Backlinks vs. Mentions: What Really Counts for AI Visibility

Two currencies people mix up

A backlink is a clickable reference from an external site to yours — technically an HTML link with your URL as the target. Search engines counted and weighted these for two decades because a link was evidence that someone found your page useful. A backlink is measurable, verifiable, and easy to audit. That's what made it the hard currency of classic search engine optimization.

A mention is different: your brand, product, or location shows up in running text with no link attached. A trade publication writes about your software, a forum thread recommends your workshop, a directory lists you with a short description. No click target, but real context. For a human reader that's often worth more than a bare link, because the mention is explained and justified in place.

Here's the split that matters: language models don't evaluate these two things the way Google did for twenty years. They read text, not just link graphs. That's why a well-placed mention with no link can matter more for your AI visibility than a technically clean backlink sitting on a page with nothing to do with your topic.

How language models actually read text

A large language model trains on enormous volumes of text. It isn't learning which page links to which — it's learning which words and names tend to appear together, and in what context. When your company name keeps showing up next to phrases like sustainable packaging, Munich, and mid-sized, the model connects those ideas. No link required; linguistic proximity alone builds the pattern. Researchers first gave this dynamic a name — Generative Engine Optimization, or GEO — in a 2023 paper from Princeton and IIT Delhi, and the core finding still holds: what you write about yourself, and what others write about you, shapes how models describe you.

When answering a question, modern AI systems also pull in live search results — retrieval. Here too, content outweighs links: the system reads matching passages and drafts an answer from them. One reason this matters: an Ahrefs analysis found only 6–8% of URLs cited by ChatGPT overlap with Google's top 10 for the same query, and about 80% of ChatGPT's cited pages don't even rank in Google's top 100. Citation and ranking are two different games. If a source describes you clearly and accurately, you're likely to land in the answer. If you're absent from the text, or only named in passing, even a strong backlink sitting quietly in the background won't save you.

The practical takeaway: visibility in AI answers gets decided at the level of language, not link equity. It's not just whether people talk about you — it's how precisely and how consistently. Contradictory or thin descriptions blur the picture a model holds of you and dilute every mention you've earned.

Backlinks still matter — just not the way they used to

Writing off backlinks would be a mistake. They still do three things that indirectly support AI visibility. First, they bring real visitors to your site, and that traffic is itself a relevance signal. Second, links help search engines discover and classify your content in the first place, which is the raw material retrieval systems draw on. Third, a share of AI answers still leans on the classic search index, where links continue to count — even as that index gets clicked less: Pew Research found users click through to a traditional result in only 8% of visits when an AI summary appears, versus 15% without one, and Ahrefs measured a 58% drop in average click-through for the #1 organic result once AI Overviews show up.

On top of that, a backlink is often the visible proof of a mention. When a trade publication writes about your company and links to it, you get both at once: context for the language model and a technical signal for the search engine. The smartest approach doesn't separate the two — it looks for the moments where a mention and a link land together.

What you should drop is the old numbers game. A hundred cheap links from unrelated sites do almost nothing for AI visibility, while ten credible mentions in the right context sharpen how a model describes you. This tracks with an Ahrefs study of roughly 75,000 brands, which found brand mention frequency correlates with AI citation rate at about 0.664 — nearly three times stronger than the backlink correlation of about 0.218. Quality and topical fit beat raw volume, more clearly now than they ever did in classic SEO.

What separates a strong mention from a weak one

Not all mentions carry equal weight. A strong mention uses your full, consistent name, places you in a clear category, and sits inside text that's actually about your topic. A concrete example: a physiotherapy practice gets more out of a paragraph in a health guide that names its location, specialty, and hours than from a stray link buried in an unrelated blog post.

The source matters just as much. Models learn which sources to trust more. A mention in an established trade publication, a reputable directory, or a known partner's site carries more weight than identical wording on an anonymous aggregator page. Spread your presence across several credible, independent sources rather than betting everything on one.

And repetition compounds. A model anchors a connection more firmly when it sees the same facts phrased similarly across many sources. Keep your core details — name, category, region — worded the same way everywhere. It's unglamorous work, but it's one of the strongest levers you actually control.

The mistake almost everyone makes

Plenty of companies assume more links automatically means more AI visibility. They buy link packages or trade placements with unrelated sites, then wonder why they still don't show up in ChatGPT. The error: they're optimizing a signal that only counts indirectly for language models, while ignoring the text that actually trains the model and gets read when it generates an answer. The same mistake shows up in a newer form — publishing an llms.txt file and assuming it changes anything. Google's John Mueller has confirmed no Google Search system reads or acts on llms.txt, and an Ahrefs review of roughly 137,000 sites that published one found about 97% saw zero measurable referral traffic tied to it.

