Key takeaways
- Backlinks still matter for AI search, but indirectly. They build the rankings and domain trust that get your pages into the pool AI engines cite from. They no longer decide visibility on their own.
- Brand mentions correlate with AI visibility roughly 3x more strongly than backlinks do (0.664 vs 0.218 in Ahrefs' study of 75,000 brands). In Ahrefs' December 2025 follow-up, YouTube mentions became the single strongest signal measured, at 0.737.
- The widely quoted "76% of AI Overview citations come from top-10 pages" figure is out of date. Ahrefs' March 2026 re-run put it at 37.1%, after AI Overviews switched to Gemini 3.
- Links behave like a threshold, not a dial. Below a certain authority level you are essentially invisible to AI engines. Above it, more links buy very little extra visibility.
- 84% of AI citations trace back to earned media, and visitors arriving from AI search convert at 4.4x the rate of traditional organic visitors. Budgets are shifting from link acquisition toward digital PR and entity presence accordingly.
The short answer
Every SEO conference in 2026 has at least one panel titled some version of "Is link building dead?" The panelists always land on "no, but it's changing," which is the safest answer in the industry and, for once, also the correct one.
For most of SEO's history the deal was simple: earn links, rank, collect traffic. AI search broke the last step of that chain. When ChatGPT or a Google AI Overview answers a question in place, the user often never sees the ten blue links your links helped you climb into. What they see instead is a short list of cited sources, and citation selection runs on its own logic. Links feed that logic, but from several steps back.
This guide walks through what the 2025 and 2026 citation datasets actually show, including one widely quoted statistic that has since been overturned, and what the findings mean for where your link building budget should go.
What the correlation data actually shows
The dataset everyone quotes is Ahrefs' study of 75,000 brands, which measured how strongly around two dozen signals correlate with appearing in AI answers. Here is the hierarchy, condensed:
| Signal | Correlation with AI visibility |
|---|---|
| Brand mentions on the web | 0.664 |
| Branded anchor text | 0.527 |
| Brand search volume | 0.392 |
| Domain Rating | 0.326 |
| Referring domains | 0.295 |
| Branded traffic | 0.274 |
| Backlinks | 0.218 |
| Paid ad traffic | 0.216 |
| URL Rating | 0.180 |
Three things jump out. Every brand-shaped signal beats every link-shaped signal. Backlinks sit below both Domain Rating and referring domains, which tells you domain-level trust matters more than raw link counts. And backlinks are statistically tied with paid ad traffic, which is quietly humbling: how much you spend on ads correlates with AI visibility about as strongly as how many links you have.
Ahrefs ran a follow-up in December 2025 across ChatGPT, Google AI Mode, and AI Overviews. YouTube mentions came out as the strongest single signal at 0.737, ahead of branded web mentions (0.709 in AI Mode). The same brands also tend to win everywhere: the pairwise overlap in which brands appear across the three engines was 0.779.
The distribution is brutally top-heavy. Brands in the top quartile by mention volume average 169 AI Overview mentions. The next tier down averages 14. The bottom half of brands barely registers at all. An audit of 500 SaaS sites found the same pattern in practice: the most-cited brands were rarely the ones with the biggest backlink counts.
One honest caveat, which the Ahrefs researchers raise themselves: this is correlation, not causation. Strong brands earn mentions, links, press coverage, and citations as a bundle, and no observational study can fully untangle which piece does the work. Fair. But if links were the dominant driver of AI visibility, they would not sit near the bottom of the table.
The 76% stat everyone quotes is out of date
For most of 2025 the consensus was "AI cites what already ranks." It rested on an Ahrefs finding from July 2025 that around 76% of AI Overview citations came from pages in the organic top 10. That number made the rounds in every GEO presentation, and plenty of articles still repeat it today.
Then Ahrefs re-ran the analysis in March 2026, across 863,000 keyword SERPs and 4 million AI Overview URLs. The new figure: 37.1% of citations rank in the organic top 10. Roughly six in ten citations now come from outside page one.
What changed? AI Overviews moved to Gemini 3 in January 2026, and the retrieval layer got noticeably broader. BrightEdge's independent tracking points the same direction: its September 2025 sample put top-10 overlap as low as 16.7%, while overlap with pages that rank anywhere in organic results rose from 32.3% in May 2024 to 54.5%.
