Ahrefs looked at 75,000 brands and found backlinks correlating with AI Overview visibility at 0.218. Semrush looked at 1,000 domains and found authority correlating with AI mentions at 0.65. Both numbers are real. Both get quoted constantly, usually by people arguing with each other on the internet. They should not be arguing. The two studies asked different questions about backlinks and AI search visibility, and once you line the questions up next to each other, the thing they agree on is far more useful than either headline.
Neither study is a smoking gun and neither research team claims otherwise. Ahrefs states plainly that correlation is not causation and that every factor they tested came back moderate to very weak. That caveat matters, and it is exactly why the reconciliation is worth doing carefully rather than picking whichever coefficient supports the budget you already wrote.
The two studies everyone quotes at each other
Start with what each one actually put on the x axis, because that is where the disagreement dissolves. Ahrefs ran a Spearman analysis across 75,000 brands filtered to domains above DR 40, matched against millions of AI Overview responses, and asked which brand and link metrics track with getting mentioned. Roughly 74 percent of the brands in that set had at least one AI Overview mention, so the sample is not a study of whether AI mentions happen. It is a study of who gets more of them.
Semrush took 1,000 randomly selected domains across ChatGPT, ChatGPT with Search, Gemini, Google AI Overviews and Perplexity, and asked a coarser question first: does a domain with a strong backlink profile show up in AI answers at all. That question returns 0.65. Then they asked the finer question, which is how much of the answer set a domain owns once it is already appearing. That one returns roughly 0.23. Same data, same vendor, two very different numbers, and the discourse has almost entirely adopted the first one.
| STUDY | WHAT IT MEASURED | HEADLINE NUMBER | WHAT IT ACTUALLY SUPPORTS |
|---|---|---|---|
| Ahrefs, 75,000 brands, May 2025 | Spearman correlation of link and brand metrics against AI Overview brand mentions | Backlinks 0.218, branded web mentions 0.664 | Mentions track AI visibility about three times as closely as raw backlink counts do |
| Semrush, 1,000 domains, October 2025 | Authority Score against whether a domain is mentioned at all across five AI surfaces | Pearson 0.65 | Strong link profiles are strongly associated with being in the candidate set |
| Semrush, same dataset | Authority Score against share of voice once a domain already appears | Pearson 0.23 | Additional authority stops paying once you are inside the set |
| Ahrefs, same dataset | Referring domains rather than raw backlink count | 0.295 | Domain count is a slightly better proxy than link count, and still weak |
| Semrush, same dataset | Follow links versus nofollow links against AI visibility | 0.334 versus 0.340 | The attribute buyers negotiate hardest for is not the one associated with the outcome |
Read down that table and the story stops being a fight between two vendors. It becomes one finding stated twice: link equity is associated with admission, and weakly associated with everything after admission. Ahrefs measured mostly the after, and got 0.218. Semrush measured the admission and got 0.65, then measured the after and got 0.23. The numbers were never in conflict. The framing was.
Backlinks and AI search visibility: a gate, not a dial
A gate is binary and a dial is continuous, and almost every link building budget in existence is priced as though the thing being bought is a dial. More links, more visibility, linearly, forever. The data supports a different shape. There is a threshold effect on the way in, and then a long flat stretch where the marginal link buys progressively less of the thing you actually want.
Spearman correlation with AI Overview brand visibility across 75,000 brands, Ahrefs, May 2025. Published coefficients rendered on a 0 to 100 scale for comparison. The study's own caveat applies: these are associations, not proven causes, and all of them are moderate at best.
Look at what sits at the top of that chart. Branded web mentions at 0.664 and branded anchors at 0.527 are both measures of how often the web says your name. Domain Rating, referring domains and raw backlinks, the three metrics that every link vendor prices against, cluster at the bottom between 0.33 and 0.22. The gap between the top of the chart and the bottom is roughly three to one, and it maps almost perfectly onto the difference between being talked about and being linked to.
