://$DOMAIN/guides/carbon-steel" \
| grep -o 'href="https\?://[^"]*"' \
| sed 's/href="//;s/"$//' \
| grep -v "$DOMAIN" \
| sort -u \
| while read -r url; do
code=$(curl -s -o /dev/null -m 10 -w '%{http_code}' -L "$url")
case "$code" in
200) ;;
000) echo "unreachable: $url" ;;
*) echo "$code: $url" ;;
esac
doneA 200 is not proof the page still says what you cited, only that something answers. For the links your argument actually rests on, open them once a year and read the first paragraph. That is the only check that catches a document being silently rewritten, which happens more often than it disappearing.
Questions this check raises
- Do outbound links hurt my rankings?
- No. The idea comes from a misreading of how PageRank was described in the 1990s, and it has not been true of any modern system. Pages that cite sources are, if anything, treated as more trustworthy, and refusing to link out costs a reader the ability to verify what you told them.
- Should outbound links open in a new tab?
- That is a usability question rather than an SEO one, and the honest answer is that it should be the reader choice. Forcing a new tab takes away their back button and is a particular problem on mobile. If you do it, mark it up so a screen reader announces it rather than silently opening a window.
- Do I need rel="nofollow" on outbound links?
- Only on links you did not editorially choose: paid placements, user-generated content, and anything you are contractually required to include. Google has separate values for those, sponsored and ugc. A citation you chose to make should be a normal link, because that is what the value describes.