Two implications. First, rankings still feed citations, since a majority of cited pages do rank somewhere. Second, the citation pool has widened well beyond page one. Ahrefs found that 18% of AI Overview citations that don't rank come from YouTube, with Reddit, forums, and niche publications filling out the rest. If your site can't crack page one for a head term, you are no longer automatically excluded from the AI answer. That is a genuine change from a year ago, and it is good news for smaller brands.
How backlinks still feed citations: the threshold effect
So links still matter. The question is how, and the clearest answer comes from a Semrush study of roughly 1,000 domains run with Kevin Indig.
Their finding: domain Authority Score correlates strongly with whether a brand gets mentioned by AI engines at all (Pearson 0.65, Spearman 0.57). But Authority Score correlates weakly with how much share of voice a brand gets once mentioned (Pearson around 0.23). Links function as an entry ticket. Past the gate, more authority buys very little extra presence.
Indig's summary of the data: "your backlink needs to hit a minimum threshold… You need a minimum investment to see the expected impact."
The practical translation is a budget rule. If your domain is weak, links are still priority one, because without baseline authority you are not in the pool at all. If you are already mid-to-high authority, incremental link building has sharply diminishing returns for AI visibility, and the money is better spent elsewhere. Practitioner sentiment matches the data here: 73.2% of SEOs in a 518-expert survey believe backlinks influence whether their content appears in AI search, which is a far cry from "links are dead," but nobody in that survey thinks more links equal more citations either.
One more nuance from the same study: ChatGPT Search and Perplexity weigh link quality least, while plain ChatGPT shows the strongest link-quality correlation. Engines differ in how much they lean on traditional authority, so check your own citation data per engine rather than assuming they behave alike.
The nofollow clue
This is my favorite finding in the entire body of research, because it reveals what AI systems actually read.
In the Semrush data, nofollow links perform essentially identically to followed links for AI visibility: 0.340 vs 0.341. Image links even outperform text links at higher authority tiers.
Googlebot treats a nofollow attribute as "do not pass equity." AI engines do not appear to care. They read the reference itself: your brand named on a credible page, in proximity to a relevant topic. The href is close to decoration.
That reframes the job. What you are really building is references, and a followed link is a bonus. A Reddit thread comparing your product to a competitor, a YouTube review, a podcast guest spot, a journalist quoting your founder. None of those need followed links to register with the systems deciding what AI says about your category.
Where citations actually come from
If mentions matter more than links, where do the actual citations come from? The 2026 data is surprisingly consistent on this.
Muck Rack analyzed 25 million cited links across ChatGPT, Claude, and Gemini, and found that 84% trace back to earned media: press coverage, third-party articles, reviews. Not owned content, and not SEO landing pages. If your AI strategy is "publish more pages on our own domain," this stat should sting a little.
Reddit appears in about 3% of AI citations overall. That sounds small until you look at what those citations do: more than 80% of them answer decision-stage prompts, things like head-to-head brand comparisons. Reddit punches far above its weight at the exact moment people are choosing between you and a competitor.
YouTube supplies 18% of AI Overview citations that don't rank in Google, and per the December Ahrefs data, YouTube mentions are now the strongest correlated visibility signal measured. Breadth compounds too: sites present on four or more platforms are 2.8x more likely to appear in ChatGPT responses.
Freshness matters more than in traditional SEO. One analysis of nearly 17 million citations found cited content runs about 25.7% fresher than organic top-10 pages. Your 2019 pillar page can hold a ranking for years and never get quoted.
The money case for citations
The uncomfortable numbers first. Ahrefs' Q1 2026 benchmark report found that AI Overviews now cut the average CTR of top-ranking pages by 58%, up from 34.5% eight months earlier. Seer Interactive measured organic CTR at 2.36% when an AI Overview is present versus 3.82% without one. Pew Research, watching real users across roughly 69,000 searches, recorded an 8% click rate when an AI summary appears versus 15% without.
Now the numbers that make citations worth fighting for. In Seer's dataset, brands cited inside an AI Overview averaged 2.07% CTR versus 0.94% for uncited brands on the same SERPs, roughly a 120% lift in clicks per impression. And per Semrush, visitors arriving from AI search convert at 4.4x the rate of traditional organic visitors.
The pie is shrinking. Citations decide who keeps a slice, and the slice you keep converts better than the traffic it replaced. Meanwhile the channel keeps growing: AI search visits rose 42.8% year over year, from 15.6 billion in Q1 2025 to 27.4 billion in Q1 2026, while Google search visits grew 2.4%.
