Mentions vs links: the new currency of authority for AI

The currency of authority changed. For generative AI it's no longer just about how many sites link to you, but in how many textual contexts your brand shows up — with or without a hyperlink — in sources the models read or learn patterns from.

This post is the meta pair of the last one. Citation-worthy content covered the artifact — which properties of the content on your site make it extractable. Here we cover the other half: how you earn textual presence off your site. Production versus acquisition.

Traditional Link Building: What Holds and What No Longer Suffices

The backlink isn't dead. It's still a signal for classic Google search, and PageRank still runs under the hood. What changed is that the generative paradigm added a new layer of rules on top.

The dissociation is already measurable. According to Ahrefs, which analyzed 75,000 brands, brand web mentions correlate with AI Overview presence at 0.664, while backlinks do so at just 0.218 — a gap of nearly 3 to 1 in favor of the mention.

That doesn't mean abandoning links. It means reordering the priority. According to Search Engine Land, LLMs "look beyond backlinks" and evaluate mentions, context, and the repeated co-occurrence between a brand and the topics it wants to be associated with.

The link is still a good vehicle — but now it's one of several, not the only one. What a generative engine weighs is something broader: your textual footprint across the whole web.

What a "Mention" Is Through an LLM's Lens

There are two kinds of mention, and the difference matters less than you'd think. A linked mention is the classic one — anchor text plus a URL pointing to your site. An unlinked mention is just your brand name or domain written in context, with no hyperlink.

A language model doesn't draw a strict line between the two. What it registers is co-occurrence: your brand appearing near the concepts of your category, in an authoritative source. The link is a navigation bonus, not the requirement.

This isn't new — it just accelerated. In 2014, Bill Slawski documented the concept of "implied links" from a Google patent, where a reference to a brand could be treated as a signal even without being a link. His caveat matters: the patent was about branded queries, not mentions on a page. The idea that text-without-link is a signal had been brewing long before LLMs.

What changed with generative models is the mechanics. According to Search Engine Land, AI engines treat unlinked mentions "much more seriously" than traditional search — because the model learns from words, not from link graphs.

Why the Mention Is the Currency of Authority for LLMs

There are three concrete mechanisms by which a mention — linked or not — becomes authority to a model. These aren't loose hypotheses; each has evidence behind it.

Training Corpus Exposure

The model learns co-occurrence patterns between your brand, your category and the source. More exposures in relevant contexts raise the probability that the model mentions you when someone asks about your category.

The same Ahrefs study confirms it, quoting its content director Ryan Law: unlinked mentions "have very little impact on SEO, but a much bigger impact on GEO… LLMs derive their understanding of a brand's authority from words on the page". The text is the input, not the link.

Real-Time Retrieval (RAG)

When the model searches live — ChatGPT with browsing, Perplexity, Claude with search — editorial presence on crawlable tier-1 and tier-2 sites sends authority signals the model weighs when choosing what to cite. Here the prerequisite is infrastructural: a bot must be able to read the source, which connects to the evidence that favors other tactics over whatever hype is in season.

Entity Disambiguation

Wikipedia, Wikidata, the Knowledge Graph and Organization schema with a complete sameAs anchor your entity as a unique node. Consistent mentions in external sources reinforce that disambiguation — the model knows "Madbotz" is this company and not something else. It's the same authority for answer engines that makes an LLM pick your mention over others just as citable.

Mention Engineering Tactics

Here's the actionable part. Each tactic with what it is, why the LLM cares, how it's executed and how it's measured with a reasonable proxy. These are acquisition work, not writing.

Modern Digital PR Aimed at Citable Placements

Pitch newsworthy stories to tier-1 trade press — Search Engine Land, MarTech, Modern Retail, Information Week — with a first-party datapoint and a defensible thesis. The LLM cares because a placement in an authoritative source is a high-quality exposure in its corpus. You execute it with original research (ideally N>500) that gives the journalist something to cite. You measure it by tier-1 placements earned per quarter.

Expert Quote Distribution

Pitch an internal expert's takes via HARO/Connectively, Featured.com or Qwoted. The LLM cares because one clear perspective can land in multiple tier-2 and tier-3 outlets, multiplying co-occurrence. You execute it with a designated spokesperson and a sharp opinion angle. You measure it by published mentions per pitch sent.

Thematic Podcast Appearances

Show up on podcasts in the client's niche, not just generic SEO or marketing ones. The LLM cares because transcripts get indexed and guest names stay in the show notes — crawlable text. You execute it by prioritizing podcasts in the real vertical. You measure it by appearances with a public transcript per quarter.

First-Party Data as Industry Research

Publish one or two datasets or studies a year with transparent methodology, a citable landing page and a release with embed code for charts. The LLM cares because original data is the most citable thing there is — Backlinko places original statistics and research among what most attracts AI mentions. You execute it by making the datapoint easy to extract. You measure it by the citations the research accumulates.

Editorial Presence in Industry Publications

Recurring columns or bylined contributions in tier-1 and tier-2 publications in the client's vertical. The LLM cares because a recurring byline builds the brand-expertise association over time. You execute it by securing a stable editorial slot. You measure it by pieces published and how long they stay indexed.