The opposite mistake is just as common. Some companies rely entirely on their own polished website and skip external validation. If you're the only one talking about you, that's an unverified claim as far as an AI system is concerned. It becomes credible enough to show up in an answer only once independent sources tell a consistent version of the same story.

The fix sits in the middle. You need a clean foundation of your own content plus external voices that back it up. Neither pure link-building nor pure self-description gets you there — what works is a coherent story told from multiple sources that don't contradict each other.

A plan for building both, deliberately

Start with your own substance. State clearly on your site who you are, what you offer, and who it's for. These facts become the template everyone else copies, and the raw material models pick up on. Then make sure the same facts show up, worded identically, across directories, industry portals, and partner pages. It's tedious work, but it feeds consistent mentions directly.

From there, invest in things worth writing about. A genuinely useful piece, a small original study, local involvement, a real product argument — these give journalists, bloggers, and peers a reason to name you. Earned mentions like these often bring a link and the context together, and they hold up far longer than paid placements that wobble every time an algorithm changes.

Finally, check whether any of it is working. Ask language models about your category on a regular basis and see whether, and how, you're named. These checks show you faster than any backlink count whether the model's picture of you is accurate, and where the gaps or wrong associations still need fixing.

• Your own website: state category, audience, and region clearly, and repeat them.
• Directories and portals: identical core facts everywhere, no inconsistent spellings.
• Earned mentions: create things worth writing about, so trade media and partners cover you unprompted.
• Consistency checks: hunt down and fix contradictory details across the web.
• Test the effect: ask models directly about your category and log how you're mentioned.

Where the weighting is heading

The direction is clear. The more people ask AI systems their questions instead of typing into a search box, the more value shifts from raw link counting toward understood context. Backlinks don't disappear — they go from being the main signal to one building block among several, while the surrounding text carries more of the weight.

That doesn't mean scrapping what you've already built. Most of what makes classic optimization good also helps AI visibility: clean content, credible sources, real relationships with partners and media. What's changed is the emphasis. The first question is no longer who links to me, but who's talking about me, and in what context.

Whoever takes this shift seriously now gets a head start. Building consistent mentions takes longer than running a link campaign, but the result holds up better and is harder for a competitor to copy. That's exactly why it's worth investing in substance and context today, instead of chasing a metric that explains less every year.

A concrete example

Picture two providers with nearly identical offerings. Provider A has built up 180 backlinks over two years but is rarely named in free-running text. Provider B has only 40 links but is named in roughly 25 trade articles, forum threads, and comparison lists — often with a clear description like 'affordable for small teams.' Ask a language model for a recommendation, and it will almost always name B, simply because it has read more descriptive context about B.

The logic is simple: a link with no surrounding text gives the model nothing it can retrieve later. Twenty-five clear mentions, on the other hand, build a dense picture of features, audience, and strengths. Try this yourself: count how often your name shows up in free text tied to a specific characteristic. That number says more about your AI visibility than your backlink count in an SEO tool ever will.

None of this means volume is irrelevant — it just means the volume that counts is measured in content-loaded mentions, not bare references. Run this exercise once for your own brand and it becomes obvious why classic link metrics point in the wrong direction here.

How this differs by industry

Mentions don't distribute the same way across industries. In B2B software, they show up mainly in comparison sites, trade blogs, and community threads where users describe real experience. The move here is to show up in genuine discussions and place trade articles that state your position clearly.

In local business — restaurants, trades, practices — review platforms, regional portals, and local editorial coverage dominate. A model often builds its assessment from review text and location references. Keep your name, location, service, and one recurring characteristic consistent everywhere so a single clear picture forms.

In e-commerce, product reviews, buying guides, and comparison lists carry the most weight. Rely only on backlinks from price-comparison portals and you're feeding the model numbers with no story attached. Figure out first where your industry's providers actually get described in narrative form, and put your energy there.

FAQ: what mentions can and can't do

Is being named everywhere enough? No. A mention with no substance — say, a bare listing in a company directory with no description — barely helps the model at all. Quality clearly beats raw frequency here: ten well-described mentions outweigh a hundred bare directory entries.

Can I just buy mentions? Not a good idea. Paid, uniform-sounding text stands out for its repetition and builds no credible picture. What lasts is a real occasion — a trade article, a study, a customer project others choose to write about. Organic mentions like that keep paying off for years.

And the limits? You control the occasion, not the exact wording someone else uses. So make your core message clear enough that it's easy for others to repeat accurately. Measure progress over months, not weeks — language models build their picture slowly, which is exactly why a solid base of mentions is worth the patience.