Domain Authority Impact on AI Citations: What 5 Million Citations Show

It's a natural carry-over assumption from the "ten blue links" era: stronger domains should get cited more by AI search engines. 

A study spanning 20,000 prompts and roughly 5 million citation URLs tested that assumption directly, and the relationship turned out to be close to nothing at all.

Domain authority, whether measured as PageRank, Harmonic Centrality, or a proprietary composite score, has been the backbone of SEO strategy for two decades: build enough authoritative links, and rankings follow. It's a reasonable instinct to assume the same logic carries over to AI Overviews, AI Mode, ChatGPT, and Perplexity. The data says otherwise, and the way it says otherwise is more interesting than a flat "no."

Domain Authority Impact on AI Citations

Researchers pulled three months of AI citation-tracking data spanning 20,000 prompts and roughly 5 million unique citation source URLs across AI Mode, AI Overviews, ChatGPT, and Perplexity, then measured how often a domain got cited against three different authority metrics: PageRank and Harmonic Centrality (both derived from Common Crawl's web graph) and a proprietary composite Domain Score. All three are different lenses on the same underlying question, how authoritative is this domain, by classic link-based logic, which is exactly the assumption the study set out to test.

The Headline Finding: Authority Barely Correlates With Citation, And the Sign is Negative

Every one of these Spearman correlations sits close enough to zero to be statistical noise. If anything, the sign leans slightly negative, the opposite of what the "stronger domains win" assumption would predict. That's an odd enough result to be worth digging into rather than taking at face value.

Why The Correlation is Negative: A Handful of Mega-Domains Skew Everything

The researchers' working theory was that massively over-represented publishers, Reddit and Wikipedia especially, which cover nearly every topic and rank well across the board, dominate the dataset so thoroughly that they distort the relationship. To test it, they removed the top 5% of domains by strength entirely and re-ran the correlation.

With the top 5% removed, the already-weak negative correlations collapsed even closer to zero across every model and metric. That's consistent with AI answer engines actively deprioritizing the very strongest domains, plausibly to diversify sources rather than defaulting to the same handful of mega-publishers for every answer, though an equally plausible alternative is simply that the strongest domains publish less genuinely citable content per page than the data might suggest.

Domain authority doesn't appear to have any meaningful impact on the likelihood of being cited by AI search engines. That's the headline. It doesn't mean off-page signals are irrelevant entirely, but the specific, classic link-authority metrics that drove traditional SEO for two decades don't carry over cleanly to AI citation.

A Stranger Wrinkle: Filtering For Blog Posts

Filtering the dataset down to URLs containing /blog/ (an imperfect but reasonable proxy for blog content) nudged PageRank and Domain Score very slightly positive across most models, but Harmonic Centrality did something stranger, swinging to a mild positive correlation (+0.162) for AI Mode while tanking to a sharper negative one (-0.226) for AI Overviews.

The likely explanation involves how these models get built in the first place: Common Crawl's bot prioritizes scraping by Harmonic Centrality, meaning high-HC domains get crawled more heavily and end up better represented in the training data behind these models. 

Which alone makes a domain a stronger candidate to be surfaced, independent of anything about the specific page. Within blog content specifically, general-topic blogs naturally accumulate higher HC scores than tightly focused niche blogs simply by touching more subjects, which may explain part of the swing. Though the researchers were candid that the inconsistency across models made this a partial explanation at best, not a settled one.

How this fits the wider research picture

This isn't an isolated finding. Separate large-scale tracking (Ahrefs, analyzing hundreds of thousands of keywords) has documented a sharp decline in how often AI Overview citations come from pages that also rank in the traditional top 10. A substantial drop occurred over the second half of 2025 into 2026, consistent with AI systems drawing on a broader, less rank-dependent pool of sources than classic search ever did.

What This Means: Content Probably Matters More Than Authority

If classic link-based authority isn't doing much predictive work, the natural implication is that something else is carrying more of the weight, and content itself is the most plausible candidate, since it's the one lever available to a smaller or newer site the same way it's available to an established one. As one industry commentator put it, in a framing that captures the practical implication well: the winning position for both traditional rankings and AI citations is content anchored in established consensus that simultaneously contributes genuine new information.

