The 2026 State of AI Search: what the data says about visibility, content, and the brands getting left behind

Somewhere right now, a potential customer is asking an AI assistant which brand to choose in your category. The AI is generating an answer. Your name may or may not appear. You have no way of knowing either way.

This is not a hypothetical. Across millions of queries and hundreds of product categories, AI search systems are making recommendations, surfacing comparisons, and building consumer understanding of which brands matter, entirely without the participation of the brands being discussed. The question every content and ecommerce team should be asking is not whether this is happening, but whether they are positioned to benefit from it or be bypassed by it.

Research compiled across major platforms including ChatGPT, Gemini, Perplexity, and Google's AI Overviews, maps the signals that determined who appeared and who didn't over the course of 2025. The findings are both clarifying and, for many brands, uncomfortable.

Freshness is not optional anymore

The most consistent signal across the entire dataset is one that brands consistently underestimate: recency. AI models treat content age as a direct proxy for trustworthiness. If a page has not been updated recently, the model's assumption is that its information may no longer reflect reality, and so it reaches for something newer instead.

The numbers make this concrete. Pages that go more than three months without an update are over three times more likely to lose visibility in AI-generated answers compared to recently refreshed pages. More than 70% of all pages cited by AI models were updated within the past twelve months. For pages earning citations in the past six months, the figure climbs even higher, suggesting that the freshness window is narrowing, not widening, as more content competes for the same citation slots.

For commercial queries, the kind that drive purchasing decisions, the bar is even higher. Around 83% of citations in commercial search contexts come from pages updated within the last year, and more than 60% from pages refreshed within six months. In fast-moving categories like fashion, beauty, and consumer electronics, models apply an even tighter window: pages older than three months see steep drops in citation likelihood.

For retail content teams, this demands a significant recalibration. The legacy model of publish once, optimise once, and leave it to perform is not compatible with how AI systems evaluate trust. A product description written eighteen months ago, a category page last touched during a site migration, a buying guide refreshed only when a product line changes: all of these are liabilities in an environment where recency is a retrieval signal.

The instinct, for many teams, will be to treat this as a content volume problem. It is not. It is a content maintenance problem, which is a different kind of challenge entirely. The question is not how to produce more content; it is how to ensure that the content you already have remains current, accurate, and competitive on an ongoing basis. That is an infrastructure question, not a campaign question.

Structure is how models understand you

Freshness gets you into consideration. Structure and coherence is what determines whether your content can be understood clearly enough to be cited.

AI models do not read content the way humans do. They interpret it, extracting meaning from signals like heading hierarchies, schema markup, and the organisation of information into scannable units. Pages that provide these signals consistently are cited at substantially higher rates than those that do not.

The data is specific: pages that follow a clean, sequential heading structure are cited at roughly 2.8 times the rate of those with fragmented or inconsistent hierarchies. Among pages cited within ChatGPT, more than 68% follow logical heading structures. Nearly 87% use a single H1 as the primary anchor. The model uses this architecture to understand what a page covers, how ideas relate to one another, and whether the content is likely to answer a given query reliably.

Schema markup compounds the effect. Pages using three or more schema types show a meaningfully higher likelihood of being cited than those using one or none. FAQ schema, in particular, appears in a disproportionate share of cited pages, mapping content directly to question structures, which is exactly the format AI models are trying to match when generating answers. Nearly 80% of pages cited across major AI platforms include structured lists, which make information easier for models to extract and re-present.

The implication for retail is direct. Enterprise product catalogs, category pages, and buying guides are often structured for visual appeal or legacy SEO conventions, neither of which is optimised for AI extraction. A category page that uses beautiful imagery and atmospheric copy but lacks clear heading hierarchy, structured attributes, and machine-readable organisation will underperform against a less polished competitor whose content is architecturally cleaner.

This is a tension that many teams will recognise. Brand aesthetic and structural clarity are not inherently in conflict, but achieving both requires intentionality that goes beyond most current content workflows.

The trust layer has moved off-site

Here is the finding that should most unsettle any brand relying primarily on its own website for AI visibility: approximately 85% of brand mentions in AI-generated answers originate from third-party pages rather than owned domains. Brands are between six and seven times more likely to be surfaced through external sources than through content they publish themselves. In one word, the brand is not in the room anymore.

This is not a temporary anomaly. It reflects something fundamental about how AI systems establish trust. When a model is asked which brand to choose, it does not simply search for that brand's claims about itself. It looks for external validation: the accumulated signals of what other sources, communities, and authoritative voices have said. A brand's own content matters, but it functions more as corroboration than as primary evidence.

The sources driving this external trust are consistent across platforms. Nearly half of all AI search citations come from user-generated and community sources: Reddit, YouTube, LinkedIn, Wikipedia, niche forums. Reddit alone appears in roughly one in five AI-generated answers, functioning as the model's reference point for peer validation and authentic user experience. YouTube is the second or third most-cited source on most major platforms, supporting educational and category-level queries. Perplexity references community platforms in more than 90% of answers; Gemini in as few as 7%, which illustrates that the weight each model places on these signals varies considerably, and that optimising for a single platform is its own form of risk.

Structured third-party formats are particularly influential. Around 90% of third-party brand mentions come from listicles, comparison pages, and review roundups: the formats that aggregate and rank options. Brands that appear consistently in these formats, and near the top of them, gain a structural advantage that no amount of owned content can replicate.