Honest Participation in Forums Where AI Documents Itself

Substantive comments on Reddit, Stack Overflow, Hacker News and Quora, signed with brand affiliation and clear disclosure where it applies. The LLM cares because these platforms feed the post-training corpus of several models. You execute it by adding real value, never promotional copy-paste. You measure it by participations with genuine traction, not by volume.

Wikipedia and Wikidata Management

When it applies, make sure the entity's knowledge panel exists, is current, and has complete references and sameAs. The LLM cares because these sources are high-trust disambiguation anchors. You execute it with documented edits and respect for conflict-of-interest policies. You measure it by the panel's completeness and currency.

How to Measure Progress

You don't need to get into the full mechanics of tracking — that's Pillar 2 territory. Three reasonable proxies are enough to know whether the needle is moving.

The first is share of mention: how often your brand shows up in ChatGPT and Perplexity answers for your category's queries, measured by hand or with tools like HubSpot's AI Search Grader, Otterly.ai or Profound.

The second is unlinked-mention monitoring — Brand24, Mention or Google Alerts with quoted queries to catch the name written without a hyperlink. The third is the citation rate that multi-LLM monitors report. None is exact, but the trend across the three tells you whether you're gaining ground.

Comparison: Link Building vs Mention Engineering

The table sums up the reframe. The right-hand column is the answer — the operating model this post proposes for the generative era. Its logic is the one the data already shows: according to Ahrefs, the three factors that correlate most with AI Overview presence are all off-site.

Anti-Patterns That Kill Mention Engineering

There are ways to do this that don't just fail to help — they subtract. These six are the most frequent ones we see operating on real sites.

• Paying for mentions without disclosure — it violates FTC rules, and models discount low-quality content farms in post-training.
• Mention spam on Reddit or Quora with promotional copy-paste — moderators take it down and degrade the source's reputation.
• Issuing press releases with no newsworthy angle — they don't run in tier 1; they get syndicated to directories LLMs filter out.
• Creating fake personas or sockpuppets to mention yourself — detectable and toxic to reputation when it's discovered.
• Confusing monitoring with engineering — measuring mentions isn't building presence; the first observes, the second works.
• Aiming only at high-DR sites without asking whether that outlet even feeds the corpus of the models that matter to your category.

The first is the most expensive, and it connects to the sister post's thesis. According to Google Search Central, content should be made "primarily to help people" and not to manipulate rankings — a bought, disguised mention fails that test, and a model ends up discounting it.

Quarterly Mention Engineering Checklist

What the editor reviews each quarter to know whether presence acquisition is advancing:

1. There's an active pitch list of tier-1 and tier-2 outlets in the vertical.
2. At least one citable first-party datapoint was produced this quarter.
3. An internal expert's quote was distributed via HARO/Connectively or similar.
4. There are appearances on thematic podcasts with a public transcript.
5. The accumulated tier-1 and tier-2 editorial presence was audited.
6. The knowledge graph was reviewed — Wikipedia, Wikidata, current sameAs.
7. A share-of-mention baseline exists to compare against.

What Madbotz Can and Can't Claim

Honesty over hype. Madbotz is a young blog — we're building our own repertoire of tier-1 mentions, and we won't claim results we don't yet have.

What we can document is the strategy we apply to ourselves. These seven Pillar 1 posts are the citable dataset we're seeding — research on the Searchability framework and the 131 check items of Visibility as an industry contribution, not as decoration. It's the same contrarian pattern as the llms.txt post, now applied to self-assessment: say what's there, not what sells.

Inflating the dogfood would break the rule that holds this post up — a mention you can't back up takes authority away from you instead of giving it.

Frequently Asked Questions

Does an unlinked mention count for AI?

Yes. What weighs for an LLM is the textual co-occurrence of your brand with its category in authoritative sources, whether or not there's a hyperlink. An unlinked mention — just the name in context — feeds the pattern the model learns about who you are.

Has traditional link building stopped working?

No. It's still a signal for Google search and PageRank is alive. What changed is that it's no longer enough: for a generative engine to cite you, textual presence off your site weighs more than the backlink count.

How do I measure whether my mention engineering works?

With reasonable proxies: share of mention in ChatGPT and Perplexity answers for your category's queries, unlinked-mention monitoring, and citation rate in multi-LLM monitoring tools. None is perfect, but together they show the trend.

How do I make my content citable for AI bots?

By letting a model extract a sentence of yours without rewriting it — a short lead that answers in the first sentence, headings that state the subtopic, real FAQs, and verifiable first-party data. We cover it in depth in the post on citation-worthy content, which deals with the artifact on your site; this post covers the complement — earning textual presence off-site.

Closing

Three takeaways:

• The unit of authority shifted from backlink to textual mention in context — linked or not.
• Acquiring presence off your site (PR, first-party research, podcasts, forums) weighs more than the backlink count for an LLM to cite you.
• Measure it with proxies — share of mention, unlinked mentions, citation rate — and don't inflate what you can't back up.

If you want to know how visible your brand is to AI systems today — and which of the 131 check items in the Searchability framework you already meet — the Visibility analyzer tells you in under 60 seconds.