A client came to us convinced that something was broken. Page one on Google had been for months, with a solid backlink profile, nothing unusual. Then they asked ChatGPT the exact question their page answers. Their brand never came up. Nobody else’s did either — the AI just answered the question itself, pulling from a page ranking nowhere near the top ten.
That’s not a bug. It’s the new normal, and it’s forcing a real rethink of what a “good backlink” is even for.
The Old Logic Still Half-Works
Backlinks haven’t become worthless. A link from a genuinely authoritative site still consistently ranks higher in its impact than one from a directory nobody’s heard of, and that’s just as true whether an algorithm or an AI model is doing the judging. No-follow links still aren’t dead weight either — Google reads them as a hint rather than a directive, and a link profile that’s suspiciously all do-follow still looks manufactured.
So the mechanics of “what makes a backlink good” haven’t moved much. What’s changed is how much backlinks matter relative to everything else.
Rankings Stopped Predicting Citations
Eighteen months ago, ranking in Google’s top 10 gave you roughly a three-in-four chance of earning a citation when that query triggered an AI Overview. By early 2026, Ahrefs ran the same analysis across 863,000 keywords and 4 million citations — and that overlap had fallen to about 38%. A separate study from BrightEdge put it closer to 17%.
Part of it is Google’s January 2026 move to the Gemini 3 model, which pulls from noticeably more sources per answer than before. Part of it is something called query fan-out: instead of matching your page against one search term, the AI silently splits the question into several related ones and gathers citations across all of them. You can win the keyword and still lose the citation, because the AI was never really comparing you against that keyword alone.
Practically, this means chasing a single ranking is a shrinking strategy. Covering a topic thoroughly — the sub-questions, the related angles — now matters more than owning one search term.
What Actually Predicts an AI Citation
If backlinks aren’t the top signal anymore, what is? Ahrefs’ study of 75,000 brands ranked the strongest predictors of AI citation, and the order is a little humbling for anyone who’s spent years optimizing purely for links:
| Signal | Correlation with AI citations | What it takes |
|---|---|---|
| YouTube mentions | 0.737 | Video content, transcripts, product walkthroughs |
| Unlinked brand mentions | 0.664 | Digital PR, press coverage, industry listicles |
| Branded anchor text | 0.527 | Consistent brand naming across the web |
| Brand search volume | ~0.35 | People actively searching your name |
| Traditional backlinks | 0.218 | Editorial outreach, high-authority placements |
Backlinks sit at the bottom — not irrelevant, but doing less work than almost everything above it. They still function as a floor: a domain needs enough authority to be trusted at all before AI systems will cite it. Past that floor, more links barely move the needle. What does move it is showing up, by name, in places AI already trusts — and staying recognizable when it does. That’s most of what we mean when we talk about how AI engines normalize a brand’s identity across sources — the same name, spelled and framed the same way, everywhere it appears.
Say It Early, or It Won’t Get Cited
One more shift worth building around: a study of over a million ChatGPT citations found that 44% of them come from the first 30% of a page. AI systems tend to grab whatever answers the question fastest and move on — they don’t reward the slow build the way a long-form SEO guide used to.
That lines up with something Nielsen Norman Group has said about human readers for years: most people leave a page within 10–20 seconds if it doesn’t get to the point. Turns out AI models and impatient humans want the same thing: the answer up top, not buried in paragraph twelve.
Duplicate content is still worth fixing for the same reason it always was — when the same page exists at two URLs, neither a search engine nor an AI model can decide which one to trust, and you risk losing the citation to neither.
Where This Leaves a Link-Building Budget
Relevance still beats volume, maybe more than ever — a handful of links from genuinely related, high-authority sites hold up over time in a way that a pile of low-quality links never does, and the AI-era penalty for the second approach is arguably worse than the old one.
But link building alone isn’t the whole play anymore. The businesses doing well right now are splitting effort three ways: enough high-quality backlinks to clear the trust threshold, real presence in the places AI already cites (YouTube, trade press, review platforms), and content built to hand over its answer in the first few hundred words. None of this is a one-time fix — citation patterns shift monthly, which is why it’s worth tracking rather than auditing once and moving on. We keep a running list of what’s actually worth measuring for AI visibility if you want a starting point.
For the client who came to us worried their site was broken — nothing was broken. Their site was still built for a search engine that increasingly isn’t the one answering the question. That’s the adjustment most brands haven’t made yet.
If you’re trying to figure out where your own backlink strategy needs to flex for this, Intelligent Links works with Australian businesses on exactly this — link profiles built to hold up in both search worlds, not just the one we’ve all spent twenty years optimizing for.
