John Mueller has said it plainly enough that even the most stubborn link-builder should have absorbed it by now: an unlinked brand mention on someone else’s site does not pass PageRank. It does not tickle the link graph. It does not, by itself, push you up a single position in Google’s traditional organic results. Yet here we are, watching the same SEO forums recycle the same “implied link” patent from 2009 as if it were classified intelligence. The actual mechanics of how brands get surfaced in search have quietly split in two. Google’s classical algorithm still runs on hyperlinks. Everything emerging around it, from entity understanding to generative AI recommendations, increasingly runs on whether the web talks about you, and what it says when it does.
The Patent SEOs Keep Misreading
The confusion starts with US7603355B2, the patent that launched a thousand blog posts about “implied links.” Read it carefully (or read Mueller’s repeated clarifications) and the distinction is obvious. The patent describes using branded search queries, typed by actual users, as a proxy for popularity and authority. It does not describe Google’s crawlers spotting your company name in paragraph text and silently converting that into link equity. “Kadabra AI reviews” as a search trend tells Google something. “Kadabra AI” sitting unlinked in a TechCentral article does not, in itself, register on the same machinery.
This frees you from a specific kind of magical thinking. The old playbook said: get mentioned, hope Google counts it as a “soft link,” watch rankings rise. That playbook was wrong. Unlinked mentions were never a backdoor into PageRank. Treating them as failed backlinks leads to bad measurement, bad reporting, and bad strategy. They were always doing something else, and that something has become more important as Google’s understanding of the web has moved from graph theory toward language models.
How Google Actually Understands Your Brand
Google’s Natural Language Processing systems map entities to topics, attributes, and relationships. Each time your brand appears in a context Google trusts, it adds another vector to that map. Not a link, but a fact. “Kadabra AI” plus “machine learning in finance” plus a publication Google recognises as authoritative becomes a data point in how Google understands what your entity is for.
This feeds directly into E-E-A-T, the framework Google’s quality raters use to judge whether a source deserves trust: Experience, Expertise, Authoritativeness, Trustworthiness. Unlinked mentions in trusted publications do not replace backlinks for traditional ranking. They do, however, compound the picture of whether your brand is a real participant in its field or a hollow site optimised for keywords. The Knowledge Graph does not care whether your name is blue and underlined. It cares whether you show up consistently, in diverse contexts, as a coherent entity.
For South African businesses, this has a practical shape. A fintech startup mentioned in BusinessTech, ITWeb, and a handful of specialised African tech newsletters is building entity recognition that no amount of directory submissions will match. The publications do not need to link. They need to be real, and they need to describe you accurately in contexts that make sense.
The Local SEO Exception Where Mentions Are Direct Signals
Unlinked mentions function as explicit, direct ranking inputs in one corner of Google’s system: local search. NAP citations (Name, Address, Phone number) scattered across the web tell Google whether your business exists at a specific physical location and whether it can trust the details it has.
Inconsistent NAP data confuses the system. Your Google Business Profile says one phone number. Yellow Pages SA lists another. A local chamber of commerce directory has an old address. Each discrepancy erodes confidence, and confidence determines whether you appear in the local pack for “plumbers Johannesburg” or vanish beneath three competitors who got their details straight.
For South African businesses, the practical checklist is unglamorous but decisive. Audit your presence across Yellow Pages SA, industry directories (SA Decor & Design for interior work, construction portals for builders, hospitality listings for restaurants), community portals, and any local news coverage. Match every character: the same abbreviation rules, the same phone format, the same street address (not “Shop 4” on one site and “Unit 4” on another). These are unlinked brand mentions in the most literal sense, and they are among the few that Google’s classical algorithm treats as first-class signals.
Why Generative AI Cares Even Less About Your Links
The real shift is happening outside Google’s traditional results entirely. AI Overviews, Gemini, Perplexity, ChatGPT. These systems do not crawl the live web and calculate PageRank in real time. They synthesise from training data, massive corpora of web pages, books, articles, and forums ingested during development. What they learn are associations, frequencies, and contextual patterns.
When a user asks Perplexity for “the best project management software for remote teams in South Africa,” the model is not running a link popularity contest. It is retrieving what it has learned about which brands cluster with which problems. If Asana, Monday.com, and a local contender like ClickUp (or a genuinely South African alternative) appear repeatedly in training data alongside that use case, they surface. The original mentions may have been linked or unlinked. The model does not distinguish. It cannot click.
This changes the value equation for brand mentions fundamentally. A linked mention in a high-authority site still matters for traditional SEO, still drives referral traffic, and still builds the profile that feeds entity recognition. But an unlinked mention in a niche forum, a Reddit thread, a MyBroadband discussion, or a local newsletter now carries potential value in a completely separate system. This system is growing fast, younger users increasingly default to it, and Google’s own interface is absorbing it.
Google’s AI Overviews already demonstrate this. The citations they surface often include pages where brands are discussed without hyperlinks. The model found the association useful. The link status was irrelevant.
What This Looks Like in Practice
The implication for South African marketers and SEO practitioners is a split strategy, not a replacement strategy. Backlinks remain essential for classical ranking. Anyone telling you otherwise is selling something. But a strategy optimised only for backlinks is now a strategy blind to half the search ecosystem.
Start with tracking. Use Google Alerts for your brand name, product names, key executives, and common misspellings. Manually sweep Twitter, LinkedIn, Facebook, and local forums like MyBroadband. For deeper work, paid tools like BrandMentions, Prowly, or Brand24 offer sentiment analysis, competitive share-of-voice, and coverage of sources manual search misses. The goal is not vanity metrics. It is understanding whether your brand is being discussed in contexts that align with what you want to be known for.
Then move to influence. Public relations aimed at South African tech journalists. Thought leadership that actually contains thought, not repurposed press releases. Participate in industry events that get covered. Create content good enough that people reference your conclusions even when they do not bother linking to your source. The aim is density of association: your brand plus specific problems, specific solutions, specific sectors.
Structure that content for AI synthesis. Use clear statements of what you do and for whom. Maintain consistent framing across pages. Avoid the clever ambiguity that human copywriters sometimes prize. An LLM reading “we help ambitious brands unlock synergistic growth” learns nothing. One reading “Kadabra AI builds custom machine learning pipelines for South African financial services companies” learns an association it can retrieve.
The Measurement Problem Nobody Has Solved
We do not yet have clean ways to measure whether ChatGPT or Gemini is recommending your brand. You can run manual prompts and log results. You can watch referral traffic from AI interfaces, though attribution is murky. You can track brand search volume as a lagging indicator. But there is no Search Console for LLM citations, no ranking report for “mentioned in response to user query about remote team tools.”
This means the strategy requires faith in mechanism over immediate feedback. You build the associations, you monitor the mentions, you trust that frequency and context in training data translate to visibility in generated responses. It is unsatisfying. It is also where the field is heading.
What to Do This Week
If you are running SEO for a South African business, the actionable split looks like this. Keep your link-building programme, but audit whether your brand mention strategy even exists. Check your NAP consistency across local directories. Set up Google Alerts if you have not. Search your brand name on Perplexity and Gemini with queries relevant to your services. See what comes back. If nothing, or if competitors dominate, you have a gap that backlinks alone will not close.
The algorithms are diverging. Google’s classical system still wants links. The emerging AI layer wants to know whether you are a real entity, discussed in real contexts, for real reasons. Unlinked brand mentions were never the secret backlink SEOs hoped for. They are something else, something now more consequential, and the practitioners who adjust their measurement and their tactics accordingly will be the ones whose brands show up in the answers people actually read.
