Brand is the only compounding asset in AI search

For eighteen years, the SEO conversation has been organised around a single implicit assumption: that the page is the unit of optimisation. You build pages, you rank pages, you measure pages. Everything downstream — content strategy, technical audits, link building, schema — is a page-level activity dressed up in different language.

That assumption has quietly broken. And most of the industry hasn't caught up.

The evidence has been accumulating for months. iPullRank's analysis of 13.1 billion search events shows zero-click behaviour is now bimodal — some categories are effectively closed loops where Google answers directly, others still generate clicks, and the gap between them is widening. SparkToro just launched brand affinity data in its audience reports because agencies and in-house teams kept asking for it. Rand Fishkin published a piece a few days later arguing that brand monitoring is now foundational to controlling how you appear in AI systems. Google's new Search Console AI opt-out toggle forces every publisher to make a distribution decision that pages can't answer — only brand strategy can.

Read those signals together and a pattern emerges. The industry has spent two years arguing about which pages get cited in AI Overviews, which schema markup helps, which structured data formats matter most. All of it is useful. None of it is the point. The point is that AI systems don't cite pages the way search engines rank pages. They cite entities they've already learned to trust. And the mechanism by which they learn to trust an entity is brand.

This is the piece I've been trying to write for six months. Here it is.

The page-level model is running out of runway

Everything about how the SEO industry thinks was built during an era when the URL was the atomic unit of the web. Google indexed pages. Pages ranked. Traffic flowed to pages. Even link building — which is technically about relationships between domains — got operationalised at the page level, with tools scoring individual URLs on individual keywords.

The unit of retrieval has shifted from the page to the entity. And entities are branded.

That worked because search engines were, effectively, page-retrieval systems. You typed a query, they returned the ten pages most likely to satisfy it, and the game was to be one of those ten pages.

AI-mediated discovery doesn't work that way. When ChatGPT answers a question about content marketing tools, it doesn't retrieve ten pages and let you pick one. It generates an answer that names three or four brands, occasionally with links, often without. When Perplexity summarises a topic, it cites sources — but the sources it cites are dominated by domains it has learned to treat as authoritative on that entity, not the ten pages most closely matched to the query string.

The unit of retrieval has shifted from the page to the entity. And entities are branded.

This is not a semantic distinction. It has direct operational consequences. A single well-optimised page can win in the old model. In the new model, a single page in isolation is almost invisible — because the AI system isn't asking "which page best matches this query?" It's asking "what does the training data tell me about this topic, and which brands are consistently associated with it?"

The businesses that will still get discovered in five years are the ones the models have already heard of. Everyone else is fighting for scraps at the citation layer, and the citation layer is shrinking.

Brand is the only signal that compounds across every AI system simultaneously.

What the SparkToro launches actually reveal

I want to focus on SparkToro because Rand Fishkin's last two moves tell you something interesting about where sophisticated marketers are actually spending time.

The brand affinity feature launched on July 28th. The pitch is straightforward: SparkToro now shows you which brands an audience is most familiar with and most likely to purchase from. In Fishkin's walkthrough, he uses it to identify Humble Bundle, Kotaku, and Devolver Digital as brands his indie-gaming audience recognises. His explicit use case: agencies pitching clients want to demonstrate they understand the brand ecosystem their client operates in, and in-house marketers want to know which brands are competing for their audience's attention.

Five days earlier, he'd published a piece on brand monitoring — arguing that tracking mentions of your brand name across the web is now "super critical for AI." His reasoning: if you want to control how your brand appears in AI tools, you have to start by monitoring how people are already talking about it. Because the language people use to describe your brand becomes the language AI systems use to describe your brand.

Put these two products together and the thesis is clear. The most valuable work in marketing right now isn't optimising your own pages. It's mapping the brand ecosystem around your audience, monitoring how your brand is described in that ecosystem, and shaping that description over time.

Notice what's not in that thesis: keyword research, content briefs, link building targets, schema markup. Not because those things don't matter — they still do — but because they're downstream. They're expressions of a brand strategy, not substitutes for one.

For eighteen years, agencies have been selling brand as a soft, unmeasurable thing you do after you've handled the "real" performance channels. That framing was always slightly wrong. Now it's completely wrong.

