"Build more backlinks" is one of the most repeated pieces of SEO advice, and it rests on a correlation anyone can see: sites with more links tend to rank and get more traffic. That part is true. But "tend to" hides the real question, and almost nobody answers it with data: does getting more links cause the traffic, or does the kind of site that earns traffic also happen to earn links along the way? Those are very different businesses to run, and only one of them is worth paying an agency for.
This is original research from Stackra's own corpus, joined against two independent public datasets, no proprietary data. Full methodology, every honest limitation, and the exact numbers are below.
The Data
Three sources, joined on registrable domain.
- •Link authority: Common Crawl's webgraph, three annual releases (2024: 173M domains, 2025: 162M, 2026: 121M). Counts how many distinct domains link to each target.
- •Traffic: the Chrome UX Report (CrUX), sampled quarterly for 14 consecutive quarters (January 2023 through April 2026), giving every site with real Chrome traffic a popularity rank.
- •Business classification: 1.15 million US business websites, classified by industry and age.
Two honest limits, stated up front rather than buried in a footnote. CrUX measures all Chrome traffic (organic, direct, paid, referral combined), not organic search specifically, so the headline result below is a conservative, diluted read of the link-to-organic effect. And Common Crawl's link graph is noisy release to release, which we handle by trusting patterns that survive the noise, never a single-site delta.
Finding 1: The Correlation Is Real, and It Survives Controls
Among 9.68 million domains, the odds of reaching real traffic climb in a clean, monotonic staircase as referring-domain count rises. This holds up as an independent effect, not an age or industry proxy: a controlled logistic regression on the US corpus, holding age, business type, CMS, schema markup, and local-business status constant, scores AUC 0.928. Authority's effect on the odds (roughly 180x across tiers) dwarfs age's independent effect (roughly 1.4x), though the two interact: the same referring-domain tier pays off at an 18% real-traffic rate for established sites versus 2% for brand-new ones. Authority is necessary but not sufficient. Its payoff is age-gated.
| Referring domains | Reach real traffic |
|---|---|
| 0-5 | 7% |
| 21-100 | 23% |
| 101-1,000 | 58% |
| 1,000+ | 83% |
World corpus, 9.68 million domains. Source: Stackra analysis of Common Crawl webgraph joined to CrUX traffic rank.
Finding 2: Direction Is Genuinely Ambiguous
A naive cross-lagged panel model on 5.07 million domains across three annual snapshots (2024, 2025, 2026) looked like traffic leads links by 15.5x. That ratio turned out to be a measurement artifact of how sticky each variable is: link percentile barely moves year to year (autocorrelation 0.89-0.91) while traffic rank moves far more (autocorrelation 0.56-0.80), so a naive comparison mostly measures which variable has more "new" year-to-year variance to explain, not which one is doing the explaining. The standard fix for exactly this problem is a partial correlation that controls for each variable's own prior-period level before measuring the cross-lag coefficient. Rerun that way, on the same 5.07 million domains: links predicting next-year traffic scores a partial correlation of -0.081, traffic predicting next-year links scores -0.069, a 1.17x difference instead of 15.5x. The two directions are close enough to call comparable, with links leading only slightly. That alone should make anyone selling "links first, growth follows" nervous.
Finding 3: Buying Links Without a Business Behind Them Fails, Hard, at Scale
We identified 60,291 sites that spiked their referring domains while sitting at zero real traffic, the group most likely to prove links cause growth if the pitch were true. We tested it across 12 different threshold variants (three gain multiples, four ending link-count bands) so no single cutoff could be doing the work.
An earlier pass of this analysis claimed 76-79% of this cohort "lost" the acquired links within a year, framed as link decay. That number does not survive a control comparison: sites with stable link counts over the same window show the identical loss rate. It was measuring Common Crawl's normal crawl noise, not anything specific to manipulated sites. We pulled the claim rather than keep a wrong stat on the page.
| Cohort | Sites | Reached real traffic |
|---|---|---|
| Full manipulation cohort, 12 threshold variants | 60,291 | 0.4% or less, every variant |
| Best-performing single cell in the entire table | 779 | 0.26% |
| Tier-stratified rerun, all starting traffic tiers | 118,660 | flat, no relationship below the top tier |
Source: Stackra analysis of Common Crawl webgraph (2024-2026 releases) joined to CrUX traffic rank, May-July 2026.
The plain-English explainer this finding sits inside: what a backlink and a referring domain actually are, and why 'more' is not automatically better.
A decision guide built directly on this finding, for anyone currently being pitched a link-building package.
Finding 4: There Is a Sweet Spot, and It Moves With the Site
For sites that already have real traffic and real links, the dose-response is an inverted U, not a straight line. A separate two-year panel isolated sites with no visible traffic in 2025 and tracked what happened by 2026 as their referring-domain count changed: breakout rate (crossing into a higher traffic bucket) rose in a clean, monotonic line from 0.82% at zero link change up to 7.49% at +1,000 referring domains gained, a roughly 9x relative improvement. Real, but 92.5% of even the biggest link gainers stayed stuck. Links help at the margin. They are wildly insufficient alone.
| Referring domains | Improvement rate |
|---|---|
| 101-300 | 34% (peak) |
| 300+ | declining |
The 'right' link count shifts higher at every tier up. At the very top tier, there is no ceiling in the data at all: more real editorial links keep helping.
