Almost every link building statistic you have read this year traces back to a study published in 2018 or 2019. Several of the most repeated ones say the opposite of what the original research found.
We went back to the primary sources, checked what they actually measured, and then ran our own analysis of how Google's AI Overviews pick which pages to cite. This page is the result: the numbers that survive scrutiny, the ones that don't, and what the 2026 data says about where link building is heading in 2027.
Why most link building statistics are unreliable
There is a specific failure mode in this corner of the industry. A roundup cites a second roundup, which cites a third, which eventually cites a real study. By the time the number arrives on the page you are reading, the date has been stripped, the methodology has been lost, and the caveat the original author wrote has been deleted.
We checked the pages currently ranking for "link building statistics" in the US. The top ten results carry titles promising 38, 50, 55, 70, 80 and 105 statistics respectively. What almost none of them carry is the publication date of the study behind each figure.
That matters because link building in 2019 and link building in 2026 are not the same activity operating in the same search environment. A correlation measured before AI Overviews existed cannot tell you anything about a SERP where an AI answer sits above every organic result.
The three studies almost everything traces back to
Strip the roundups away and a surprisingly small number of primary sources are doing the work.
| Study | Published | Sample | What it actually measured |
|---|---|---|---|
| Ahrefs, backlink growth and rankings | Sept 2018 | 10,000 keywords, 200,000 pages | Rate of new referring-domain acquisition vs SERP position |
| Ahrefs, links from pages with traffic | Feb 2019 | 44,589 keywords, top 10 each | Whether links from trafficked pages correlate better than links generally |
| Aira, State of Link Building | Third annual edition | 270 SEO practitioners | Self-reported tactics, budgets and priorities |
Two of the three are now six and seven years old. The third is a practitioner survey, which measures what SEOs say they do — useful for understanding the market, useless as evidence that a tactic works.
What the primary sources actually say
Here is where it gets uncomfortable. Several widely repeated statistics are not simplifications of the original research. They are reversals of it.
"Links from pages with traffic are more valuable"
This one is repeated constantly, usually with a link to the Ahrefs study that supposedly proves it. That study, published 19 February 2019 across 44,589 non-branded keywords, concluded the opposite.
The number of backlinks from pages with traffic correlated slightly less with rankings than total backlink count. The number of referring domains with traffic did not correlate more strongly than total referring domains. Tim Soulo, who ran it, wrote that he had expected the reverse and that the data did not support it.
The metric that did correlate most strongly across the whole study was URL Rating — a page-level authority score that barely appears in any statistics roundup.
The study also produced a genuinely striking figure that almost nobody quotes: in only one SERP in every 2,229 did every top-ranking page have at least one link from a page with organic traffic. Roughly a fifth of SERPs had no such page at all.
"Top-ranking pages gain X new links per month"
The source is from 18 September 2018. It found that most pages ranking first grew their followed referring domains by between 5% and 14.5% per month, with the median growth rate declining as position declined.
The part that gets cut from every retelling is the second finding: the study observed no statistically detectable relationship between a page's backlink growth rate and whether its ranking actually moved. Links correlated with position. Link velocity did not predict change in position.
That distinction is the whole ballgame for anyone buying links on a monthly retainer, and it is almost universally dropped.
The study is also internally inconsistent in a way worth knowing about. Its methodology describes three-month measurement windows, while its headline figures are expressed per month. The author flags that the referring-domain graph includes nofollow links and recommends treating it as a rough estimate only.
What survives
Stripping out everything that fails inspection leaves a short, defensible list:
- Referring domains correlate with organic traffic at scale, across a dataset of roughly 920 million pages. This is the most robust finding in link building and it has held up across repeated measurement.
- Page-level authority correlates more strongly than domain-level authority. URL Rating beat every domain metric in the 2019 study.
- Correlation is not causation, which every one of these studies states explicitly in its own text and which no roundup reproduces.
That is a thinner evidence base than the industry's confidence implies. If you have been told that a specific number of links at a specific authority threshold produces a specific ranking outcome, nobody has demonstrated that.
Our own data: AI Overviews cite pages that don't rank
Because the existing research predates AI Overviews entirely, we ran our own analysis on 23 October 2026.
The question was simple. When Google generates an AI Overview, does it cite the pages that rank underneath it, or does it reach somewhere else? If it reaches somewhere else, then ranking first and being cited first are two different jobs, and link building has to serve both.