The zero-click data is a brand story in disguise

iPullRank's zero-click analysis, published a few days ago, is the most rigorous public data we have on how search behaviour is actually changing. The headline number — around half of Google searches end without a click — is the one everyone will cite. But the interesting finding is buried further down.

Mike King's team found that clicks disproportionately go to sites the searcher didn't name in their query. That's a small phrase with enormous implications. It means the searcher typed something generic — "best CRM for small business," "how to fix a leaking tap," "content marketing platform" — and clicked through to a specific brand they weren't looking for by name.

Which means Google (and increasingly ChatGPT, Perplexity, and Gemini) are effectively acting as brand introducers. They're pointing searchers at brands the searcher didn't already know about. The click doesn't go to the highest-ranking page in some neutral sense. It goes to the brand the algorithm has decided is most representative of the query.

That's not a page-level ranking outcome. It's a brand-level trust outcome dressed up in ranking language.

The other finding worth sitting with: Google absorbs about a third of Wikipedia's would-be traffic through direct answers, but barely dents Reddit's. Why? Because Wikipedia's value proposition is factual answers, which Google can now generate directly. Reddit's value proposition is community, opinion, and lived experience — which Google can't replicate without borrowing Reddit's brand credibility to underwrite the answer.

The pattern holds everywhere. The sites that survive zero-click are the ones whose brand carries meaning the AI can't substitute for. Everyone else is being disintermediated.

<!-- Pull quote will insert after this section -->

The measurement problem hides the brand problem

Here's the trap most enterprise teams are falling into.

The measurement infrastructure for AI search is genuinely broken. Google's Search Console AI Overview reports are opaque and misleading. Citation-tracking tools are rebadged log analysis at 10x markup. ChatGPT and Perplexity provide almost nothing usable in terms of referrer data. The result is that most marketing teams can't tell you, with any precision, how their brand is performing in AI-mediated discovery.

The natural response to that measurement gap is to focus on what you *can* measure — which is still, largely, page-level organic search performance. So teams keep publishing content, keep optimising for keywords, keep tracking rankings and impressions, because that's the data that exists in a dashboard.

The problem is that page-level metrics are increasingly measuring a shrinking share of the discovery pie. The Similarweb data I wrote about last week showed 95% of ChatGPT users still use Google — which means AI search is layering on top of traditional search, not replacing it. But that layering is where the interesting behaviour is happening. And it's the layer your dashboards can't see.

So teams end up optimising harder for the surface they can measure while the surface they can't measure quietly becomes the more important one. That's the loop. And we built it.

The way out isn't better AI search tracking (though that would help). The way out is investing in the one asset that pays off in every channel simultaneously, measured or unmeasured: brand.

When your measurement is broken, invest in the thing that doesn't need measurement to work.

What the opt-out toggle really asks

Google's new Search Console AI opt-out toggle looks, on the surface, like a technical setting. Check a box, and your content is excluded from AI Overviews, AI Mode, and Discover's AI features.

The toggle isn't asking a question about AI. It's asking a question about brand governance.

But NewzDash's tracking data — which I covered when it landed — shows that Top Stories carousels are now rendering inside AI Overviews on trending news results. Which means opting out of AI Overviews may mean opting out of Top Stories placement. Which means the decision isn't really about AI. It's about whether your brand is willing to trade a chunk of surface-level visibility for a defensive posture against generative use.

That's not a page-level decision. There's no version of that decision that can be made URL-by-URL. It has to be made at the brand level, by someone who understands what the brand is worth in each distribution channel and what it's willing to trade.

Most marketing teams don't have that person. They have SEO managers who think in pages, PR managers who think in mentions, and content managers who think in publishing calendars. The person who thinks about the brand as a distributed asset across a dozen surfaces — Google, ChatGPT, Perplexity, YouTube, Reddit, LinkedIn, TikTok, the open web, plus whatever launches next month — often doesn't exist in the org chart.

The toggle isn't asking a question about AI. It's asking a question about brand governance. And revealing that most companies don't have a governance layer capable of answering it.

The counterargument, taken seriously

Here's the strongest version of the objection to everything I've just written.

