Do Backlinks Matter for AI Search Like ChatGPT?

Indirectly, and unevenly. Assistants answer by retrieving from a search index, and links are one of the things that get a page into an index and keep it eligible, so links matter to the extent that retrieval is search-shaped. What links do not do is make a model cite you. Citation follows from being the source a retrieval step reaches for, which correlates with being referenced by other sites and is not the same contest as ranking. A measurement study of 55,393 queries found that nearly 30% of the domains cited in a Google AI Overview do not appear in the first-page results shown beside it. Nobody outside the vendors can decompose the selection, and anyone giving you a weight is inventing it.

Here is what has actually been measured, what it supports, and where the honest answer is "not known".

What does the measurement show?

Four studies, four different prompt sets, four different definitions of share. Read them for direction, not for a number to quote.

Two things follow and one does not.

It follows that citation correlates with being referenced by many other domains. The SE Ranking figure is the clearest version of that, at a very large referring-domain threshold.

It follows that citation is not ranking. Nearly a third of what Google's own answer engine cited was not on page one for that query, and the authors read that as a separate selection mechanism.

What does not follow is that links cause citation. No study in the table tests the mechanism. They measure which domains get cited and what those domains have in common. That is correlation, on somebody else's prompt set, for one engine, in one month.

Why links matter at all, then

Because retrieval needs an index, and Google says the index is the floor for its own AI features: a page has to be indexed and eligible for a snippet to be cited in an AI Overview. Google's ranking systems guide lists link analysis among its core systems and says PageRank "continues to be part of our core ranking systems". Links are part of what gets you crawled, indexed and treated as worth retrieving.

That is a real mechanism and a modest one. It means links buy you eligibility. It does not mean the assistant picks you once you are eligible, which is the step the studies show is decided by something else.

So what selects the citation?

Nobody outside the vendors knows, and that sentence is the most useful thing on this page.

What the data supports is a prior rather than a model. The sources engines reach for tend to be sources other credible sources already reference. Encyclopaedic references, education and government pages, long-lived community threads and trade publications recur across every study, which is why the per-engine top-ten lists look so similar.

Our own position, stated as a position: a domain's inbound provenance is worth measuring, and that is the premise behind Agent Citability, our published score for how deeply a domain sits in the part of the web that cited sources link to. The methodology is published, not open. The tier philosophy, example seed domains and release process are on the methodology page. The weights are not, so a third party cannot recompute the number, and we would rather say that than imply an openness we do not have. Scores ship as versioned releases and can fall between them when the pages holding the references change.

We have not found a study testing whether a link from a trusted domain predicts that the linked domain gets cited. That is the hypothesis the score rests on, and until it is tested properly the tiering is a well-grounded prior rather than a validated model.

Do the platforms the engines cite most help you?

This is where the link-count answer and the useful answer come apart.

Reddit, YouTube and LinkedIn are at the top of nearly every cited-domain list. A link from one of them is placed by whoever wanted it placed, in about a minute. It is not evidence that anyone vouched for you, and a self-promotional comment is removed by the community before it is anything else.

The engines cite the thread, not you. Being quoted inside a thread somebody else started is worth something. Creating the link yourself is worth close to nothing, which is why we weight links by whether the citing domain chose to publish them rather than by how often an assistant reads that domain.

The numbers also move fast. Promptwatch measured Reddit at "a steady 3.8% share of ChatGPT citations from July 18 through August 7" 2026, then "just 0.5%" between 14 and 17 August, "an 86% decline relative to the previous period", as reported by Semrush on 26 August 2026. Promptwatch "calls the size of the drop provisional and says it can't rule out a problem in its own data collection". Any ranking of cited domains is a snapshot of one engine's retrieval settings that month.

Using the Revised directory for this

If provenance is the slow half, the domain is the part you can buy rather than earn. As of 14 September 2026 the directory held 11,255 listings, every one expired and available to register, with 84% passing a low spam screen.

1. Open the source hub, which groups expired domains by the sites that reference them. It is the view that matches the question on this page.
2. Start from a source family the studies name. Wikipedia backlinks held 4,772 listings, .edu backlinks 108, .gov backlinks 42.
3. In Quality, set a minimum Agent Citability score, then set Spam signal to the low band.
4. Set Category in Basics to your own subject. A domain referenced for something else inherits nothing you can use. This is the check that disqualifies most high-scoring names.
5. Read the Linked by column before revealing anything.
6. Run your own domain and the domains an assistant cited in your test questions through the Agent Citability checker and the backlink checker. The gap between the two is the size of the job.

Filtering and sorting are free signed in or not. A plan buys reveals.

For the wider picture, which websites LLMs cite has the full study table with denominators, and how to get cited by ChatGPT covers the crawler half in OpenAI's own words.

What does not work

Buying links to get cited. The engines lean on domains that other domains chose to reference, and a bought link is the category Google names as link spam.

Assuming a page that ranks will be cited. Nearly 30% of cited domains in the Washington University study were not on page one for the same query.

Chasing a top-ten cited domain list. If ten domains carry 11 to 13% of citations, the other 87% or so is where a normal site lives.

Treating any published percentage as a target. Four studies, four denominators, and one of them swung by 86% inside a month.

FAQ

Do backlinks help with ChatGPT rankings? There is no ranking in ChatGPT. There is retrieval, and links help a page be indexed and retrievable. SE Ranking found sites with over 32,000 referring domains were 3.5 times more likely to be cited than sites with up to 200, which is a correlation rather than a mechanism.

Does AI search use backlinks? Indirectly. Assistants retrieve from a search index, and links are part of what puts pages in that index and keeps them eligible. No vendor publishes a link weight for citation selection.

Are AI citations the same as Google rankings? No. Being indexed is the floor for AI Overviews, but the Washington University study found nearly 30% of cited domains were not in the first-page results beside the overview.

Can I pay to get cited by an AI assistant? Not as a citation. Ads appear as ads. Nobody sells placement in the cited sources.

Do links from Reddit help me get cited? The thread gets cited, not you. Being quoted in a thread has some value. Posting your own link is placed by you and is not evidence anyone vouched for you.

Does an aged domain get cited more easily? An aged domain already referenced by sources the engines lean on starts inside the pool they draw from. Age alone does nothing, and a domain referenced for an unrelated subject does not help.

If the referenced-by-others half is the part you cannot rush, decide the domain first. The Revised directory of expired domains with backlinks names the referencing sources on every listing, with spam screening on every row, so you can check whether a domain's existing references are in your subject before revealing it.