Link building in 2026: what works and what is a liability
Links remain a ranking input and are increasingly a citation input too. The methods that survived are the slow ones, and the cheap ones now carry real downside.
Key takeaways
- Links now serve two purposes: ranking, and signalling authority to language models choosing whom to cite.
- Original data, named expertise, free tools and genuine PR are what still work.
- Paid links and mass guest posting risk devaluation at best and a manual action at worst.
- Unlinked brand mentions carry real value now, because generative engines read text rather than only following links.
Links still matter in 2026, and they matter in a second way they did not before: language models weight third-party sources and reputable citations heavily when deciding which domains to mention. That raises the value of genuine editorial coverage and lowers the value of volume, because a link that no human would cite is unlikely to influence a model that is modelling human judgement.
What still works
- Original data. A survey, a benchmark, an analysis of your own anonymised dataset. Publishing a number nobody else has is the only reliable way to be cited by people who owe you nothing.
- Genuine expertise on the record. Named practitioners answering journalist queries, contributing to industry publications, or being quoted because they know something specific.
- Tools and calculators. A free utility that solves a real problem accumulates links for years without further effort.
- Digital PR with a real story. Expensive, slow, and the most durable, because coverage in a publication people actually read carries both authority and demand.
What is now a liability
Paid links passing PageRank, private blog networks, and mass guest posting on sites that exist only to sell placements are all against Google's link spam policies, and detection has improved. The practical risk is less often a manual penalty than a quiet devaluation, where you keep paying for links that pass nothing. Either way the money is wasted.
Directory submissions at volume, comment links and forum signatures are not dangerous so much as pointless. They consume budget that would produce more return spent on one piece of original research.
Judging a link before you pursue it
Ask whether a relevant human would plausibly click it. That single test filters most bad opportunities. Beyond it, prefer topical relevance over raw domain metrics, prefer editorially placed body links over footer and author-bio links, and prefer one link from a publication your customers read over fifty from sites they have never heard of.
Unlinked mentions now count for something
Because generative engines read text rather than only following links, a mention of your brand in a relevant article has value even without a hyperlink. This changes the outreach calculus: coverage that would previously have been considered a failure because no link was included is now partially useful. Track brand mentions alongside links, and when you request a correction, ask for accuracy of the claim first and the link second.
Building the asset before doing the outreach
Most link building fails at the first step rather than the last. Teams write an ordinary article and then spend weeks asking people to link to it, when nothing about it gives anyone a reason. Reverse the order. Decide what would genuinely be worth citing in your field, build that, and the outreach becomes an announcement rather than a request.
The cheapest version of this for most businesses is a number they already have. Anonymised, aggregated data from your own operations is often genuinely novel: average response times in your category, the real distribution of project costs, what proportion of enquiries convert by channel. You are the only source for it, which is precisely why someone would cite it.
Measuring link work without deceiving yourself
Count referring domains rather than links, because fifty links from one site is one relationship. Track the proportion that are topically relevant, since an irrelevant link from a strong domain is worth less than a relevant one from a modest publication. And record brand mentions separately from links, because coverage without a hyperlink now carries real weight in how generative engines describe you.
If no relevant human would click it, no ranking system worth optimising for is going to count it.
Sources