TL;DR
- Backlinks still matter, but mostly as a threshold condition: enough referring domains to be indexed, trusted, and surfaced as a candidate. Beyond that threshold, more links yield diminishing returns for AI citations.
- Unlinked brand mentions and co-citations (your brand named alongside competitors in third-party content) are now stronger predictors of being cited by LLMs than raw backlink counts.
- ChatGPT and Perplexity cite differently than Google AI Mode: Google leans on its traditional link graph; ChatGPT leans on Bing's index plus its training data, where mentions in forums, PR, and listicles outweigh DR.
- Treat link building as distribution for mentions, not as the goal itself. A guest post that names you in a comparison list is worth more than a homepage link with no context.
- Audit both: referring domains (SEO tools) and unlinked mentions (Google Alerts, Brand24, CiteFlow). The gap between them is your GEO opportunity.
Backlinks were the backbone of PageRank for two decades. With generative engines summarizing answers instead of ranking ten blue links, the question is whether that signal still travels into the model's decision about whom to cite. The honest answer is "partly" — and the nuance matters if you're allocating budget between link building and digital PR.
What the recent studies actually show
The most cited dataset is the Semrush and Kevin Indig "Ghost Citations" study, which analyzed 454 prompt-and-domain combinations across multiple AI engines. The pattern they surfaced: a sizable share of cited domains were not the top-ranking SEO results for the same query. Pages with modest backlink profiles routinely surfaced in AI answers when they had strong topical alignment and were mentioned across the broader web.
The Digital Bloom's 2025 LLM visibility report, which synthesized over 680 million citations and referenced the Princeton/KDD 2024 GEO paper, reached a similar conclusion: citation-worthy content correlates more strongly with structured evidence (statistics, quotes, named entities) than with raw link metrics. Backlinks help the page get crawled and trusted, but they don't determine which sentence gets quoted inside an answer.
A March 2026 analysis from The Growth Spice framed this as a threshold effect. To appear in traditional Google top-10 results, a minimum number of referring domains is typically required. For AI citations, that minimum is lower — and once you clear it, additional links produce flatter returns than topical authority and mention density do.
Ranking links vs co-citations vs unlinked mentions
Three distinct signals get lumped together as "backlinks." They aren't interchangeable.
Ranking links are the classic SEO asset: a dofollow link from a relevant domain that passes authority. These still help Google AI Mode because AI Mode is built on top of the same retrieval stack as core Google Search. If your page can't rank, AI Mode is less likely to cite it.
Co-citations are instances where your brand and a competitor are named together in third-party content — comparison posts, listicles, "best of" roundups, Reddit threads. LLMs ingest these patterns at training time and at retrieval. When a user asks "what are the alternatives to X?", the model leans heavily on which brands repeatedly appear alongside X across the corpus, with or without links.
Unlinked mentions are brand-name references with no hyperlink — podcasts transcribed to text, news articles, forum posts. Conductor's AI search primer distinguishes citations (clickable source links rendered in the answer UI) from mentions (your brand named in the answer body without a link). Both have business value; only one shows up in referral analytics.
The Ahrefs 2025 brand-mentions study that's been circulating in GEO circles found mentions correlated more strongly with LLM citation frequency than referring domains did. That doesn't mean links are dead — it means they're necessary but not sufficient.
What this means for how you build links in 2026
A few practical shifts I'd make if I were running a GEO program right now:
- Optimize link prospecting for context, not authority. A DR-30 listicle that places you in a "top 7 tools for X" comparison is more useful than a DR-80 sidebar mention. The listicle creates a co-citation pattern the model can learn.
- Pursue PR placements where your brand gets named, even unlinked. Trade publications, podcasts, expert quotes in journalist roundups. These feed the mention graph that ChatGPT and Perplexity weight heavily.
- Keep doing technical link hygiene for Google AI Mode. Elementor's writeup referenced a Seer Interactive analysis across 10,000 finance and SaaS questions showing that AI Mode citations skew toward pages that also rank in the top 10. If you abandon traditional link building, you cede AI Mode.
- Measure both link velocity and mention velocity. Set up tracking for unlinked brand mentions and track them as a leading indicator of LLM citation lift over 60–90 days.
How to audit your current position
A 30-minute audit:
- Pull referring domains from Ahrefs/Semrush and segment by whether the linking page also mentions you in the body. Bare-anchor links are weaker GEO signals than contextual mentions.
- Run brand-name searches in ChatGPT, Perplexity, and Google AI Mode. Note which domains the engines cite. Compare to your link profile. Gaps reveal which mentions are missing.
- Search Reddit, Hacker News, and industry Slack archives for your brand. These corpora are heavily represented in LLM training data and are nearly invisible in standard backlink tools.
- Identify the top 10 comparison/listicle URLs that rank for "[competitor] alternatives" in your category. If you're not on those pages, that's a higher-leverage fix than chasing another DR-70 link.
FAQ
Are backlinks weighted the same across ChatGPT, Perplexity, and Google AI Mode?
No. Google AI Mode inherits the traditional Google ranking signals most heavily, so links matter most there. Perplexity blends web search with model reasoning and weights freshness and citation density. ChatGPT relies on Bing's index plus training-data priors, where unlinked mentions in forums and PR carry substantial weight.
Should I stop link building and focus only on digital PR?
No. Treat them as complementary. Links keep you indexable and rankable, which feeds the retrieval layer of every AI engine. PR and mentions shape which of those indexed pages get quoted. Cutting either pillar weakens the other.
How many backlinks do I need before AI engines will cite my domain?
There's no published threshold, but the pattern from the studies above suggests "enough to rank on page one for your target queries" is a reasonable floor. Below that, you likely lack the trust signals to be retrieved at all.