We took eight commercial US SERPs — a mix of link building, agency, SaaS software and definitional queries — and compared every domain cited in the AI Overview against the domains occupying the organic top 10 for that same query.
All eight returned an AI Overview. Across them, the AI cited 48 domain-level sources. Twenty-six of those also appeared in the organic top 10. Twenty-two did not.
Who gets cited without ranking
The domains the AI reached for but Google's organic results did not are the interesting part. A pattern shows up quickly.
On best crm software, the AI Overview cited G2, Forbes and PCMag. None of the three appeared in the organic top 10. On project management software, it cited G2 and Wikipedia, neither of which ranked. On best seo tools, four of the five cited sources were absent from the top 10 — including two sites most SEOs have never heard of — while Backlinko, Semrush and Surfer all ranked and none were cited.
Two categories are overrepresented among the cited-but-not-ranking group:
- Review aggregators and directories. G2, Forbes Advisor, PCMag, Capterra-style platforms. These are structurally designed to be quoted: named entities, explicit comparisons, clean lists.
- Small, highly specific pages. On link building services the AI cited a DR 18 blog post. On what is domain authority it cited a UK agency blog and an Australian one, in a US SERP.
Neither group is winning on authority. They are winning on being easy to extract a sentence from.
What this does not prove
This is an eight-SERP snapshot, not a large-scale study, and we would rather say so than dress it up. Forty-eight citations is enough to show a pattern exists and nowhere near enough to size it precisely. The overlap ranged from 20% to 75% across individual queries, which is wide variance on a small sample.
What it does establish is that the question is real. If AI citation and organic ranking were the same thing, we would expect overlap near 100% and we measured 54%.
Which link building strategies actually have evidence behind them
Most "best tactics" lists are rankings of popularity, not effectiveness. The Aira survey, for instance, reports what 270 practitioners say they use. That tells you what the market does. It does not tell you what works.
Sorted by the strength of the evidence rather than the strength of the claims:
| Strategy | Evidence quality | What the data supports |
|---|---|---|
| Earning links from topically relevant domains | Strong | Referring domain count correlates with organic traffic across ~920M pages. Relevance is the consistent qualifier across every ranking-factor analysis. |
| Digital PR / original data | Moderate | Produces links from domains that cannot be bought. Evidence is largely case-study level, not controlled. |
| Competitor gap analysis | Moderate | Sound as a prospecting method. No study shows gap-sourced links outperform other links. |
| Broken link building, resource pages | Moderate | Works because it offers the publisher something. Reply rates are well documented; ranking impact is not. |
| Guest posting at volume | Weak | Widely used, poorly evidenced. Google's link spam systems explicitly target it at scale. |
| Hitting a monthly link quota | Contradicted | The 2018 study found no detectable relationship between link growth rate and ranking movement. |
| Buying links on marketplaces | Contradicted | Algorithmically neutralised rather than penalised, which is why it looks like nothing happens. |
The bottom two rows are where most link building budget goes. That is the uncomfortable part of reading the evidence honestly.
If you are currently working to a monthly link target, the link velocity calculator will show you what your acquisition rate looks like relative to your existing profile — which is the measure the 2018 data actually supports, rather than a raw monthly count.
The statistics page paradox
One more finding from our own SERP data, because it reframes why pages like this one exist.
We pulled the full US SERP for "link building statistics". The combined organic traffic of the entire top ten is roughly 250 visits per month. These are not traffic assets.
What they are is link assets. BuzzStream's statistics page carries 224 referring domains. SearchLogistics' carries 117. DemandSage's carries 139. A dedicated statistics microsite in that SERP has accumulated 694 referring domains on a domain with a DR of zero.
And the page currently ranking second is on a domain with a DR of 16, outranking sites at DR 77, 78, 90 and 91 for the same query.
Two conclusions follow. First, authority is not what wins this SERP, which should tell you something about how much of your strategy should rest on domain-level metrics. Second, statistics pages are published to be cited, not to be visited — and almost nobody measures them that way.
Where this goes in 2027
Here is the thesis, stated plainly so you can disagree with it.
A link's value is splitting into two separate currencies
For twenty years a link did one job: it passed authority that moved a ranking. Every metric we built — DR, DA, UR, link velocity — measures that single job.
