Link building in the age of AI search
An SEO manager opens her dashboard to the same pattern she's watched for months: rankings holding steady, impressions climbing, clicks sliding week after week.
The traffic isn't going to a competitor. An AI summary sitting above the blue links is absorbing it.
Pew Research found that users click a traditional search result in only 8% of searches where an AI summary appears, compared with 15% when there's no summary, and just 1% click a link inside the summary itself.
That one shift changes what a link is worth, how you should earn it, and which pages the algorithm even bothers to read. Link building still matters. The real question is what kind of link building matters now.
Do backlinks still move the needle in AI search?
Backlinks still count, but not the way they did two years ago. Answer engines don't rank ten blue links; they assemble a paragraph and pick a handful of sources to trust. Those sources get chosen on signals that overlap with traditional SEO without being identical to it.
An Ahrefs study of 863,000 keywords found that only 38% of pages cited in Google's AI Overviews also rank in the top 10 for the same query, down from 76% earlier in the year. The pages the AI trusts and the pages Google traditionally ranks are drifting apart. A link that helps you climb from position 12 to position 6 might do very little for whether an AI answer names you at all.
What AI answer engines are actually citing
Two patterns show up over and over when researchers audit AI citations. Brand mentions across the open web, linked or not, carry a lot of weight. And earned media tends to outperform owned content when a user asks about a specific brand.
When the AI wants a verdict on your company, it usually prefers to read about you somewhere else. That reframes link building. The point isn't only to pass authority to your pages; you want your brand to exist, with context, in complete sentences, on the sources answer engines already trust.
Which kinds of links are worth chasing now
Not every placement helps in this environment. A few categories consistently pull their weight:
- Editorial coverage on real publications. A named writer, an outlet with a masthead, and a page that reads like journalism. LLMs often quote these sources when a user asks a general question about your industry.
- Expert commentary and quoted mentions. When your founder or specialist is named as the source of a claim, that mention can outperform a raw link. The sentence around your name is what trains the model.
- Community and forum presence. Threads on high-trust discussion sites regularly show up in AI answers, not as spam drops, but as substantive contributions that get upvoted and referenced.
- Industry data and original research. Publish a number no one else has, and other writers cite you. Each of those citations doubles as an AI training signal.
What's fading: cheap directory placements, low-context guest posts on generic sites, and anything that reads like it was written for a crawler rather than a person.
How to rebuild the program for 2026
Start with the pages that deserve to be cited. Answer engines pull from content that reads as clear, specific, and sourced. If your landing pages are thin or your claims are unsupported, more links won't fix the underlying problem; they'll point authority at a page the AI has no reason to quote.
From there, the work is less about volume and more about placement quality. A useful shift in vetting: stop counting links per month and start asking whether the sites you're earning coverage on are the sites your prospects, and their AI assistants, already read. For a deeper walk-through of how to weigh that trade-off before signing with an agency, listen to the the Link.Build podcast on what separates a great link building service from a mediocre one podcast on what separates a great link building service from a mediocre one.
A practical sequence looks like this:
- Audit your current mentions. Search your brand across the sources AI tools quote most. If you're absent, that's your first target list.
- Publish something worth citing. Original data, a specific framework, a clear point of view. Vague thought-leadership rarely gets pulled into answers.
- Pitch for context, not links. A paragraph that names your brand and explains what you do beats a bare backlink in a resource list.
- Measure differently. Track share of voice inside AI answers alongside referral traffic and rankings. If you never appear in the summary, the ranking is losing its meaning.