We found no links to external sources

The check walks every crawled page looking for at least one outbound link that can count as a link to a source. A single qualifying link anywhere on the site is enough and no finding is raised.

What does not count as a source link (and this is the heart of the check): the link has to lead off your domain and also outside the excluded domains listed in config/authority-domains.json. Those are your own social profiles, CDNs and similar. Without that, a link to your own Facebook page would count as “citing a source”, which would be nonsense.

What the check does not do:

• It does not judge the quality of the source. A link to a blog of unclear provenance counts the same as a link to a peer-reviewed study.
• It does not count how many there are. One link on the entire site silences the finding just as a hundred would. It is a yes/no check for the whole site, not a measure of how habitually you cite.
• It does not verify the link works. An outbound link to a dead page still counts; the crawler never visits other domains, so we have no way to know.
• It does not check that the link relates to the text. A partner link in the footer satisfies the condition just like a citation in an article.
• It does not measure the other two factors from the experiment discussed below: neither quotations nor concrete statistics. That is a known gap, not an oversight.

The finding is site-level and appears at most once in the report; severity is warning.

This is the only check in the entire audit backed by a peer-reviewed experiment, which is why it gets more space here than its technical simplicity would suggest.

Researchers at Princeton published a study at KDD 2024 (GEO-bench, 10,000 queries) that black-box tested what measurably lifts a page's visibility in the answers of generative search engines. They measured three things:

It is the best-supported lever in the field and, importantly, it is about content, not technique. No markup, no file at the root of your site, no setting. Just how the text is written.

Where we have to be honest. The experiment measured what happens when source links are added to the text. Our check measures only whether at least one exists anywhere. That is a far cruder question, and a 28 % lift does not follow from it. It is a necessary first step, not that result.

A second admission: the two factors that scored even higher are not measured by us at all. Quotations and statistics would have to be judged on substance rather than by an HTML element. If you are going to take something up, take those, and the strongest of them (+41 %) has no code of ours at all.

And one thing you will meet in articles about authority that has no documented support: that domain authority, backlinks or domain age affect AI visibility. The major citation-tracking tools do not analyse those factors at all; it is a classic SEO claim transplanted into GEO without evidence. We keep it out of the score.

One link silences the finding, but doing only that would miss the whole point. So the advice comes in two levels, and the second is the one that matters.

The minimum: add real source links wherever your text already refers to sources in words. Copy that says “according to research” or “the manufacturer states” has a specific document behind it. Link that document.

<!-- Weak: a claim with nothing behind it -->
<p>Research suggests demand for heat pumps is growing.</p>

<!-- Strong: a concrete number, a named source, a working link -->
<p>Heat pump sales rose 18 % in 2025
  (<a href="https://example.org/statistics-2025">Trade Association, annual report 2025</a>).</p>

The part that matters more: add the other two factors from the experiment to your text, even though our audit does not measure them. Those are where the measured effect was strongest:

1. Concrete numbers instead of general claims. “Fast delivery” carries no information. “Delivered within two working days, on 94 % of orders” does, and it is quotable.
2. Verbatim quotations. A sentence from an expert, a customer or a document, presented as a quote with an attribution.
3. Links to sources for both of the above.

What to watch out for, so the work is not wasted:

• Link the primary source, not an article about it. A link to the report carries different weight from a link to a blog post covering it.
• rel="nofollow" on a source link makes no sense. It is for paid and untrusted links; citing a source is neither.
• Do not hide links behind a redirector. A URL going through your own /redirect?url=… is not, to a reader or a parser, a link to that source.
• Do not link for the sake of the count. The check counts one, and the experiment measured relevant citations, not a list of links in the footer.

And to close with what holds most strongly across the whole audit: this is the only recommendation where we can show you a measured number. If you have time for one thing from the entire report, make it this one.