Our citation data suggests a second job has appeared that the first set of metrics does not measure. Being the source an AI answer reaches for is not the same as ranking, because 46% of the time the cited source was not in the top 10 at all. Whatever qualifies a page for citation, organic position is not a reliable proxy for it. We go deeper into the practical side of this in our guide to AI link building.
The two currencies have different inputs:
- Ranking currency accrues from authority and relevance, and the existing correlational evidence still holds. This is not going away.
- Citation currency accrues from being structurally quotable and independently corroborated. A model reaching for a claim appears to favour sources that state it cleanly and sources that other sources agree with — which is why directories, review aggregators and tightly scoped pages keep appearing above better-linked competitors.
What changes in the brief
The 2026 link building brief reads: get me N links at DR X. We expect the 2027 brief to read: make me the corroborated source on topic Y.
That is not a rhetorical difference. It changes the unit of measurement from link strength to coverage breadth — how many independent domains state the same verifiable fact about you. One DR 80 link and eight DR 30 mentions are roughly equivalent in the old currency. In the citation currency they are not close, because corroboration across independent sources is the thing being rewarded.
Three things to do about it now
- Make your claims extractable. If the fact a model needs is buried in a paragraph of narrative, it is harder to lift than the same fact in a table with a named subject. This is the cheapest change available and it has nothing to do with links.
- Count domains that state your fact, not links that point at your page. An unlinked mention that corroborates a claim may be worth more in the citation currency than a followed link that says nothing specific. Most reporting cannot see this at all.
- Stop reading monthly link counts as progress. The only study that measured velocity against ranking movement found no relationship. If your reporting leads with links built, it is leading with the metric with the weakest evidence behind it.
What we think stays true
Relevance keeps beating volume. Links from domains that would plausibly cover your topic unpaid keep outperforming links from domains that would not. Nothing in the AI data changes that; if anything, corroboration rewards it more heavily, because a relevant source is a more credible corroborator.
And the underlying mechanism — independent sites vouching for a page — is not a Google implementation detail. It is how both ranking systems and language models decide what to trust. That is why we do not expect links to stop mattering. We expect the measurement of them to change faster than the practice does.
Methodology and limitations
So you can judge the numbers rather than take them on trust.
- Our AI Overview analysis: eight US commercial SERPs pulled on 23 October 2026 via the Ahrefs API, spanning link building, agency, B2B SaaS and definitional queries. For each, we recorded every domain cited in the AI Overview and every domain in the organic top 10, then measured overlap at domain level. Sample: 48 cited domains. Results reported both including and excluding YouTube.
- Known limitations: small sample; US only; single point in time; AI Overviews are personalised and volatile, so another pull on another day will differ. Domain-level matching treats two URLs on one domain as one source. We report a pattern, not a population estimate.
- Third-party figures are cited to the primary study with its publication date, not to a roundup. Where a study's own conclusion conflicts with the way it is commonly quoted, we have said so.
- What we did not do: we did not run a controlled experiment, and neither did any study cited here. Everything on this page is correlational.
Frequently Asked Questions
How many backlinks do I need to rank?
No study has established a number, and any source quoting one is extrapolating. The defensible version: referring domain count correlates with organic traffic at scale, the right target depends on what the specific SERP already has, and page-level authority correlated more strongly than domain-level authority in the largest study to test both.
Are backlinks still a ranking factor in 2026?
Yes. The correlational evidence between referring domains and organic traffic remains the most robust finding in the field. What has changed is that ranking is no longer the only outcome worth measuring.
Do links help with AI Overviews and AI search?
Indirectly, and not in the way most people assume. In our sample, 46% of AI Overview citations went to domains outside the organic top 10, so ranking well is not sufficient and may not be necessary. Corroboration across independent sources appears to matter more than the authority of any single link.
Is link velocity a real ranking factor?
The one study that measured it found no detectable relationship between backlink growth rate and ranking movement, despite finding a correlation between link count and position. Velocity is better understood as a profile-health signal than a ranking lever.
Why are most link building statistics unreliable?
Because the industry cites roundups instead of studies. The two most-quoted datasets are from 2018 and 2019, publication dates are routinely stripped, and at least one widely repeated claim states the opposite of what its source study concluded.
What should I measure instead of links built?
Referring domains rather than raw backlinks, relevance of the linking domain to your topic, page-level authority of the specific target, and — increasingly — how many independent domains state the same verifiable claim about you.