The implication is that AI visibility is, in large part, an off-site reputation problem. Brands that have invested heavily in their own content infrastructure but neglected their external presence across publications, review platforms, community forums, and comparison sites, are likely to find themselves significantly underrepresented in AI-generated answers, regardless of the quality of what they publish on their own domains.

Visibility is volatile by design

One of the more disorienting findings in the data concerns the nature of AI search visibility itself. Unlike traditional search rankings, which are relatively stable over days and weeks, AI-generated answers are rebuilt from scratch each time a query is submitted. The model reassesses, rebalances, and arrives at a new answer, which may or may not include the same brands as the previous run.

The consequence is that only around 30% of brands remain visible across two consecutive answers to the same query. Just one in five maintains consistent presence across five successive runs. For brands accustomed to thinking about SEO in terms of stable positions and predictable traffic, this represents a fundamental shift in how visibility should be understood and measured.

The good news, such as it is, is that most disappearances are temporary. More than half of brands that drop from an answer resurface within two runs. The model is not forgetting them permanently; it is cycling through sources to provide diversity and freshness. Brands with stronger signals, fresher content, broader citation presence, and clearer off-site validation, return more quickly than those without.

This points to a different framework for measuring AI visibility. Rather than asking "do we rank?", the relevant question is "how consistently does our brand return?" Consistency of reappearance is a function of signal strength, not position. Brands that earn both citations (direct links to their content) and mentions (references without links) show 40% higher likelihood of resurfacing across consecutive answers. Yet only about 28% of AI-generated answers include brands with both types of signal, meaning dual-signal presence is rare, and therefore disproportionately valuable.

Zero-click is reshaping the funnel, without eliminating it

Google's rollout of AI Overviews has accelerated a pattern that was already emerging: more queries resolving inside the results page without a click through to any website. Zero-click behaviour has increased substantially following the AIO launch, and Google's own usage data tells an interesting story: visits to Google increased by around 9% over a nine-month period, while time-on-site and pages-per-visit declined, as users found answers faster and left sooner.

For brands, the temptation is to treat this as pure loss. It is more nuanced than that. Pages cited inside AI Overviews saw page views grow by more than 21% in the period following the rollout, compared with just over 1% growth for pages not surfaced in Overviews. Appearing inside an AI Overview drives exposure even when it does not drive a click. The brand is seen. The brand is associated with the answer. That is not nothing.

What is perhaps most striking is the source composition of these citations. Around 60% of AI Overview citations come from URLs that are not ranking in the top twenty organic results. This is a meaningful departure from traditional search logic, where visibility was tightly correlated with ranking position. AI systems are drawing on a wider, differently-weighted pool of content, and brands that assume their organic positions will translate directly to AI visibility are likely to be wrong.

The practical consequence is that the AI content strategy and the traditional SEO strategy are related but not identical. Brands need both. But they need to understand that what earns them a position in organic results is not always what earns them a citation in AI-generated answers. The game has more than one board.

The shifts that defined 2025

It would be easy to read all of this as a set of disconnected tactical observations. Taken together, they describe something more coherent: a fundamental restructuring of how search works, who it favours, and what content must accomplish.

2025 marked the year AI Overviews went from experiment to standard, now appearing on a significant share of queries and resolving an increasing proportion of them directly in the results page. Google reinforced its emphasis on experience and trustworthiness as ranking signals, favouring content that demonstrated genuine expertise over content produced at scale without editorial grounding. Search became multimodal and conversational: users are no longer only typing keywords, they are asking questions, uploading images, and expecting synthesised answers rather than link lists.

Off-site sources including Reddit, YouTube, comparison roundups, and editorial publications moved from supporting signals to primary citation sources for AI systems evaluating brand credibility. Freshness became table stakes in a way it was not even eighteen months ago. And the link between traditional organic ranking and AI citation proved weaker than most practitioners had assumed.

For retail brands, each of these shifts compounds the others. A brand whose content is structurally well-optimised but chronically stale will lose ground to a competitor that publishes less elegantly but maintains its pages consistently. A brand with strong owned content but minimal off-site presence will be underrepresented in precisely the answers where buying decisions begin. A brand with good organic rankings but no explicit strategy for AI citation may find that its traditional performance masks a growing gap in the channel that is capturing an increasing share of early-stage discovery.

What this means for your content operation

The research, taken together, describes a visibility landscape that is structurally different from the one most retail content teams were built to navigate.

It is not enough to publish good content. That content must be maintained at a quarterly cadence, at minimum, that keeps it inside the freshness window models apply when assessing trust. It must be structured in ways that allow models to extract and interpret it accurately: clean heading hierarchies, rich schema, organised lists, direct answers to the questions your customers are actually asking.

And it must be supported by a presence the brand does not directly control. The reviews, the forum mentions, the comparison pages, the editorial roundups: these are not peripheral to AI visibility. For many brands, they are the primary mechanism through which AI models understand what the brand is, what it offers, and whether it deserves to be recommended.

None of this is comfortable for organisations accustomed to controlling their own narrative. The brand voice that took years to develop cannot survive intact through every AI intermediary that synthesises, paraphrases, and recontextualises it. What can survive is the factual substance beneath the voice: the clear product attributes, the specific differentiation, the honest description of who a product suits and who it does not. The AI extracts facts, not atmosphere.

The brands that will earn consistent visibility in AI search are those that treat content not as a creative output but as an infrastructure layer, one that requires active maintenance, systematic structure, and external validation to perform.

The data is unambiguous on this. Whether your content operation is built to deliver it is a different question, and one worth asking before the answer is decided for you.