Do backlinks still matter for AI citations?

Backlinks matter far less for AI citations than they do for Google rankings, and the published correlations put them closer to a threshold than to a driver. Enough links to be crawled, indexed and treated as a real site appears to help; more links beyond that point does not reliably buy more citations.

The two systems are choosing different things, and that is what makes the split coherent. A ranking system is choosing which whole documents to list for a query, and link-based authority is a reasonable proxy for which documents a lot of people found worth referencing. An answer engine is choosing which passages to assemble into a reply, and a strong domain with no passage that answers the question has nothing to contribute.

What this changes in practice is where a budget goes, not whether authority is ever worth having. A company with a serious link-building line item and no page that answers its buyers' actual questions is spending against the weaker of the two signals. A company with excellent pages and almost no external presence has a different problem, and it is more often a mentions problem than a links problem.

The absolute version of the claim is wrong in both directions. Backlinks are not dead, and they are not the lever. Most of the confident content on this question in 2026 argues against a straw position, so the measurements are worth reading directly.

What does the data say about domain authority and AI citations?

Domain authority explains surprisingly little of the variation in AI citations once somebody measures it directly. According to Clairon's 2026 analysis of domain authority scores against AI citation behaviour, authority scores accounted for roughly 3% of the variation in AI citation probability, which makes a DA or DR number close to useless as a predictor of whether a specific page gets quoted.

Rank position tells the same story from a different angle. The ConvertMate GEO Benchmark 2026, an observational study of 12,500 queries across 8,000 domains, found 83% of AI citations coming from pages outside Google's organic top 10. Since link authority is one of the larger inputs to that top 10, a cited population drawn mostly from outside it is hard to reconcile with links being the dominant citation signal.

The strongest single piece of evidence is about what survives statistical controls. Discovered Labs' 2026 citation analysis, built on more than two million citations across ChatGPT, Claude, Perplexity and Gemini using fixed-effects regression, found that vocabulary alignment between a page and the way buyers search was the only page-level signal with a causal effect that survived controlling for domain authority, at an effect size of β=+0.37. Most other signals collapsed towards zero once domain was held constant.

That last finding is also the trap in every naive study of this question. High-authority domains get cited more and tend to have better-built content, so a correlation that does not control for domain makes content structure look like the cause of citations and authority look like a proxy for quality. Our methodology page explains why that control changes the conclusion so completely.

Why do backlinks matter less to an answer engine than to Google?

Backlinks matter less to an answer engine because the unit being selected is a passage, not a page, and a link says nothing about any particular passage. Retrieval pulls candidate chunks of text that look like answers to the question, reranks them, and the model writes from what it was handed. A link pointing at the page from elsewhere does not make a specific paragraph in it a better answer.

Content-side changes, by contrast, move the number measurably. In "GEO: Generative Engine Optimization" (ACM SIGKDD 2024), Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande tested content modifications across roughly 10,000 queries on their GEO-bench benchmark and reported visibility gains in the range of 22% to 41%, with adding citations and quotations among the most effective changes. Those are edits to the page, requiring no external agreement from anybody.

Engine architecture also splits the answer. Google's AI surfaces sit on top of an index whose ranking has always used the link graph, so link authority leaks into AI Overviews and AI Mode through the candidate set they draw from. ChatGPT leans on a search index plus its own retrieval, where mentions in forums, listicles and press carry more weight than a domain metric. Treating the four engines as one system is what produces contradictory advice, and our comparison of why a business shows up in ChatGPT but not Google AI Overviews works through the divergence.

Links are also a slow signal. Acquiring them takes months, and the citation picture rotates faster than that, so the effect of a link campaign lands after the question it was meant to win has already shifted. That makes links a poor primary lever however strong the signal is.

Do brand mentions work better than backlinks?

Brand mentions outperform backlinks in every published correlation on AI visibility, by a margin large enough to change where a budget should go. According to Ahrefs' 2025 analysis of brand visibility factors across 75,000 brands, brand web mentions correlated at 0.664 with visibility in Google AI Overviews against 0.218 for backlinks. Ahrefs' 2026 follow-up on the same 75,000 brands extended the pattern to ChatGPT and Google AI Mode, where YouTube brand mentions were the strongest individual signal measured, at roughly 0.737.

The reason mentions win is mechanical. An unlinked sentence naming your company in a page an engine retrieves is text the model can read, quote and reuse. A link is a pointer, and a pointer from a page the engine never retrieves for your buyer questions contributes nothing to an answer about your category.

