Do backlinks still matter?
Yes, but less than they did. Google says links are no longer a top-three factor, and Ahrefs' 2025 study still finds a 0.255 correlation between referring domains and rankings. Both statements are true at once.
In short
- Ahrefs' study of one million keywords, published in January 2025, found a correlation of 0.255 between referring domains and rankings.
- The same study found the correlation for comparable queries fell from 0.27 in 2019 to between 0.22 and 0.24 in 2024.
- Google's Gary Illyes said in September 2023 that links are not a top-three ranking factor and in April 2024 that Google needs "very few links" to rank pages.
- Court exhibits from the US antitrust case show links still feed two core parts of Google's ranking, which are topical relevance through anchors and site quality through PageRank.
- For AI assistants, brand mentions correlated with visibility at 0.664 and Domain Rating at 0.326 in Ahrefs' 2025 study of 75,000 brands.
Backlinks still matter, and they matter less than they used to. That sounds like a dodge, but it is what the evidence shows. Google’s public statements have played links down since 2023. The data, and Google’s own internal documents shown in court, say links remain built into how pages are ranked. Both sides are set out below with the numbers, so you can judge how much of your budget links deserve. The question is worth taking seriously because links are the most expensive part of SEO to acquire, and someone new to the subject will hear confident voices saying that links are everything and equally confident voices saying they are finished.
The case that links matter less
Google’s own people have said so, more than once. In September 2023 Gary Illyes of Google, speaking at Pubcon, said: “I think they are important, but I think people overestimate the importance of links. I don’t agree it’s in the top three. It hasn’t been for some time.” In March 2024 Google edited its documentation to describe links as “a factor” where it had said “an important factor”, as reported by Search Engine Land. In April 2024 Illyes said Google needs “very few links” to rank pages. Each of these on its own could be put down to a spokesperson managing expectations, but together they form a consistent line held across several months, and the documentation edit is a written change that somebody at Google chose to make.
The data points the same way. Ahrefs found that the correlation between referring domains and rankings for comparable queries fell from 0.27 in 2019 to between 0.22 and 0.24 in 2024. A referring domain is a separate website that links to a page, and a correlation measures how closely two things move together, so a falling figure means the number of linking sites has become a slightly weaker guide to where a page ranks. The fall is small. It is still a fall, and it runs in the direction Google’s statements suggest.
There is also a reason for the decline. Testimony in the US antitrust case confirmed that Google uses click data heavily. A system called Navboost works on 13 months of clicks, and Google’s Pandu Nayak called it “one of the important signals”. Machine learning systems share the load too. Links are now one input among several strong ones. This makes sense when you consider what a link tells a search engine. It is an indirect sign that people value a page, passed on by whoever chose to link. What searchers click is a more direct sign of the same thing, so once a system can draw on clicks at that scale, the indirect signal has less of the work to do.
The case that links still matter
The first point on the other side is that the correlation is still there. Ahrefs studied one million US keywords and published the results in January 2025.
| Measure | Correlation with rankings |
|---|---|
| Referring domains | 0.255 |
| Followed referring domains | 0.250 |
| Backlinks | 0.248 |
| Domain Rating | 0.131 |
These are modest numbers, and they have fallen only slightly. A correlation is not proof of cause, because pages that rank also attract links. But no other off-page measure in the study did better. The order of the rows carries its own lesson. The count of separate linking sites comes out ahead of the raw count of links, and both come out well ahead of Domain Rating, which is Ahrefs’ own score for the strength of a whole site’s link profile.
The second point is that links are built into Google’s ranking. An exhibit from the antitrust case, reported in May 2025, describes the structure. Topical relevance is built from three signals: anchors, body text and clicks. Anchors are the clickable words of links. Site quality, which the exhibit calls “incredibly important”, takes PageRank as an input, and PageRank is described as “a single signal relating to distance from a known good source”. Links therefore act in both halves of the system, once in deciding what a page is about and once in deciding whether the site can be trusted. This is stronger evidence than any correlation, because it describes how the system is built and does not depend on inferring cause from patterns in the results.
The third point is that Google keeps fighting link spam. Its spam policies were last updated in August 2026 and still open with a detailed list of link schemes, meaning arrangements made to manipulate rankings through links. In August 2026 Google said link spam is “handled by separate systems” from its general spam updates. A company does not maintain a separate system for a signal that no longer counts. The effort spent on defending the signal is indirect evidence of what the signal is worth.
The fourth point is that the market is still paying. SEOs put the acceptable price of a quality link at about $509 on average, according to a March 2026 survey of 518 SEOs by Editorial.link. BuzzStream’s 2026 State of Digital PR report found that the share of digital PR teams spending more than $20,000 a month doubled to 8.8%. In a Zyppy survey of 131 SEOs, reported second-hand and so less certain, 54.8% placed backlinks among the top three factors. Spending is weaker evidence than the court exhibit, since a whole industry can be wrong about what works. Even so, these are people who watch their own rankings closely, and their budgets are moving up.
