Why Backlinks Don't Predict AI Citations (and What Does)

One stat is doing most of the work in every "GEO agency" blog post right now: brand mentions correlate with AI visibility at 0.664, backlinks at only 0.218. It's a real number from a real study. It's also being applied far more broadly than the study itself supports. The more interesting, more useful version of this argument is the nuanced one, not the headline one.

What the data actually shows: for Google's AI Overviews specifically, brand mentions do predict visibility considerably better than backlinks do. For ChatGPT specifically, the same relationship appears to hold, but much more weakly. And for B2B SaaS in particular, the platforms everyone assumes matter most for this (G2, Capterra, Gartner) explain almost none of the variance in who actually gets cited. Backlinks aren't irrelevant either. They still determine whether a page gets crawled and indexed at all, a precondition nothing else on this list substitutes for.

Where the 0.664 vs. 0.218 number actually comes from

The stat traces to Ahrefs, not to any of the agencies now repeating it. Ahrefs analyzed 75,000 brands (filtered to Domain Rating above 40) and ran Spearman rank correlations between several metrics and AI visibility. Branded web mentions came out at 0.664; backlinks at 0.218; URL rating lower still, at 0.18 (Ahrefs, "An Analysis of AI Overview Brand Visibility Factors"). That's a real, large-sample, transparent-methodology finding, worth taking seriously.

The detail that gets dropped when this stat gets repeated: the study's title says AI Overviews. That's Google's AI-powered SERP feature, not ChatGPT. A related Ahrefs analysis breaking correlation down by platform found something meaningfully different: branded-mention correlation was strong for Google AI Overviews (~0.65), moderate for Perplexity (~0.30), and considerably weaker for ChatGPT specifically (~0.15) (Ahrefs, "Google Seems More Biased Towards Big Brands Than ChatGPT and Perplexity"). If that platform breakdown holds up, the popularized "3x" framing is largely a Google AI Overviews phenomenon being applied to ChatGPT by extension, not a finding actually measured on ChatGPT itself. Ahrefs' own reporting on this line of research includes the caveat directly: "correlation isn't causation... that doesn't mean improving these metrics will automatically boost your AI visibility."

Why this might be true at all

Nobody has published a study from inside OpenAI or Anthropic explaining the mechanism. We checked Anthropic's own engineering writeup on retrieval directly, and it describes embeddings, lexical matching, and reranking, with nothing about mention frequency or brand consistency across the web (Anthropic, "Contextual Retrieval"). So the explanation for why mentions might beat backlinks is still a theory, not a documented fact, however settled it sounds in agency blog posts.

The most interesting version of that theory comes from Seer Interactive: the argument is that citation is largely a post-hoc process. The model first recalls which brand to recommend from what it already knows, then retrieves a source to back that recommendation up, rather than the reverse. Their supporting figure: when a brand is mentioned in an LLM's response at all, its citation rate runs around 53%; when it isn't mentioned, that drops to roughly 11% (Seer Interactive, John Lovett, Mar 24, 2026). If that's roughly right, it explains why backlinks, which shape a link graph rather than a model's training-data familiarity with your name, would carry less weight here than they do for classic search ranking.

What this looks like specifically for B2B SaaS

This is the part where the popular narrative ("get reviews, get cited") runs into the most friction.

Kevin Indig, publishing on G2's own Learn Hub, analyzed 30,000 AI citations across 500 G2 software categories and ran a regression on review count against citation share. The result: a 10% increase in review count associates with roughly a 2% increase in citation share, a coefficient of about 0.1, explaining less than 2% of the variance. His own conclusion: review count is "a small piece of a larger puzzle," not the lever it's often sold as.

Separately, a DerivateX study of 40 B2B software categories and 188 cited sources found G2 and Capterra received zero direct citations across every category tested; Gartner appeared twice, both times via user-review pages rather than analyst research (covered by AiThority, Jun 19, 2026). A small, unbranded consulting blog in that same dataset out-cited Forbes, Reuters, and Gartner combined.

Put those together with the Quoleady finding covered in our companion post on B2B SaaS citation: nearly every tool ChatGPT recommends does have G2/Capterra reviews, but review volume barely moves where you land once you're in. A consistent picture emerges. Review platforms function as a gate you need to clear to be considered, not a dial you turn to get cited more. Treating "get more G2 reviews" as a direct AI-citation strategy is, on the current evidence, an overstatement of what reviews actually do here.

The case for backlinks that doesn't disappear

None of this makes backlinks pointless, and it's fair to name where they still matter.

A page has to be crawled and indexed before any of this is possible. None of the correlation studies above account for pages that were never discovered in the first place, which is still a real, unglamorous precondition backlinks help with.

Google's AI Overviews also still draw substantially from standard organic rankings, even as that reliance appears to be decreasing over time. For that specific surface, the traditional SEO signals that backlinks feed into haven't stopped mattering. They've just stopped being the whole story.

What this actually means to do

The most useful version of this argument, more useful than either "backlinks are dead" or "just get more reviews," looks like this:

• Brand mentions genuinely seem to matter more than backlinks for AI citation, but the size of that advantage depends heavily on which AI system you're talking about. Treat the Google-AI-Overviews-specific number as exactly that, not a universal constant.
• The mechanism is plausibly that models recall and recommend brands they already know from training, then find a citation to support it, which argues for being talked about consistently across the web, not just linked to.
• For B2B SaaS specifically, G2/Capterra/Gartner presence is closer to a minimum bar than a growth lever: worth having, not worth over-indexing on.
• Backlinks still gate whether you're discoverable at all, and still matter more than the headline stat implies for Google's AI Overviews specifically.

This is also, plainly, the argument for measuring your own Share of Answer directly rather than assuming any one lever (mentions, reviews, or backlinks) is doing the work for your specific category. The studies above disagree with each other enough that the only way to know which lever matters for a given brand is to actually track the citations, not infer from someone else's 75,000-brand average.

---

Sources cited: Ahrefs (two separate studies), Anthropic engineering blog, Seer Interactive, Kevin Indig/G2 Learn Hub, DerivateX/AiThority.