Where the citations themselves come from points the same way. AirOps' 2026 State of AI Search report, which tracked citation patterns across ChatGPT, Perplexity, Google AI Overviews and Gemini, found about 85% of brand mentions reaching AI systems originated on third-party pages instead of owned domains, and about 48% of citations came from community platforms such as Reddit and YouTube.

The operational difference is who you talk to. Chasing links means asking for a specific HTML element on somebody else's page, which is a transaction many publishers now price. Chasing mentions means being easy to write about and worth naming, which is a different and usually cheaper ask, and our page on whether you need third-party mentions covers how to pick which surfaces to pursue.

If not backlinks, what should a link budget buy instead?

A link budget converts best into the three things an answer engine actually reads: pages that match the question, mentions on pages that get retrieved, and presence in the comparison content that buying questions pull. The table below compares where the same money goes.

Where the money goes	What it buys for AI citations	Who has to agree
Paid or outreach link building	A weak signal here, and a real one for classic Google rankings	A publisher, per link
Rewriting existing pages for vocabulary match	The heaviest measured page-level signal	Nobody outside your company
Publishing answers to uncovered buyer questions	Eligibility for questions you currently cannot win	Nobody outside your company
Digital PR aimed at unlinked mentions	The strongest off-site correlation in the published studies	A journalist or editor, per story
Inclusion in comparison and best-of pages	Presence on shortlist questions that retrieve listicles	The page owner, per entry
Review platform profiles and customer reviews	Quotable third-party description of what you do	Your customers
Server-side rendering and structure fixes	Whether any of the above can be read at all	Your own engineers

Read the third column before the second. Everything your own team controls can be done this quarter, and everything requiring an outside party arrives on their schedule. A budget reallocation that moves money from the bottom rows to the middle two usually produces visible movement faster, which also makes the next budget conversation easier.

Does this mean you should stop link building?

Stopping link building entirely is the wrong conclusion, for two reasons that have nothing to do with sentiment about links. Google's AI surfaces draw candidates from an index whose ranking uses the link graph, so links keep buying influence there even where they buy little in ChatGPT. And classic organic search is still a channel that pays for itself independently of anything an answer engine does.

The threshold argument is the second reason to keep some of it. A site with almost no external links tends to be crawled less, trusted less and included in fewer candidate sets, which is a floor effect rather than a ranking effect. Clearing the floor matters; climbing far above it does not appear to pay the same way, which is why our page on whether small sites can get cited by AI treats authority as a threshold to clear, not a dial to turn.

What is worth stopping is the specific practice of buying volume. Bulk link packages, directory submissions and low-quality guest posts were a weak play for rankings and are close to worthless for citations, because the pages hosting them are not retrieved for anybody's buyer questions. Spending there is the easiest line in a marketing budget to cut with no measurable loss.

It helps to treat a link as a side effect of something more useful. A placement in a publication your buyers actually read gives you a mention in retrievable text, a named source an engine can quote, and a link, in that order of value. Pursuing the link alone throws away the two more valuable parts of the same placement.

How do you tell whether authority is your problem at all?

Authority is your problem only in a specific, recognisable pattern, and most companies that believe it is theirs have a different one. Run your buyer questions across the engines, record the cited URLs, and look at the domains that won instead of at your own absence.

If the winners are mostly large, well-known domains and no company of your size appears anywhere in the cited set, authority may genuinely be acting as a gate on that question. That pattern is most common on broad, high-volume category questions and least common on the specific, narrow questions where buyers reveal real intent.

If companies similar to yours are cited and you are not, authority is not the constraint, because the engines are clearly willing to quote companies at your level. The difference is then on your pages or in your third-party presence, and the measurement tells you which: check whether the cited pages are those companies' own or somebody else's writing about them.

If nothing of yours is cited for any question, including ones where you are plainly the best answer, check retrieval and rendering before concluding anything about authority. A client-rendered site or a blocked crawler produces total absence that looks exactly like an authority problem and costs an afternoon to rule out, which our page on checking whether AI can read your website walks through.

The diagnosis matters because the two problems have different price tags. Building domain authority is a multi-year programme; fixing vocabulary and structure on twenty pages is a few weeks of writing. A free visibility assessment runs the questions and reports which pattern you are in, which is the step that keeps a budget from going to the expensive answer when the cheap one applies.

What else do people ask about backlinks and AI citations?