Link Building in 2026: What the Evidence Says Still Works

Every link building guide worth reading was written for a world where the destination is a page of blue links. That world still exists. It is simply no longer the only one, and the guides have not caught up.

The most useful evidence on what changed comes from an unlikely source.

Does link building still work in 2026?

Link building still works for classic search rankings, and does very little for whether an AI assistant mentions your brand. Those are two different jobs, and most guides still treat them as one.

Ahrefs analysed 75,000 brands to see which factors correlate with brand mentions in ChatGPT, AI Mode and AI Overviews. One of their headline conclusions, in their words, is that link and content volume has little impact on AI visibility.

That deserves a moment. Ahrefs sells access to one of the largest backlink indexes in the world. A finding that links matter less than assumed runs against their own commercial interest, which is exactly why it is more credible than the same claim from someone with nothing to lose. Their own link building guide, all 6,264 words of it, does not mention AI search once.

So the honest answer to whether link building still works depends entirely on which surface you are asking about, and the numbers below separate the two.

What does the evidence actually show?

The strongest correlate of AI visibility in the study was YouTube mentions, at roughly 0.737, ahead of every other factor tested. The signals that matter cluster around presence and recognition rather than around links.

Two cautions before anyone acts on that table. These are correlations across 75,000 brands, not proof that one thing causes another, and large brands score well on nearly every signal at once, which inflates everything. And the 0.779 overlap figure carries a practical implication of its own: optimising separately for each assistant is largely wasted effort, because they mostly agree.

What the numbers change is where the effort goes, and the answer is less about links than most of the industry has been assuming.

What does that change about how you build links?

The shift is from accumulating links to being mentioned, and the two are not the same activity. A link is a technical artifact on a page. A mention is your brand appearing in a context a model reads, whether or not anything is clickable.

Three practical consequences follow. The first is that coverage without a link is no longer a failure. A piece in a trade publication that names you and does not link you is still doing the job for the surface where models read text rather than follow hyperlinks.

The second is that YouTube stops being a separate marketing channel and becomes part of the same job. If mentions there carry the strongest correlation in the study, then a video series is a visibility investment rather than a nice-to-have.

The third is the least comfortable: the tactics that scale link volume are the ones this evidence says matter least. Bulk guest posting, directory submissions and paid placements were always about volume, and volume is the variable with the weakest relationship to AI visibility. Which raises the obvious question of what actually deserves your time.

Which link building tactics still earn their time?

The tactics worth keeping are the ones that produce a mention a human chose to make, because those work on both surfaces at once. A link earned by being genuinely useful counts for classic search, and the mention around it counts for AI visibility. Everything else optimises for one surface at the expense of the other.

• Digital PR. The strongest version of this, because it produces coverage, mentions and links from one piece of work. We cover the campaign types and the pitching process separately in digital PR for link building.
• Original data. Publishing something nobody else has is the most reliable way to be cited, by journalists and by models. The Ahrefs study this page is built on is itself an example of the tactic working.
• Video. On the study's own numbers, the strongest correlate of AI visibility. Most SEO teams do not own this channel and treat it as somebody else's job.
• Genuinely free resources. Tools and lists people link to because they use them. Ours sit at free backlinks and across the free tools.
• Being the obvious answer. Branded search volume correlates with AI mentions across all three assistants. Work that makes people search for you by name feeds the same signal.

Notice what is missing from that list, because the omissions are as informative as the entries.

What should you stop doing?

Stop buying volume. Paid links breach Google's published spam policy, and on this evidence they buy the variable with the weakest relationship to the surface that is growing.

Stop treating a nofollow link as a wasted placement. It carries a mention, and mentions are the currency the study points at. Stop measuring the programme on link count alone, because a number that goes up while brand mentions stay flat describes activity rather than progress. And stop optimising separately for each assistant when they agree with each other 78 per cent of the time.

If you want to check what you already have rather than what to build next, the questions below cover the ones people ask most.

FAQ

Do backlinks still matter for Google rankings?

Yes. The Ahrefs study measured AI brand mentions, not classic rankings, and nothing in it suggests links stopped working for the blue-link results. The finding is narrower and more useful than "links are dead": links do one job well and another job poorly.

How many backlinks does a site need?

There is no target number, and this evidence makes chasing one worse rather than better. Content volume showed almost no relationship with AI visibility at 0.194, and link volume was reported as similarly weak. A smaller set of links attached to real mentions outperforms a larger set that is not.

Does this mean I should stop link building?

No. It means link building earns its place when it produces mentions as well as links, which is what earned coverage does and what bought placements do not. The activity changes shape rather than stopping.

Which AI assistant should I optimise for?

All three mostly name the same brands, with an output overlap correlation of 0.779 in the study. Separate strategies for ChatGPT, AI Mode and AI Overviews mainly duplicate effort.

Check your own link profile

Two free things here do the work that follows from this page. The Backlink Gap Finder shows which sites link to your competitors and not to you, which is where earned opportunities usually hide. If you are judging a specific link rather than planning a programme, good backlinks versus bad backlinks covers the test Google actually applies. The Competitor Backlink Scanner runs a deeper version of the same analysis inside Claude.