01Notes · 23 August 2026
AI citations vs backlinks: what the SEO data actually says
Quick answer
In 2026 research covering 1,645 AI citations, 72.8% of cited pages had no page-level referring domains at all. Link count showed almost no independent effect on citation position.
The wrong conclusion is "stop link building". The right one is narrower: links are not the mechanism by which a page gets cited, so a link programme aimed at citations is aimed at the wrong target.
Key takeaways
- 72.8% of AI-cited pages had zero page-level referring domains. Links were not the entry condition in that sample.
- Ranking still matters enormously. Top-three pages appeared in AI citations 98.9% of the time, and links help you rank.
- The direct mechanism is being mentioned in a retrieved source, which a mention achieves and a link is not required for.
- Most link programmes do not target mentions. They filter for domain authority, which is a different list.
- Bought links are now a liability in two places. Google’s spam policies cover generative responses explicitly.
- The right target list comes from measurement, not from a domain-authority threshold.
The finding, stated precisely
Across 1,645 AI citation observations spanning 31 topics, 72.8% of the cited pages had no page-level referring domains. Not few. None. And link count showed almost no independent effect on where a page placed in AI citations.
Precision matters here, so two clarifications. This is page-level referring domains, not domain-level authority. And it is a correlation study, not a controlled experiment. It tells you what cited pages looked like, not what caused them to be cited.
Even with those caveats, it is a hard number to reconcile with an off-page programme built entirely on link acquisition.
The conclusion people jump to, and why it is wrong
"Links are dead, stop building them." That reading is wrong and expensive.
The same body of research found that pages holding traditional top-three positions appeared in AI citations 98.9% of the time. Ranking remains the strongest single gateway to citation, and links remain one of the strongest inputs to ranking.
So links matter, indirectly, through ranking. What the data undermines is the direct claim, that acquiring links to a page is how you get that page cited in an AI answer. That specific mechanism is not what the numbers show.
| A backlink | An unlinked mention | |
|---|---|---|
| Helps classical ranking | Yes, directly | Weakly, if at all |
| Puts you in a retrieved source | Only if the source names you | Yes, that is what it is |
| Required for AI citation | No — 72.8% had none | Effectively yes |
| What programmes target | Almost all of them | Almost none |
| How it is selected | Domain authority filter | Whether the source actually gets retrieved |
| Risk if bought | Link scheme, now covering AI responses too | Inauthentic mentions are also in scope |
Read the fourth row. Almost every off-page programme is optimising the column that the citation data says is not required, and ignoring the one it says is.
The mechanism, plainly
An AI answer is assembled from sources the system retrieved. If a comparison article names six vendors and you are one of them, you can appear in the answer — whether or not that article links to you.
That is why reading the source list is the most useful diagnostic available. The question is not "who links to my competitor". It is "which pages get retrieved for my category, and do they mention me".
Those are different questions with different answers, and only one of them is what a standard backlink audit tells you.
Your off-page budget, reconsidered
Not the amount. The target list.
A conventional programme builds a prospect list by filtering for domain authority above some threshold, then pitches for links. A citation-aware programme builds the list from the sources that actually get retrieved for your query family, then works to be accurately represented in them — linked or not.
Sometimes that is a placement. Often it is smaller and faster: an outdated comparison article that can be corrected, a directory entry with the wrong pricing, a review profile nobody has activated. That is the work we do, and the unglamorous corrections usually move first.
Why cheap links got riskier, not safer
It would be easy to read all this as permission to care less about link quality. The opposite is true.
Google’s spam policies now explicitly cover generative AI responses as well as search results. A bought link profile was already a ranking liability. It is now a liability across both surfaces, and it compounds quietly until it does not.
Everything we do off-page is pitched and earned. The distinction, and why it decides whether the work survives a policy update, is set out here.
What this means for you
- Audit mentions, not just links. Most tools report one and ignore the other.
- Build the prospect list from retrieved sources, not from a domain-authority threshold.
- Do not read this as permission to buy cheap links. The risk went up, not down.
02Common questions
Questions this raises
Do backlinks still matter for AI citations?
Less directly than most people assume. In 2026 research, 72.8% of AI-cited pages had no page-level referring domains at all, and link count showed almost no independent effect on citation position.
So should we stop link building?
No, and that would be the wrong conclusion. Links still matter for classical ranking, and ranking still matters — top-three pages appeared in AI citations 98.9% of the time. What changes is that links are not the direct mechanism for citation.
What is the mechanism then?
Being mentioned in the sources a system retrieves. A comparison article that names you without linking to you still puts you in the answer. That is a mention, not a link, and most link programmes do not target it.
Are brand mentions better than backlinks for AI search?
For citation specifically, the evidence points that way. For overall search health, links still do work mentions do not. The honest answer is that they do different jobs and you need both.
Does this mean cheap links are fine now?
The opposite. Google's spam policies now explicitly cover generative AI responses, so a bought link profile is a liability in two places rather than one.
How should this change our off-page budget?
Shift some of it from acquiring links to earning mentions in the specific sources your category's answers draw on. Same discipline, different target list, and the target list comes from measurement, not from a domain-authority filter.
Sources
Every figure on this page comes from one of the following. Where a claim is ours rather than a source’s, it carries an evidence label instead.
- Rankability — State of AI Search, 1,645 citation observations across 31 topics, 2026. Source for the 72.8% zero-referring-domain finding, the 43.3% vs 15.5% query breadth finding and the link-count finding. Independent study
- Google Search Central — spam policies covering link schemes, now explicitly in scope for generative AI responses, and Optimizing your website for generative AI features (15 May 2026, updated 10 July 2026). Primary
- Searlixa delivery across client engagements since 2018, labelled Observed with sample and date. Our own
Ask an assistant
Hand this page to an assistant and ask it to summarise or challenge the argument. If it gets something wrong about us, that is itself a useful finding.
→Next step