Where links matter most and least
The Ahrefs study broke the correlation down by type of query. It was stronger for local queries, at 0.33, and for high-volume, branded and informational queries. It was weaker for commercial and transactional queries. Local queries are searches for something nearby, informational queries are searches to learn something, and commercial and transactional queries are searches made by someone comparing products or ready to buy. So the single headline figure hides a spread, and where your own pages fall in that spread affects how much links can be expected to do.
That last finding surprises people, because commercial pages are where most link budgets go. One likely explanation is that few sites link to product and category pages naturally, so other signals carry more of the weight on those results. Consider a shop that sells kitchen knives. Other sites may happily link to its guide on sharpening, and almost none will link to the page listing its knives for sale, and the same is true of every competitor. When nobody in a set of results has many links to the competing page, links cannot do much to separate them. It does not mean links are useless for commercial pages. It means you should not expect a simple line between link counts and positions there.
How both sides can be right
The court documents settle the argument. Links are still wired into Google through anchors and PageRank. Click data and machine learning now do part of the job links once did alone. “Links matter less” and “links still matter” are both accurate, because the first describes a change in share and the second describes a role that has not been removed.
Illyes’ “very few links” comment fits this too. A page needs enough links to be found and trusted. Beyond that point, in many markets, more links add little. In the hardest markets, where every competitor has good content and a fast site, that point sits much higher. This is the most useful way for a beginner to hold the idea. Links work like a threshold that varies by market. A page in a quiet niche may clear it with a handful, while a page competing against established national brands may need a great many before it is even in contention, and in both cases piling on more links after the threshold is passed brings less and less.
What has changed most is which links count. Google says its systems neutralise links they consider manipulative, so the cheap, off-topic link bought in bulk is the one losing value fastest. To neutralise a link is to ignore it, so the buyer gets nothing and is rarely told. An editorial link from a relevant publication, one that an editor chose to give, has not lost value in the same way. Much of the argument over whether links still work comes from people talking about different kinds of link without saying so.
Links and AI assistants
AI assistants add a second reason to earn coverage. In Ahrefs’ study of 75,000 brands, published in 2025, the number of times a brand was mentioned across the web correlated with its visibility in Google’s AI Overviews at 0.664. Domain Rating came in at 0.326 and the raw number of backlinks at 0.218. AI Overviews are the written answers Google places above some of its results, and the figures say that how widely a brand is talked about tracked its appearance in those answers far more closely than its link counts did.
Read that with care. It compares correlations and does not show cause. Large brands have more of everything, so mentions, links and visibility in AI answers may all be reflections of size. And links do not drop out of the picture: links whose anchor text is the brand name correlated at 0.527, and a separate study by Seer Interactive found that Google rankings, which depend partly on links, were a strong predictor of mentions in AI answers. The practical reading is that a link inside real coverage now pays twice, once for rankings and once as a mention. A link placed on a page nobody reads can at best pay once. Brand mentions and citations covers that side.
What this means for your budget
The first thing to do is check the page before the links. If your page does not match what searchers want, links will not fix it, and this is the cheapest mistake to rule out because it costs nothing to compare your page with the ones that rank. The second is to measure the gap. Count referring domains to the pages that rank for your target term and compare them with yours. That tells you whether links are your problem. If the pages above you have about the same number as you do, the answer lies somewhere else, and spending on links would be spending on the wrong thing.
If links do turn out to be the gap, buy fewer, better. One link from a relevant, read publication beats many from sites that exist to sell links, because the second kind is the kind Google’s systems are built to neutralise. And count mentions as a result. Coverage with no link, or with a nofollow link, still has value, both for the readers it reaches and for the AI assistants discussed above. See the tactics chart for what each way of earning links costs and risks. If you are starting from zero, your first 100 links puts the tactics in order.
Common questions
Are backlinks still a ranking factor in 2026?
Yes. Google's own ranking guidance still lists links as a factor, and exhibits from the US antitrust case show anchors and PageRank inside its core ranking signals. Their weight has fallen compared with a decade ago.
Is link building dead?
No. Budgets are rising, and SEOs put the acceptable price of a quality link at about $509 on average, according to a March 2026 survey of 518 SEOs by Editorial.link. What is losing value is the cheap, irrelevant link bought in bulk.
Can a page rank without backlinks?
Sometimes. Google's Gary Illyes said in 2024 that Google needs very few links to rank pages. In competitive commercial markets, the pages on page one usually have more referring domains than the pages below them.
Do backlinks matter for ChatGPT and AI Overviews?
Indirectly. Ahrefs found that brand mentions track AI visibility much more closely than link counts. Links whose anchor is the brand name were still among the strongest factors, and AI assistants draw on pages that already rank.

