Backlinks

A backlink is a hyperlink on someone else's site pointing at yours. Search engines use links for two distinct jobs — finding pages, and judging them — and conflating the two is where most link strategy goes wrong.

The two jobs a link does

Discovery. A crawler reaches a page by following a link to it, reading it in a sitemap, or being told about it directly. A page with no inbound link and no sitemap entry is, for practical purposes, invisible: nothing points at it, so nothing fetches it. This job is binary and unglamorous, and it is the one that most often explains a page that "isn't ranking" — it was never crawled or indexed.

Endorsement. A link from a page an engine already trusts is treated as an input to how that engine assesses the target. This job is probabilistic, heavily filtered, and not something a publisher can inspect. It is also the one that attracts manipulation, which is why the policy surface around it is so large.

What the attributes actually change

Google documents three rel values for qualifying an outbound link:

• `sponsored` — advertisements and paid placements. Preferred for paid links.
• `ugc` — user-generated content, such as comments and forum posts.
• `nofollow` — the fallback, for when the other values do not apply and you would rather Google not associate your site with, or crawl, the linked page.

The important caveat is in the same document: links marked with these attributes "will generally not be followed", but the linked pages "may be found through other means, such as sitemaps or links from other sites, and thus they may still be crawled". nofollow withholds an endorsement. It does not block access, and treating it as access control is a mistake. That is what robots.txt and authentication are for.

What counts as spam

Google's spam policies define link spam as creating links to or from a site "primarily for the purpose of manipulating search rankings", and name the cases:

• buying or selling links for ranking purposes, including exchanging goods or services for them
• excessive link exchanges, or partner pages built solely for cross-linking
• using automated programs or services to create links
• requiring a link as part of a Terms of Service, contract, or similar arrangement, without letting the third-party content owner qualify the outbound link
• text advertisements or text links that do not block ranking credit
• links from low-quality directory or bookmark sites
• forum comments with optimised links in the post or signature

Two adjacent policies matter to anyone acquiring links through placements. Site reputation abuse covers third-party content published on a host site mainly because of that host's established ranking signals — the guest-post-farm pattern, named. Expired domain abuse covers buying a lapsed domain to host content of little value on top of its history.

The through-line is intent plus disclosure. A paid placement is not forbidden; an undisclosed paid placement passing ranking credit is.

Why this looks different when the reader is an agent

An assistant that retrieves documents cites what it fetched. It does not compute a link graph at answer time, and it does not have a "backlinks" field to weigh. So the naive version of the question — do backlinks make ChatGPT cite me? — has no direct mechanism behind it.

The indirect path is real, and it runs entirely through candidacy. AI search systems assemble a candidate set before generating anything, frequently from a conventional search index. Links affect whether your page is in that set, in two ways: they are how the page was discovered at all, and they are one input to the ranking that decides which handful of results the retrieval step actually pulls. Once your page is in the pool, links stop mattering and the passage does — whether the claim is stated in one self-contained sentence with its qualifier attached. That is the split GEO turns on, and it is why a link campaign and a rewrite are not substitutes for one another.

There is a third, softer effect that is easy to overstate and wrong to ignore: a link is usually accompanied by a sentence describing you. Those sentences are text on the linking page, and that page can itself be retrieved. What the internet says about you when it links to you is content in the corpus. See source citation.

Failure modes

• Treating `nofollow` as a robots directive. It withholds credit; it does not prevent a fetch.
• Buying attention instead of relevance. A link from a high-traffic page about an unrelated subject puts you in front of nobody who was looking for you, and sits inside exactly the pattern the site-reputation-abuse policy describes.
• Chasing counts. A count is not a measurement of anything an engine exposes. Two links from pages that genuinely discuss your subject are not equivalent to two hundred from directories.
• Ignoring the anchor and the surrounding sentence. The paragraph around a link is what a retrieval system reads. A link buried in a list of logos carries a URL and no claim.
• Leaving discovery to links alone. If a page is only reachable through one external link, its indexing is hostage to that page staying up. A sitemap and internal links cost nothing.
• Reclaiming lost links late. A link to a URL that now 404s is a discovery path that has been cut. Redirects preserve it; a custom 404 page does not.

Frequently asked questions

What is link spam, formally?

Creating links to or from a site primarily to manipulate search rankings, per Google's spam policies.

Which rel value goes on a paid link?

sponsored is preferred; nofollow remains acceptable. An unattributed paid link is the first listed example of link spam.

Does nofollow prevent crawling?

No. Google says such links are generally not followed, but the target may still be crawled if found another way.

Do backlinks cause AI citations?

Not directly. They influence discovery and ranking, which decide whether your page is in the candidate pool a retrieval step draws from; the passage decides the rest.