That framing matters because it names two things at once: you need to cover what's already established and expected on a topic, and you need to add something a generic summary of existing sources wouldn't already contain.

What to Actually Do About It

1. Don't abandon off-page work entirely: The study measured three specific link-based authority metrics; it didn't test every possible off-page signal, and some (brand mentions without a link, community discussion, review-site presence) plausibly still matter even if classic link authority doesn't.

2. Research what's already established before writing: Understanding the key entities, subtopics, and facts that AI systems and top-ranking pages already treat as baseline coverage for a topic is the "established consensus" half of the equation; skipping it means competing against content that already covers the basics you're missing.

3. Add something that isn't already out there: Original research, proprietary data, expert quotes, and genuinely held opinions are the "genuine new information" half of the equation, the part a purely competent-but-generic write-up can't replicate, and the part most likely to make a piece citable rather than summarizable-away.

4. Treat blog content as a reasonable home for this strategy. The (admittedly noisy) blog-post-filtered data nudging positive, combined with separate research finding blog mentions correlate more strongly with AI recommendation strength than other content types, makes blog-format content a defensible place to invest this effort.

What Not to Conclude From This

"Domain Authority is Now Completely Irrelevant"

The data shows a weak, close-to-zero relationship specific to AI citation likelihood; it says nothing about domain authority's continued relevance to traditional Google rankings, where it remains meaningfully predictive.

Off-page SEO Doesn't Matter for AI Search

This study tested three specific link-based metrics, not the full universe of off-page signals. Brand mentions, community presence, and review-site visibility weren't part of what was measured here and may behave differently.

Correlation This Weak Means Causation is Proven Either Way

A near-zero correlation is evidence against domain authority being a strong direct driver of citation, it isn't proof of exactly what does drive it. The content-quality conclusion is a reasonable inference from the pattern, not something the correlation data alone establishes.

Frequently Asked Questions

Does this Mean Small, New Sites Can Compete With Established Publishers for AI Citations?

The data is consistent with that being more possible in AI search than it ever was in traditional rankings, since classic domain authority isn't doing much predictive work. It doesn't guarantee it, a new site still needs genuinely competitive, well-researched content, since content quality appears to be carrying more of the weight that authority used to carry.

Why would AI Search Engines Deprioritize the Strongest Domains?

One plausible explanation is a form of deliberate diversification; engineers may want to avoid over-relying on a handful of mega-publishers for every answer, even when those publishers are strong candidates by raw authority. Another is simply that very strong, broad-topic domains publish citable content less densely per page than the raw authority number would suggest. The study's data can't fully distinguish between these explanations.

Should I Stop Building Backlinks?

Backlinks and domain authority remain relevant to traditional Google rankings, and traditional rankings still feed into AI Overview and AI Mode citation eligibility indirectly. What this data suggests is that pursuing domain authority specifically as an AI-citation strategy is likely a weaker investment than pursuing genuinely differentiated content.

Why Did Harmonic Centrality Behave so Differently from PageRank and Domain Score?

The leading explanation ties back to how large language models get trained: Common Crawl's scraping bot prioritizes by Harmonic Centrality, so high-HC domains are better represented in training data and become stronger surfacing candidates independent of any single page's quality. Within blog content, general-topic blogs likely score higher HC than focused niche blogs simply by covering more ground. Though the researchers were explicit that the swing between models made this a partial, not definitive, explanation.

Is This Finding Specific to Surfer's Tracking Methodology, or Does it Hold Up Elsewhere?

Separate large-scale research (notably Ahrefs' tracking of AI Overview citations against top-10 organic rankings) shows a related pattern. This is a declining relationship between classic ranking signals and AI citation over time. The exact correlation figures circulating elsewhere in the industry vary widely and aren't always methodologically transparent, so this study's detailed, disclosed methodology is a reasonable anchor even though it isn't the only data point available.

Related Searches:

• Why Your Blog Gets Traffic but No Leads
• LLM Scraped AI Answers vs API Results: Why They Don't Match
• How to Write Small Business Website Copy