Brand is expensive. Brand is slow. Brand is hard to measure. For a small business owner or an in-house marketer at a growth-stage company, "invest in brand" often lands as advice that sounds correct but doesn't cash out into anything actionable this quarter. Meanwhile, publishing a page, targeting a keyword, and tracking whether that page ranks — that's a workflow that produces measurable outputs on a predictable timeline. Telling those teams to shift budget from performance to brand feels like telling them to stop doing the thing that works in favour of something vaguer that might work eventually.

That objection has real force. I've made versions of it myself when clients have gotten too excited about brand campaigns that had no plausible short-term ROI.

But it misreads what "invest in brand" means in this context. I'm not suggesting anyone stop publishing pages. I'm suggesting they stop publishing pages *as if the page is the point*. The page is now an expression of the brand, aimed at reinforcing what AI systems already believe about the brand and expanding the entity graph the brand occupies. That's different from publishing pages to rank for keywords in a vacuum.

Practically, that looks like: fewer pages, higher quality, each one clearly authored by a real named person or team, each one linked to and from other assets that reinforce the same entity signals. It looks like earned media that gets your brand mentioned in the venues AI training data draws from. It looks like consistent, distinctive positioning that survives being rephrased by a language model.

None of that requires abandoning performance work. It requires understanding that performance work now compounds only when the brand underneath it is doing the load-bearing.

Where I'd concede

I'd concede two things.

First, this argument is more true for services businesses, publishers, and SaaS than it is for pure e-commerce. If you sell products on Amazon, brand still matters, but the distribution logic is different — you're competing inside a marketplace that has its own algorithms and its own brand hierarchy. Some of what I've written applies. Some doesn't. I'd write a different piece for pure e-commerce.

Second, "invest in brand" is genuinely bad advice if delivered without operational specifics. Most brand campaigns fail. Most brand agencies overcharge. Most brand-oriented marketing produces beautiful decks and no measurable outcomes. The point isn't that brand is magic. The point is that in an AI-mediated discovery environment, brand is the only asset whose returns compound across every channel simultaneously — which makes it structurally undervalued in a marketing culture that still budgets by channel.

The teams that will win the next five years aren't the ones with the biggest brand budgets. They're the ones who understand brand as a distribution mechanic — the thing that determines whether AI systems remember you when someone asks a question you should be the answer to.

What this means for the work

If the page is no longer the unit, what is?

The unit is the entity. Which means the operational shift is to stop thinking about your marketing as a stack of pages and start thinking about it as a coherent signal that AI systems can parse, remember, and reproduce.

That reframes a lot of the everyday work. Content isn't published to rank — it's published to reinforce a specific position the brand is trying to occupy in the AI's model of the topic. Links aren't earned to boost domain authority — they're earned to create the co-occurrence patterns AI training data picks up. Digital PR isn't a reputation tactic — it's the primary mechanism by which brand mentions get into the venues LLMs actually train on. Schema markup isn't for AI Overviews specifically — it's for making your entity legible to any system that reads structured data.

The daily tasks look similar. The strategy underneath them is completely different.

The teams still writing content briefs organised around keywords with search volume attached are optimising for a version of the web that's ten years old. The teams writing content briefs organised around entity positioning, brand associations, and audience overlap are optimising for the version of the web that exists now.

The close

I've been writing about AI search for two years now, and the pieces that have held up best are the ones that argued the fundamentals didn't change — that domain authority, quality content, technical hygiene, and earned media still drive discovery, whether the discovery surface is Google or ChatGPT or whatever launches next quarter.

I still believe that. Nothing in this piece contradicts it.

But there's a subtler point underneath. The fundamentals didn't change, but the *hierarchy* of the fundamentals did. Brand used to be one input among several. It's now the input that determines whether the others compound or evaporate.

If your marketing programme is producing measurable results today but the brand underneath it is thin, you're borrowing against a balance that's going to zero. If your brand is strong but your operational execution is scattered, you can rebuild the execution — the brand is what took years to earn.

The businesses I'm watching succeed in AI-mediated discovery aren't the ones with the best schema, the most content, or the smartest GEO consultants. They're the ones whose brands were built well enough over the last decade that AI systems learned to name them without prompting.

Everything else is downstream of that.