Finding 5: We Checked Our Free Data Against a Paid Tool, and the Story Held
Common Crawl is free and public, which invites a fair question: how good is it, really? We cross-checked 500 real SaaS domains against DataForSEO, a commercial backlink index built from a completely separate crawl. The ranking agreed 94% of the time (correlation 0.94). The raw count ran 4 to 9 times lower depending on site size, in a predictable, correctable way: fitting a power law across the 499 clean domains gives DFS_rd ≈ 19 × (CC_rd)^0.76 (R² = 0.898), with residual scatter of about ±1.66x at one standard deviation. Then we did the harder thing: isolated organic search traffic specifically, the one channel links are actually supposed to move, for 100 real small SaaS businesses across three years of history (199 domain-year transitions), and reran the whole cross-lagged test on that channel alone. Same answer, more starkly: the organic-leads-links partial correlation came back +0.20, versus +0.01 (indistinguishable from zero) for links-leads-organic, positive for organic at every traffic tier and negative for links at every tier except the very top (+0.15 there).
That last check is the one that matters most to a skeptical reader. The obvious objection to everything above is "your traffic metric includes non-organic traffic, of course links don't look causal." We tested that objection directly, on the exact channel it targets, isolating organic search specifically, and it did not hold up.
What This Actually Means
Links are downstream of something else, most likely brand strength, content quality, and product-market fit: the things that make a site worth visiting in the first place. Traffic tends to respond slightly faster than links do, which is exactly what you would expect if both are symptoms of the same underlying cause rather than one causing the other.
- •Real, organic links do help at the margin, especially for a genuine business stuck at the bottom of the traffic distribution. The sweet spot is real, and it moves with your traffic tier: 101-300 referring domains at the bottom tier, higher at every tier up.
- •Manufacturing a link spike without the business to back it up is not a growth lever. It shows up in the data as clearly as it would in a manual review.
- •If you are being pitched a link-building package right now, the 60,291-site finding above is the exact scenario that package is selling. It failed 99.6%+ of the time.
Where do you actually stand? Look up your own referring-domain count with any backlink checker (including free ones, since our own cross-check shows rank-order holds up even where the raw count runs low) and place it against the table in Finding 1. Under 20 and stuck, you are below the inflection point entirely, no amount of manufactured links will substitute for the traffic-earning work. Past 100, you are in the range where the effect is real.
How We Built This
This research combines Stackra's own corpus with two independent public datasets. Here is exactly how it is built, so anyone can weigh how much to trust it.
- •Link authority: Common Crawl's domain-level webgraph, three annual releases (2024: 173M domains, 2025: 162M, 2026: 121M), raw in-degree (count of distinct linking domains) verified against 3.9 billion raw edges, 500/500 exact match.
- •Traffic: Chrome UX Report (CrUX), 14 consecutive quarterly snapshots, January 2023 through April 2026, via HTTP Archive's public BigQuery tables.
- •Business classification: Stackra's own corpus of 1.15 million verified US business websites, industry and age labeled.
- •Cross-sectional and controlled-model analysis: 9.68 million world domains for the threshold curve; US-corpus logistic regression (AUC 0.928) for the controlled version.
- •Cross-lagged panel model: 5.07 million domains with data in all three annual snapshots, partial correlations controlling for each variable's own prior-period level (the standard fix for autoregressive persistence in a CLPM).
- •External validation: DataForSEO's commercial backlink index, 500-domain cross-check for rank correlation and count calibration (499 used after dropping one non-representative outlier), plus a 100-domain, three-year historical organic-traffic pull (199 domain-year transitions) for the channel-isolated cross-lag test.
- •Every date, crawl partition, and release name was verified against that source's own fixed metadata (GCS folder names, dated BigQuery columns, documented crawl windows), not inferred by comparing datasets against each other.
See Where Your Own Site Stands
Curious whether your own site's link profile looks like the sweet spot or the spike? Stackra runs a full audit, no backlink purchase required.
Frequently asked questions
Common questions about this research and what it means for your link-building decisions.
Do backlinks actually help SEO?
Real, organic links from relevant sites correlate with better outcomes, especially for a genuine business stuck at the bottom of the traffic distribution, where 101-300 referring domains shows the strongest improvement rate in this data. But the correlation runs both ways: sites that already attract traffic also tend to attract links, so more links is not a guaranteed lever, it is a marker that often travels with one.
Does buying backlinks work?
Not based on this data. Among 60,291 sites that spiked their referring domains while having zero real traffic to begin with, tested 12 different ways, 99.6%+ never reached meaningful traffic. The single best-performing cell in the entire table still only succeeded 0.26% of the time.
How many backlinks does a website actually need?
This data does not support a single target number. At the bottom traffic tier, the peak improvement rate sits at 101-300 referring domains; past that, more links start correlating with worse outcomes, not better. The right range moves higher at every traffic tier up, and at the top tier there is no ceiling at all in the data.
Is Common Crawl's link data accurate?
Cross-checked against DataForSEO, a paid commercial backlink index, on 500 real domains: rank-order agreement was 94% (correlation 0.94). The raw count runs 4 to 9 times lower than DataForSEO's, following a power law (DFS_rd is about 19 times Common Crawl's count raised to the 0.76 power, R² = 0.898), so which sites have more links relative to each other holds up well even though the absolute count needs correcting.