Do Outbound Links Help or Hurt Your AI Citations?

Old-school SEO trained everyone to hoard link equity. Keep every link internal, never send a reader away, treat an outbound link like a leak in the bucket. For AI citations that reflex is not just dated - it is backwards, and now there is peer-reviewed work that says so.

An answer engine does one thing when it builds a response: it decides which sources it trusts enough to quote. A page that cites named, linked sources reads like it did the work. A page with zero outbound references reads like an island, and islands are hard to verify. So the engine reaches for the page it can check.

What the research actually says

This is not my hunch. The foundational GEO study - "Generative Engine Optimization" by Aggarwal et al., out of Princeton, Georgia Tech and the Allen Institute for AI, presented at KDD 2024 - tested which on-page changes actually move you into AI answers. They ran the experiments across thousands of queries on a purpose-built benchmark, GEO-bench.

The three highest-impact moves were not keyword tricks. They were Cite Sources, Quotation Addition, and Statistics Addition - adding credible citations, direct quotes, and sourced numbers. Together they lifted visibility in generative-engine responses by up to 40%.

Read that again. The single biggest lever the researchers found was citing other people's sources - the exact thing the old playbook told you to avoid.

A 2025 follow-up makes the point from the engine's side. "Source Coverage and Citation Bias in LLM-based vs. Traditional Search Engines" compared what AI search engines surface against what classic search returns. The domains the AI engines favored shared a profile: structured, hierarchical HTML, easier-to-read text, and more outbound links to reputable sources. The pages that link out get picked more often, not less.

Why it works

Three mechanisms, and I see all three in the pages that get cited.

1. They prove you did the homework. A claim with a named, linked source behind it is verifiable. The model follows the chain, confirms the claim, and quotes you with confidence. A bare assertion gives it nothing to lean on.
2. They put you in good company. Linking to high-authority sources associates your page with them. Models weight this the way an academic weights the references in a paper - the company you keep says something about you.
3. They make extraction clean. "According to [source], X" is a structure models love to lift, because the attribution is already built in. You are handing the engine a pre-formatted citation it can drop straight into an answer.

The one case where they hurt

Outbound links hurt when the company you keep is bad. Spammy, irrelevant, or so dense the page reads like a link farm - a model judging trust downgrades you for it, the same way it downgrades a page buried in ads.

The signal is about your neighbors, so keep good ones. Twenty links to thin aggregators hurt. Two links to primary sources help. The count is not the point; the quality of who you point at is.

There is a softer, non-citation cost too. If every outbound link drops the buyer onto a competitor's page at the exact moment they were about to convert, that is a merchandising problem, not an AI one. Open external links in a new tab, place them with some thought, and do not strip your sources to dodge it.

How I actually do it

The rules I ship, in order:

1. Link to primary sources, not aggregators. Cite the original study, the official docs, the first-party announcement - not the blog that summarized them. Models weight primary sources higher because the chain is shorter and easier to verify. The GEO paper, not the LinkedIn post about the GEO paper.
2. Name the source in the sentence. "Princeton's GEO study found..." is more extractable than a naked hyperlink on the word "here." Inline attribution survives being lifted out of context; a bare link does not.
3. Add a real number or a direct quote. Statistics Addition and Quotation Addition were two of the three winning moves in the research. A sourced "up to 40%" beats "significantly improves" every time.
4. Audit your outbound links before you ship. Crawl the page with Screaming Frog to catch dead links and redirects, and sanity-check that every target is a source you would be happy to be quoted next to. Two great citations beat twenty filler ones.
5. Keep it clean. Use rel="noopener" on external links, link only where the claim needs a source, and never paste in a citation you have not read. A wrong link is worse than no link.

The tell

Notice what this post just did. Every claim about the research is linked to the paper it came from, by name. That is not decoration - it is the same move I am telling you to make, and it is why a model can quote this page without guessing.

From auditing 50+ brands, the pages that get cited most are almost never the ones hoarding link equity. They read like a well-sourced briefing - confident claims, named references, real numbers, clean structure. That is the page a model wants to quote, and now there is a benchmark that agrees.

Citing other people's sources is step one. The GEO study's other two top moves were adding quotes and adding stats - and that is exactly what our Quote Booster automates: it inserts real, attributed expert quotes into your product and content pages, the kind AI assistants lift straight into an answer.

Give the engines something quotable to quote. Try the Quote Booster, free.