Citation signals

Do backlinks still matter for AI search?

By Arnav Mukherjee, founder of TofuBofu · July 6, 2026

A founder I was talking to last week had spent two years and a real budget building backlinks. Good ones, editorial placements, the kind an SEO agency is proud of. His Google rankings were solid. Then he ran an AI visibility scan and found ChatGPT recommending three competitors and not him. His first question was the honest one: "So was all that link building a waste?"

No. But it was not the lever he thought it was. Backlinks still matter for AI search, just far less than they matter for Google, and far less than something most founders under-invest in: plain brand mentions. Here is what the data actually shows, and how to reweight your effort because of it.

0.66
correlation between unlinked brand web mentions and appearing in AI Overviews, versus just 0.22 for backlinks, across a study of 75,000 brands. Ahrefs, 2025

What the 75,000-brand study found

Ahrefs looked at 75,000 brands and measured how strongly different factors correlated with a brand showing up in AI Overviews. The three strongest signals were all off-site, and all about being named rather than being linked: brand web mentions at 0.664, brand anchors at 0.527, and brand search volume at 0.392. Backlinks came in well behind at 0.218. A follow-up study extended the analysis to ChatGPT and Google AI Mode and found branded mentions still leading, correlating in the 0.66 to 0.71 range.

Correlation is not causation, and nobody outside the labs knows the exact recipe. But the direction is consistent and it is large. Mentions correlate roughly three times more strongly than backlinks. Ahrefs also found that the brands earning the most web mentions showed up in AI Overviews up to ten times more often than the next group down. That is not a rounding error. It is a different game.

How strongly each signal correlates with AI visibility

Brand web mentions 0.664 Brand anchors 0.527 Brand search volume 0.392 Backlinks 0.218 Correlation with AI Overview presence across 75,000 brands (Ahrefs). Mentions lead links roughly 3 to 1.

Why AI values a mention over a link

A backlink is a web-graph signal. Google's ranking system was built to read links as votes, so for twenty years links were the currency. A large language model does not work that way. It learns from language: articles, forum threads, transcripts, reviews, documentation. In that world the unit of trust is not a hyperlink, it is a sentence where a credible third party names you in context.

That is why an unlinked mention can outperform a link. "We moved our helpdesk to [your company] and cut ticket time by half," written in a Reddit thread with no link at all, is exactly the kind of specific, contextual, third-party statement an engine can lift and repeat. A link in a resource-page footer, with your brand name nowhere in the surrounding text, gives the model almost nothing to work with. The link helps Google. The sentence helps everyone.

So backlinks are not dead, they are demoted

This is the part people overcorrect on. Backlinks still do real work. Google AI Overviews is built on Google's own index and ranking, which links still influence, so a strong link profile can carry you into AI Overviews even though mentions correlate more strongly. And in practice a good editorial link almost always arrives wrapped in a brand mention, so quality link building produces both at once. The mistake is not building links. The mistake is treating links as the whole strategy when the data says they are now a minority of it.

The cleanest way to think about it is proportion. A B2B services firm in 2022 might reasonably have spent 70 percent of its off-site effort on links and 30 percent on mentions and PR. In 2026, given how AI answers are formed, closer to 30 percent links and 70 percent mentions fits what the engines actually reward. Same total effort, reallocated.

What to do

1. Audit where you are mentioned, not just linked

Search your company name in quotes across Google, Reddit, YouTube, and your category's review sites. Count unlinked mentions too. That inventory is your real AI-visibility footprint, and it is usually thinner than the backlink profile suggests.

2. Get named in context by credible third parties

The goal is sentences that describe what you do and for whom, on sites AI reads: a Reddit answer, a G2 or Clutch review, a guest article, a podcast transcript. Specific beats generic, because specific is what gets quoted.

3. Keep building quality links, stop over-indexing on them

Do not fire your link building. Just stop treating link count as the scoreboard. Prefer editorial placements that name you in the body text, because those double as mentions.

4. Never buy links or mentions

Paid link networks and fake mentions get devalued or wiped, and they give the engines nothing durable. Earned mentions are the asset. Bought ones are a liability waiting to be discounted.

See which mentions are actually getting you cited

Run a free scan to see what ChatGPT, Claude, Gemini, and Perplexity say about your space, and where the gaps are.

Get your free audit

Frequently asked questions

Do backlinks still matter for AI search?

Yes, but much less than they used to, and much less than brand mentions. In an Ahrefs study of 75,000 brands, unlinked brand web mentions correlated with appearing in AI Overviews at 0.664, while backlinks correlated at only 0.218. Backlinks still help build the Google ranking and authority that feeds AI answers indirectly, so they are not dead. They are just no longer the main lever for getting cited by AI.

What matters more than backlinks for AI visibility?

Brand mentions. AI engines learn from text, conversations, transcripts, and reviews, where what counts is that a credible third party names your company in context, with or without a hyperlink. In the Ahrefs data, the three strongest signals were all off-site and mention-based: brand web mentions (0.664), brand anchors (0.527), and brand search volume (0.392). Backlinks trailed at 0.218.

Why do AI engines value mentions over links?

Because a large language model does not read the web as a link graph the way Google's ranker does. It ingests language: articles, forum threads, reviews, and video transcripts. In that substrate, being named by a trusted source is the signal. A hyperlink is invisible to the model unless the surrounding text also mentions you. So an unlinked mention in a Reddit thread or a G2 review can matter more than a link buried in a footer.

Should I stop building backlinks?

No. Backlinks still build the domain authority and Google rankings that feed Google AI Overviews indirectly, and a good editorial link almost always comes with a brand mention anyway. The change is one of proportion. Where a services firm might once have split effort 70 percent links and 30 percent mentions, in 2026 something closer to 30 percent links and 70 percent mentions fits the data better for AI visibility.

Where should I focus instead of link building?

The four channels that produce the most brand mentions: community conversations like Reddit, Quora, and niche forums; creator and YouTube content; third-party review sites like G2, Capterra, Clutch, and Trustpilot; and earned editorial PR. These are the places AI engines read most and where being named in context does the most work.

Do backlinks help Google AI Overviews?

Indirectly, yes. Google AI Overviews draws on Google's own index and ranking, which backlinks still influence. So a strong link profile can help you appear in AI Overviews even though the mention signal correlates more strongly. For ChatGPT, Claude, and Perplexity, which rely less on Google's ranking, the mention signal matters even more relative to links.

How do I know where my brand is mentioned?

Search your company name in quotes across Google, Reddit, YouTube, and the review sites for your category, and note both linked and unlinked mentions. An AI-visibility tool goes further by showing what ChatGPT, Claude, Gemini, and Perplexity actually say when a buyer asks about your space, so you can see which mentions are translating into citations and which gaps to close.

Sources and further reading

Keep reading: Why Reddit, G2, and third-party mentions move AI answers · How customer reviews feed AI recommendations · If I'm good at SEO, am I good at AEO?