Skip to content

Link building: what the data says about how much links still matter

Author: Matteo Pellegrini

Link building is the work of getting other websites to link to yours. The premise is as old as Google: a link is a vote, and whoever collects more of them from trustworthy sources climbs the rankings.

The premise still holds, but far less than most guides suggest. At Pubcon Pro in Austin, in September 2023, Google's Gary Illyes said people overestimate the importance of links and that they are not among the top three ranking factors: "they haven't been for some time", Search Engine Land reported on 25 September 2023. Meanwhile a second visibility channel has opened up, the answers generated by language models, where links count for even less.

Below are the numbers from three studies on large datasets, and what changes in day-to-day work if you take them seriously.

It is the part of off-page SEO that deals with inbound links: links that start on a domain other than yours and land on one of your pages. If the split between the two areas is not clear, start with the comparison between on-page and off-page SEO. It should not be confused with internal links, which you place yourself between the pages of your site to spread relevance within it.

Three elements define an inbound link: the domain it comes from, the page that hosts it and the clickable text, the anchor text. On top of these sits the rel attribute, which tells Google how to treat it: nofollow for links you do not vouch for, sponsored for paid ones, ugc for comments and user-generated content.

The trade started out as indiscriminate collecting and was cut down to size in April 2012, when Google released Penguin and began hitting link profiles built to order. Since then the question has not been how many links you have, but where they come from. The theory is in our guide to backlinks; this article is about what to do.

The most awkward figure for anyone selling link packages comes from Ahrefs. Tim Soulo analysed 44,589 non-branded keywords with 2,000 to 5,000 monthly searches, looking at the top 10 pages and at the links they receive from pages that themselves get organic traffic.

The result: only one SERP in 2,229 had every page in the top position backed by links from pages with traffic. In about 20% of the SERPs analysed, none of the top 10 pages had a single one. The study, published on 19 February 2019 and updated on 18 December 2025, concludes that you can rank on page one without that kind of link. The authors point out that correlation does not prove causation, and it is worth repeating.

Anyone working in B2B niches sees this every day. In the Macropix case we published in our guide to SEO for manufacturing, the Milan-based LED wall manufacturer went from position 88 to position 2 for "monitor pubblicitario" (advertising display) on Google Italy, in a market where the main keyword, "ledwall", gets 4,400 searches a month. It did not happen by buying links: it happened because no competitor covered that technical vocabulary. Other ranking factors outweigh the link profile when competition is weak, and weak competition is the normal state of B2B search.

In May 2025 Louise Linehan and Xibeijia Guan published an analysis for Ahrefs of 75,000 brands, using Spearman's coefficient to measure which signals correlate with a brand's visibility in Google's AI Overviews. The table below is the full ranking.

SignalCorrelation with AI Overview visibility (Spearman)
Branded web mentions0.664
Branded anchor text0.527
Branded search volume0.392
Domain Rating0.326
Referring domains0.295
Brand organic traffic0.274
Total backlinks0.218
Number of site pages0.170
Source: Ahrefs, An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied), 26 May 2025. Spearman correlation coefficient across 75,000 brands.

A brand mention, even without a link, correlates three times more strongly than total backlinks. In the same study, 26% of the brands analysed never appear in AI Overviews, and brands in the top quartile for web mentions collect up to ten times more appearances than those in the quartile below.

Semrush reached the same place by another route. On 16 October 2025 Margarita Loktionova, with Aleksandr Drozdov and with Kevin Indig of Growth Memo, published a study of 1,000 randomly sampled domains, tracking mentions across five platforms: ChatGPT, ChatGPT with Search, Gemini, Google AI Overviews and Perplexity. Authority Score correlates with AI mentions at 0.65 (Pearson).

The detail that undermines half of the link market is another one. Against mentions in AI answers, followed links correlate at 0.504 (Spearman) and nofollow links at 0.509. They are the same thing. If a nofollow link is worth as much as a dofollow one for getting cited by a model, the premium you pay for dofollow buys only half the benefit. For the rest of the game, see our guide to generative engine optimisation and the one on how to appear in ChatGPT answers.

Domain metrics (Domain Rating, Authority Score, Domain Authority) are third-party estimates, not Google numbers. They help you rule things out quickly, not decide. Five questions decide.

  • Does the site get organic traffic on topics close to yours? A domain with high metrics and zero traffic on relevant subjects is an inflated domain.
  • Is the page hosting the link indexed? Look it up on Google with the site: operator. If it is not there, the link does not exist for anyone.
  • How many outbound links does that page have to unrelated domains? A page linking to a dentist, a casino and a window manufacturer is a page that has been sold.
  • Does the link bring people? Referral traffic is measured in GA4 and separates a useful link from a decorative one.
  • Is the anchor natural? A healthy profile is dominated by brand-name anchors and bare URLs, not by the same exact-match keyword repeated.

The fifth question is the one people get wrong most often, and we covered the consequences in the article on black hat SEO techniques.

What actually works for B2B companies

Almost every company in a supply chain has a website, and many use social media for marketing. Suppliers, distributors and installers share customers and topics, yet they hardly ever link to each other. It is the most overlooked source of links there is, and it costs nothing.

Supply chain pages. Suppliers, distributors, resellers, certified installers. Every manufacturer has a list of partners who publish a "where to buy" or "our suppliers" page. Asking to be listed there takes a phone call, not a campaign.