This is not an argument that links stopped working. Ahrefs ran a controlled disavow study in 2024 in which removing links produced traffic drops of 13.3 to 18 percent, and restoring them recovered 99 to 144 percent of that traffic. Classic organic search still runs on links and will keep doing so. The claim here is narrower and more awkward: the specific outcome you are buying links for in 2026, appearing and persisting inside generated answers, is the outcome links are worst at delivering per dollar. We saw the same separation between surfaces when we pulled apart how Reddit citations and Reddit referral traffic move independently, and the lesson is the same one: two things that used to rise together now have to be funded and measured separately.
The follow attribute is the most expensive thing on the invoice
Here is the finding with the most direct effect on a purchase order, and it is the one that gets the least airtime. In the Semrush data, nofollow links correlate with AI visibility at Pearson 0.340 and Spearman 0.509. Follow links come in at 0.334 and 0.504. Nofollow is nominally ahead on both measures, which nobody should overinterpret, but the honest read is that the two are indistinguishable. For the AI visibility half of the job, the rel attribute is not doing work.
| LINK TYPE | PEARSON | SPEARMAN | WHAT IT MEANS AT THE POINT OF PURCHASE |
|---|---|---|---|
| Follow links | 0.334 | 0.504 | The default request in every outreach template, and the thing placements are priced against |
| Nofollow links | 0.340 | 0.509 | Statistically indistinguishable from follow for AI visibility in this dataset |
| Image links | 0.415 | 0.538 | The strongest link format in the study, and almost nobody asks for one |
| Text links | 0.334 | 0.472 | The format every brief specifies by default, and the weaker of the two |
Now put that next to the market. Across a survey of 500 SEO professionals published by Reporter Outreach, 76 percent said they would pay 300 dollars or more per link, and 31 percent sit in the 500 to 1,000 dollar band. A meaningful share of that premium is being paid for a follow attribute on a text link. If the buying goal is classic rankings, that premium is defensible and the follow attribute genuinely matters. If the buying goal is showing up in AI answers, and for a growing number of budgets it now is, the premium is buying an attribute the data cannot distinguish from its free alternative. A brand mention in a well-read publication with a nofollow link, or no link at all, tests roughly as well on this outcome as the placement you paid a premium to make follow. The structural work that makes a page quotable in the first place, which we set out in the enterprise standard for structured data in AI search, does more for the same money.
“You are paying a premium for an attribute that tests as neutral, on the surface you bought the link for.”
What the industry says it believes, and what it does
The most telling numbers in this whole picture are not correlations at all. They come from the same survey of 500 SEO professionals, and they describe an industry that has updated its beliefs without updating its invoices.
That fourth card is the resolution hiding in plain sight. Nobody has to be talked into digital PR. It is already the top rated tactic by the people doing the work. What has not caught up is the unit of account: campaigns are still scoped, priced, reported and renewed on link counts, so the mention that a campaign generates is treated as a nice side effect rather than the primary asset. The asset and the receipt have swapped places, and the invoice has not noticed.
How to budget backlinks for AI search visibility
None of this argues for cancelling link building. It argues for splitting one budget line into two, because it has been funding two different outcomes under one name. Our generative engine optimization work treats them as separate line items with separate measurement, and the pattern held on the Zenity engagement, where the coverage that moved citation share and the links that moved classic rankings came from overlapping but not identical work.
One honest caveat before you take this into a planning meeting. Both studies are correlational, both were run by companies that sell tools measuring the metrics they studied, and neither can tell you what happens inside the engines. The Ahrefs authors say the correlation and causation point themselves. What makes the conclusion usable anyway is that the two independent datasets converge on the same number for the same question, from different directions and different samples. Convergence is not proof. It is the strongest thing available while the engines stay closed, and it is a much better basis for a budget than the single cherry-picked coefficient that currently anchors most of these conversations. The original write-ups are worth reading in full: the Ahrefs analysis of 75,000 brands and the Semrush backlinks and AI search study.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.