Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
What the 2026 Citation Studies Actually Found
Across the largest studies available, brand mentions predict AI visibility about three times more strongly than backlinks: a 0.664 correlation for web mentions against 0.218 for backlinks in Ahrefs' 75,000-brand analysis. And roughly 83% of AI citations point at third-party sources rather than brand-owned sites, per Profound's study of 680 million citations. The question is settled in direction, if not in magnitude.
The debate stays noisy because most takes cite no data at all. Four large studies now exist, and they are worth reading side by side before you renew a link-building retainer.
| Study | Sample | Key finding |
|---|---|---|
| Ahrefs (Aug 2025, extended Dec 2025) | 75,000 brands | Web mentions correlate 0.664 with AI Overview visibility, backlinks 0.218; YouTube mentions measured highest at 0.737 |
| Profound (Aug 2024 to Jun 2025) | 680M citations | Roughly 83% of AI citations point to third-party sources, about 17% to brand-owned sites; only ~11% of domains are cited by both ChatGPT and Perplexity |
| Semrush (2025) | AI responses across platforms | Reddit appeared in 40.1% of AI responses, Wikipedia in 26.3%, YouTube in 23.5%; a separate 1,000-domain study examined backlinks and AI search |
| Evertune | 200M prompts | Even the top-cited domain on a platform rarely exceeds about 5% of citations; citation share is long-tail |
One honest methods note before you quote any of these numbers. The Semrush and Profound figures use different denominators: Semrush measures the share of responses that contain a source at least once, while Profound measures share of total citations. Reddit at 40.1% of responses and no domain above roughly 5% of citations are not contradictory findings, they are simply not comparable. And several practitioner reports found review platforms moved AI visibility more than Reddit did for B2B products, so the Reddit-first advice circulating in 2026 does not generalize to every category. The studies agree on the headline, mentions over links, and disagree at the edges. That is what real data looks like.
Why Mentions Outpredict Links in AI Retrieval
AI assistants retrieve and cross-check passages; they do not compute link equity at answer time. A backlink is one vote on one page. A consistent mention pattern across reviews, communities, and listicles is corroboration the retrieval layer can actually verify while composing an answer, which is why mentions carry the stronger statistical signal.
Walk through what happens when someone asks an assistant to recommend a product. The engine fans the question out into sub-queries, pulls a pool of candidate pages, and looks for names that recur with consistent descriptions across independent sources. Nothing in that loop resolves a link graph. What it rewards is the same brand appearing in a review platform's category page, a comparison listicle, a community thread, and a publisher's roundup, each describing the business the same way. Profound's 83% third-party figure is the direct consequence: assistants prefer to cite sources that talk about you over sources you control.
This does not mean link authority stopped existing. PageRank-style signals still shape which pages get crawled, indexed, and ranked, and that happens upstream of retrieval. A page that never ranks rarely enters the candidate pool, so links act on AI visibility indirectly, the way a road network acts on where trucks can deliver. The Evertune finding matters here too: with no domain exceeding about 5% of citations on a platform, there is no monopoly position to buy. Visibility is spread across a long tail of sources, which favors breadth of mentions over depth of authority on any single page.
Where Backlinks Still Do Real Work
Three places. Classic organic rankings, which Google's AI Overviews largely recycle, so ranking still gates AIO presence. Crawl discovery and indexing, where a few real links remain the fastest way a new domain earns trust. And referral traffic from pages people actually read. The honest conclusion is reallocation, not abandonment.
The recycling point deserves emphasis because it is where the "links are dead" framing falls apart. AI Overviews draw heavily on pages that already rank in the classic top results, and links remain a meaningful input to those rankings. A page that cannot crack the first page of Google has a much thinner path into the AI answer built on top of it. Ahrefs' 0.218 correlation is weaker than the mentions signal, but it is not zero, and upstream effects are exactly the kind a correlation on the answer layer understates.
Indexing is the second job. A domain with no inbound links is a domain crawlers have little reason to visit, and no retrieval layer can cite a page that never entered an index. The third job is the one link vendors stopped mentioning: a link on a page real buyers read sends real visitors, independent of any algorithm. That was always the defensible case for digital PR, and it survives the AI transition intact. What does not survive is the spreadsheet-driven purchase of links nobody will ever click, on pages nobody will ever read, for authority no assistant will ever compute.
What to Spend the Link-Building Budget on Instead
Redirect the budget toward the surfaces the citation studies say assistants actually quote: review platforms, the listicles already being cited in your category, genuine community participation, and consistent entity descriptions everywhere your business appears.
- Review-platform presence and recency. Fresh reviews on the platforms your category is judged on, since several practitioner reports found review platforms moved B2B AI visibility more than Reddit did.
- Placement in cited listicles. Ask the assistants who they recommend, note which roundups they cite, and pitch inclusion in those specific pages rather than buying placements on pages no engine retrieves.
- Genuine community participation. Reddit and niche forums appear in a large share of responses, but only honest, disclosed participation survives moderation and reads as consensus rather than astroturf.
- Consistent entity descriptions. One name, one one-sentence description, everywhere, so cross-checking engines resolve your brand with confidence. Keeping your own answer pages current helps too, since retrieval favors fresh sources, the same reason refreshing old posts outperforms only writing new ones.
The step-by-step version of this, from entity cleanup to extraction structure, is in our guide to getting recommended by ChatGPT, and several of the platforms in our AI SEO software comparison now track mention visibility alongside rankings. One warning before you reallocate: do not swing the freed budget back into cheap links. Paid guest posts and parasite placements are named spam violations with algorithmic enforcement since August 2025, which makes buying links both less effective and more dangerous than it has ever been.
Questions People Ask About Links and AI Visibility
- Are brand mentions more valuable than backlinks for AI search?
The largest study available says yes for visibility inside AI answers: Ahrefs measured 75,000 brands and found web mentions correlate 0.664 with AI Overview visibility against 0.218 for backlinks, roughly a three-to-one gap. Correlation is not causation, but every major 2025-2026 citation study points the same direction. Links still matter for the classic rankings that feed AI Overviews, so the honest answer is both, weighted toward mentions.
- Do backlinks matter for ChatGPT recommendations?
Indirectly. ChatGPT does not compute link equity when it composes an answer; it retrieves pages and cross-checks which names appear consistently across them. Links still influence which pages get crawled, indexed, and ranked in the indexes ChatGPT retrieves from, so they shape the candidate pool without deciding the citation. Being described consistently on the third-party pages that pool contains is the stronger lever.
- Should small businesses stop building links in 2026?
No, but most should stop paying for them. Paid guest posts and parasite placements are named spam violations with algorithmic enforcement since August 2025, and the citation data shows mentions predict AI visibility far better anyway. Keep earning natural links through content worth referencing, and move the paid budget into review platforms, community presence, and the listicles AI engines already cite.
- Do AI search engines use PageRank?
Not at answer time. Assistants retrieve candidate passages from an index and select what to quote based on relevance, consistency, and extractability. PageRank-style link authority still operates upstream, deciding which pages enter those indexes and rank well enough to be retrieved at all. So link authority gates the pool while mention consensus decides the citation.