Trade associations, industry bodies, clusters, trade fairs. Exhibitor profiles at a trade fair stay online for years and live on domains that Google considers trustworthy. The same goes for trade association member directories and the catalogues of regional business networks.

Data nobody else has. This is the slowest method and the only one that earns links without asking for them. A survey of your own orders, an average industry price list, delivery times measured over three years: anyone writing about that topic has to cite the source, and the source is you. It also works because it is the kind of content language models pick up with a citation, as the table above shows.

Unlinked mentions. If someone names you without linking, the instinct is to ask for the link. The Ahrefs data says the mention on its own is already the signal most correlated with AI visibility. Asking for the link is still worthwhile, but it is no longer the main point of the call.

Vertical digital PR. A trade publication with 4,000 readers in your market is worth more than a national newspaper with a million readers who do not buy what you sell. The rule applies to SEO and to lead generation alike.

Google's spam policies for Google web search, in the version updated on 15 May 2026, define link spam as creating links with the primary purpose of manipulating rankings, and list among the violations: exchanging money for links or for posts that contain links, exchanging goods or services for links, sending free products in exchange for reviews that include a link, guest posts and press releases with optimised anchors, footer or template links distributed across multiple sites, links in widgets, and forum comments with built anchors.

This is not a grey area open to interpretation. It is the list of things almost every link package on sale contains. Google also states the remedy: paid links should carry rel="sponsored" or rel="nofollow", and at that point the link passes no ranking value. Anyone selling paid dofollow links is, by definition, selling a declared violation.

There is a second front that affects the advertorial market directly. The site reputation abuse policy, in force since March 2024, was clarified by Google on 19 November 2024. Chris Nelson of the Search Quality team wrote that using third-party content on a site to exploit that site's ranking signals violates the policy "regardless of whether there is first-party involvement or oversight", as Barry Schwartz reported on Search Engine Land. Enforcement comes through manual actions visible in Search Console. The column bought on a local newspaper site, with the piece written by the agency and hosted in a section the paper does not oversee, is exactly the case described.

There is no number, and anyone who gives you one is making it up. There is a benchmark: the median number of referring domains of the pages ranking ahead of you for that single keyword. Not the domain, the page. It takes two minutes to read with any tool, and it varies tenfold between a niche informational query and a contested commercial one.

On timing there is no reliable public data to quote, so we do not quote any. The only verifiable point is that a link starts to count once Google has crawled the page hosting it, and you can read that date in the crawl report.

What to measure

  • Referring domains, not total backlinks. Fifty links from the same domain are still one domain.
  • Brand mentions over time, with and without links. It is the signal at the top of the table.
  • Rankings for the keywords you have worked on, not the site average.
  • Referral traffic in GA4, to see which links bring people and not just signal.
  • Presence in AI answers, which is now a separate channel: the AI SEO statistics give the order of magnitude.

If you are weighing up a quote, the questions to ask whoever sends it are the ones we list in the guide on what an SEO agency really does. Our approach to off-page work is part of our SEO service, and the measured results are in our projects.

What nobody puts in a quote

The market keeps selling links not because they work better, but because they are the only off-page activity that fits neatly into a quote. Ten links, a fixed price, delivery in sixty days: it can be invoiced, reported and closed. Producing the data that makes a company worth citing has no line in a price list, no delivery date and no predictable return. As long as buyers judge on what can be reported rather than on results, link building will stay a recurring cost instead of an asset that builds up. The interesting part is that language models are rewarding precisely the part that cannot be invoiced.

Is link building still worth doing?

Yes, but it carries less weight than people claim. In 2023 Google's Gary Illyes said links are not among the top three ranking factors. In the Ahrefs study of 75,000 brands, referring domains correlate at 0.295 with AI Overview visibility, against 0.664 for brand mentions. Links remain a useful signal, no longer the main one.

Is buying backlinks against Google's rules?

Yes. Google's web search spam policies explicitly list exchanging money for links, or for posts that contain links, as link spam. Google states the remedy: paid links should carry rel="sponsored" or rel="nofollow", which however removes the ranking value the link was bought for.

Are nofollow links any use?

For visibility in AI answers, yes, as much as followed links. In the Semrush study of 1,000 domains published on 16 October 2025, followed links correlate with AI mentions at 0.504 (Spearman) and nofollow links at 0.509. For classic Google rankings, the followed link is still the one that passes signal.

Is it better to have a few strong links or many weak ones?

A few relevant ones. The useful test is not the domain metric but whether the linking site gets organic traffic on topics close to yours, whether the page hosting the link is indexed and how many outbound links it has to unrelated sectors.

How many backlinks do you need to rank?

There is no number that holds in general. The working benchmark is the median number of referring domains of the individual pages ranking ahead of you for that keyword, not the domain total. It changes by an order of magnitude between a niche query and a contested commercial one.

Is a brand mention without a link worth anything?

In the Ahrefs data on 75,000 brands, branded web mentions are the signal with the highest correlation (0.664) with visibility in Google's AI Overviews, above Domain Rating, referring domains and total backlinks. A mention without a link is worth more than the market gives it credit for.

Matteo Pellegrini

Matteo Pellegrini

I'm a Business Developer, and at Visilay I focus on developing data-driven SEO, Google Ads, and CRO strategies. I love historical museums, have been practicing Karate for as long as I can remember, and on weekends I enjoy exploring Italian villages in search of authentic local food.

Keep reading

Dig deeper into the topic

All articles