What to do differently in 2026
Here is the strategy shift the data supports, play by play:
| Old play | What the data supports now |
|---|---|
| Chase referring domain counts | Hit the authority threshold, then stop counting |
| Dofollow or it doesn't count | Any credible reference counts, linked or not |
| Guest posts at scale | Digital PR built on original data |
| Optimize for page one | Be citable everywhere AI looks, including YouTube and Reddit |
| Report rankings to the board | Report citation share and AI-referred conversions |
A few notes on the new plays. Digital PR is rated the most effective link-related tactic by 48.6% of SEOs in that same expert survey, and it does double duty: the earned coverage that pulls links is also what AI engines cite. Original surveys and proprietary data are the best asset class for this, because journalists link to them and AI engines quote them.
Keep entity information consistent everywhere your brand appears, because AI engines resolve brands as entities, not URLs. Treat YouTube and Reddit as first-class channels rather than social side projects. Reddit rewards genuine participation, and the citations it generates land at the decision stage of the funnel.
One thing not to bother with: paying for directory listings in the hope of AI citations. There is no evidence business directories directly increase citations, though consistent business information still matters for local visibility.
What's dead or dying
llms.txt. Ahrefs checked 137,210 domains and found 97% of published llms.txt files received zero requests from any bot. SE Ranking found no citation connection across roughly 300,000 domains. Google's John Mueller compared it to the keywords meta tag, which is about as dead as an SEO artifact gets. If a vendor sells you llms.txt creation as a GEO fix, keep your wallet closed.
Cheap paid links. 66.5% of backlinks built over the past nine years are already dead, per Ahrefs. The average purchased niche edit runs $361.44 and the average guest post link $77.80. You are buying depreciating assets to feed a signal that correlates with AI visibility at 0.218.
Link counts as a KPI. Fewer than 8% of marketers still believe backlinks are a pure numbers game. The profession has moved even if some budgets haven't: nearly 70% of marketers say links have a big impact on traditional rankings, and almost half spend $5,000 to $10,000 a month on link building. The spend is real. Links still work for Google, where the top organic result averages nearly 4x the backlinks of positions two through ten. The question is whether that spend is pointed at the target that now decides discovery.
How to measure whether it's working
What to track
You cannot manage citation share you do not measure. At minimum, track your citation share across the major engines, which of your pages get cited, offsite mentions (your brand named on third-party pages AI cites, even without a link), Reddit and YouTube citations specifically, when AI crawlers visit your site and what they read, and the actual traffic and conversions arriving from AI platforms. That last one matters most, because mentions without traffic are just vanity.
Tools worth a look
Tools like Promptwatch cover most of this in one place, including the dedicated Reddit and YouTube citation reports most competitors skip, plus crawler logs that show the crawl-to-citation path on your own site.
If you already live in the Ahrefs ecosystem, Brand Radar extends it to AI brand monitoring and prompt tracking.
For the mention side of the equation, Brand24 tracks web and social mentions across 25 million-plus sources in real time, which matters when mentions correlate 3x more strongly than links.
And since 84% of citations trace to earned media, a media intelligence platform like Meltwater helps you watch the PR coverage that ends up feeding AI answers.
| Tool | What it shows you | Best fit |
|---|---|---|
| Promptwatch | Citations, offsite mentions, Reddit/YouTube reports, AI crawler logs, AI-referred conversions | Teams that want monitoring and content fixes together |
| Ahrefs Brand Radar | AI brand mentions and prompt tracking | Teams already on Ahrefs |
| Semrush AI Visibility Toolkit | AI share of voice and mention tracking | Per-domain tracking on a budget |
| Brand24 | Real-time web and social mentions | Brands focused on the mention signal |
| Meltwater | Earned media coverage and PR intelligence | Digital PR programs |
Pricing in this category changes monthly, so verify current numbers before committing. Entry-level AI visibility tracking generally runs $99 to $400 per month, with Promptwatch offering a free tier for testing the waters.
The verdict
Backlinks in 2026 are a qualifier, not a multiplier. They build the rankings and domain trust that get you into the pool AI engines draw from, and below a certain authority threshold you are not in the pool at all. But once you're in, links stop being the lever. What gets you quoted is everything else: mentions in earned media, presence on YouTube and Reddit, fresh content with clear claims, and a brand that shows up consistently across platforms.
So keep building links, up to the point where your domain is trusted. Then spend the rest of the budget making sure AI has something worth quoting, and a way to see whether it's working. The brands averaging 169 AI mentions against their competitors' 14 are not the ones with the biggest backlink profiles. The data is clear on that.