Do Backlinks Still Matter for AI Recommendations?

Backlinks still matter for AI recommendations, but not in the way they mattered for Google rankings. A link is no longer a vote that adds to a score. It's now one of several ways a model learns that other people take you seriously — and, just as importantly, a path its sources traveled to find and describe you. The links that help are the ones that put accurate, corroborating information about you in places a model is likely to read. The ones that did nothing but inflate a metric are worth even less than they used to be.

I want to be upfront that this is my practitioner read, not a statement about any model's internal mechanics. Nobody outside the labs knows exactly how a given assistant weighs a link, and the honest answer to most "does X still matter" questions is "it depends, and it's changing." But the pattern I keep seeing is consistent enough to act on, and it points in a clear direction. So let me walk through what a backlink used to do, what it does now, and where I'd actually spend the effort.

The old job of a backlink

For twenty years, a backlink was essentially a vote. Google's whole early advantage was treating a link from one page to another as a signal of trust, and the more trusted pages that pointed at you, the higher you climbed. That created an entire industry around acquiring links — some of it legitimate, a lot of it not. The underlying logic was quantitative: links were countable, their value flowed through a graph, and your ranking was, in part, a function of how much of that value you'd accumulated.

That system had a known weakness. Because links were countable and valuable, they were gameable. Private blog networks, paid links, reciprocal schemes, directory stuffing — a huge amount of effort went into manufacturing the appearance of trust rather than the thing itself. Search engines spent two decades fighting that, and anyone doing serious SEO learned to distinguish a link that reflected real endorsement from one that was just a number on a dashboard.

What AI actually does with a link

AI answers don't work by tallying a score and ranking pages. When an assistant answers a question about who to use or trust, it's drawing on two things: what it absorbed during training, and — for the systems that retrieve live information — what it can pull from the web in the moment. In both of those, a link plays a different role than it did in classic search.

The first role is corroboration. A model trying to decide whether to name you is, in effect, asking how many independent, credible sources describe you the same way. A link from a respected publication isn't valuable because it passes authority through a graph — it's valuable because it is a credible source saying something about you, in a place the model likely read. The link and the mention travel together. What the model takes away isn't "this site got a vote," it's "an independent source that I trust describes this business as X."

The second role is retrieval and discovery. For assistants that browse or that were trained on crawled data, links are still the paths that lead to information. A mention of you on a well-connected, frequently-cited page is more likely to be found, read, and remembered than the same words sitting on a page nothing points to. In that sense links haven't lost their function as the web's connective tissue — they still determine, in part, what gets discovered. They've just stopped being a currency you bank and started being a map of where your information lives.

The third role is entity confirmation. When several reputable sources link to and describe the same entity consistently — same name, same category, same facts — they help a model resolve who you actually are and reduce the odds it confuses you with someone else. That's less about authority and more about identity, and it's one of the quietest but most durable things good links do for you now.

Which links move the needle now

Because the job changed, the value of different kinds of links changed with it. Here's how I sort them.

• Editorial and earned mentions still matter — more, if anything. A journalist, analyst, or respected operator writing about you, with a link, is exactly the kind of independent corroboration a model is looking for. The link is almost incidental; the context around it is the asset.
• Relevant, in-context links beat generic ones. A link from a page that's genuinely about your topic, surrounded by substance that describes what you do, carries meaning. A link dropped into an unrelated footer or sidebar carries almost none, because there's no useful context for a model to absorb.
• Corroborating links compound. One source saying you're the go-to for something is a claim. Several independent sources saying it the same way is a pattern a model will repeat. Links that reinforce a consistent story about you are worth more together than any one of them is alone.
• Manufactured links are close to dead weight. Private blog networks, paid link placements, low-quality directories, reciprocal swaps — these were always risky in SEO, and in the AI context they do almost nothing, because they add no credible, independent description of you. They're links without the corroboration that made links useful in the first place.

The throughline is that a link is now only as valuable as the context and credibility wrapped around it. A link with no accurate information about you attached is just a path to nowhere interesting. A link embedded in a credible source that describes you well is a small, durable deposit in the record a model reads.

The links you'll never see on a dashboard

Here's a shift that trips people up. In classic SEO, the links that mattered were the ones pointing at your domain, and you could see them in your tools. In AI recommendations, some of the most valuable corroboration lives in links you don't control and can't easily track — a Reddit thread that cites a review site, a roundup article that references an industry report that mentions you, a forum answer linking to a comparison that happens to include you.

Models read the web as a connected whole, not as a list of who links to your homepage. So a lot of your real "link equity," in the loose sense, is the web of accurate references to you sitting inside other people's content — content you didn't build and a backlink checker won't fully surface. This is why I tell people to stop obsessing over their own domain's inbound-link count and start asking a different question: when someone writes about my category, is accurate information about me within easy reach of what they'll cite? That's the modern version of "do I have good links."

What I'd actually do about it

If you came here asking "should I still do link building," my answer is: do the half of it that was always real, and drop the half that was always a game. Concretely, that means earning mentions in credible, relevant places — being quoted, being included in legitimate roundups and comparisons, publishing things worth referencing, showing up where your category gets discussed. Every one of those creates the thing that now matters: an independent, in-context, accurate description of you that a model can find and corroborate.

Stop spending on volume for its own sake. Buying a hundred low-quality links to move a metric was dubious SEO and is near-useless for AI visibility. The effort is better spent making sure the handful of credible sources that do describe you describe you accurately and consistently, and that there are enough of them to form a pattern rather than a single data point.

And measure the right thing. Instead of tracking inbound-link count as a vanity number, track whether the independent record about you is growing in quality and consistency — more credible sources, saying accurate things, in places that get read and cited. That's slower and harder to put on a dashboard, but it's what the models are actually reading.

The honest limits

A few cautions, because I'd rather you plan around reality. Links alone won't carry you — a pile of good references can't compensate for a thin or contradictory core of information about who you are and what you do; the links corroborate a story, they don't invent one. The weighting is also opaque and shifting, so anyone who tells you a link from site X is worth precisely Y for AI visibility is guessing with false precision. And none of this is instant; corroboration accrues slowly, the same way reputation always has.

But the core answer holds. Backlinks still matter, because the thing a good link represents — a credible, independent source taking you seriously, in context, where a model can read it — matters as much as it ever did. What's dying is the idea of a link as a countable, bankable score. What's very much alive is the link as a piece of the public record that tells an AI who you are and whether to recommend you. Build for the second one, and the fact that the first one is fading will work in your favor.

About the author

Scott Tischler is the Founder & Chairman of AIrecommend.ai and a practitioner-authority on AI search and Answer Engine Optimization. With 20+ years in marketing technology — including American Express, MetLife, and UBS — and executive and professional study at Wharton, Harvard, and Oxford, he helps businesses become the ones AI recommends.