If you’ve been tracking AI visibility for more than a few months, you probably noticed that AI models are substantially more volatile compared to traditional search rankings – and the next few months are likely going to bring this home more than ever before.
Q1 of 2026 already gave us a preview of what’s coming. OpenAI, Anthropic, and Google all shipped major model updates within the same two-week window, including Claude Sonnet & Opus 4.6, OpenAI’s GPT-5.3, and Gemini’s 3.1 Pro. In the wake of these updates, Claude even overtook ChatGPT in U.S. App Store downloads.
What’s more, agentic AI has seen a meteoric rise in the last three months thanks to the hype around OpenClaw, which OpenAI swooped up along with its creator in February, and the more recent release of a major Claude update that enables the AI to control its users’ PCs.
While these developments arguably constitute the most important AI events since the initial release of ChatGPT – both in general and for generative engine optimization – several signs indicate it’s going to get even more intense very soon.
This piece covers the most noteworthy shifts already in motion and what they mean for brands building a GEO strategy through Q2 2026.
| Area | Likely Q2 2026 shift |
|---|---|
| AI search behavior | Greater emphasis on source trust, corroboration, and freshness |
| Brand visibility strategy | More pressure to combine GEO, PR, reviews, and strong entity clarity |
| Measurement | Higher demand for recurring prompt tracking and mention-quality analysis |
| Competitive landscape | More brands entering GEO, increasing pressure on weak or generic tactics |
The AI audience is becoming more volatile
As already mentioned, March saw the Claude app climbing to the number-one spot on both the Apple App Store and Google Play in the United States, displacing ChatGPT from the top position. Appfigures showed Claude reaching roughly 149,000 daily U.S. downloads versus 124,000 for ChatGPT. Google’s Gemini, while also popular, sat much further behind.
That shift didn’t come from nowhere. Initially, OpenAI’s Pentagon partnership kicked off a new boycott movement that pushed a meaningful chunk of consumers to try alternatives. But as someone who uses both ChatGPT and Claude on a daily basis, it’s pretty clear that Anthropic’s models are just well ahead of its competitors for most tasks a regular user may need at this time.
Furthermore, it’s not just users switching from one platform to another. Many are already choosing AI tools by use case – coding in one, research in another, shopping in a third – rather than defaulting to a single platform.
As of March, ChatGPT is still the dominant AI engine for most consumer and business use cases. But even if it can hold this position, which is far from guaranteed, we can almost certainly expect the market share to be much more spread out than previously.
For brands, this means optimising for one engine is no longer a viable long-term strategy.
Continue reading: How ChatGPT Decides Which Businesses to Recommend
New model releases will make AI visibility harder to predict
While we’ve just seen new model releases across the board in February, all major AI labs are already preparing to churn out novel models in the next few months. These frequent updates aren’t necessarily new, but the sheer volume of overlapping releases and the more substantial advances compared with previous years are creating a compounding visibility problem: each model update can change what gets cited, how sources are filtered, and which content formats get prioritised.
GPT-5.4, which OpenAI already rolled out to Codex users in March, is likely going to reach regular app users in the coming weeks. Having already used this model for a few weeks, it’s a definite improvement over the 5.3 version, but don’t expect anything groundbreaking here. And the effects on AI visibility are likely to be minimal.
What we really need to talk about are the more substantial upgrades expected in the coming months.
Most notable is Anthropic’s Mythos (aka Capybara) model. The development of this model was accidentally leaked by Anthropic around the end of March and has since been confirmed by the company in an exclusive Fortune interview.
Early information describes this model as a significant upgrade to Opus, with even further improved reasoning and coding capabilities – even a potential global cybersecurity risk. Crucially, this model is also expected to be considerably more expensive than Opus, which means it is most likely not going to be used by the average consumer, at least in its early days. Still, it will likely set a new benchmark for AI models.
Similarly, OpenAI is rumored to be working on a brand-new model (codenamed Spud), which might be released in the coming months as GPT 6.0, while DeepSeek is expected to release its V4 model as early as April.
Last but not least, prediction markets currently assign a 45 percent probability to a release of Gemini 3.5 before the end of June.
Even where the exact product details remain unconfirmed, the direction is clear: labs are shipping or preparing more model tiers, which makes both audience distribution and visibility mechanics less stable. For GEO teams, each model update should be treated like an algorithm update – something that can change citation patterns, source preferences, and content formatting requirements without warning.
Recommendation logic is getting stricter
The models aren’t just getting more powerful. They’re getting pickier about what they cite.
Semrush found citation shifts that are the strongest available evidence here. When ChatGPT dramatically reduced its reliance on Reddit and Wikipedia in September, the drop wasn’t caused by those sites getting worse.
Semrush’s Head of Organic and AI Visibility, Sergei Rogulin, attributed it to “an attempt to avoid over-citing on certain websites, to be less biased toward them, while generating answers.” In his assessment, “ChatGPT has become more resilient to manipulation attempts.”
The data backs this up. ChatGPT’s total number of cited sources jumped roughly 80% in October, indicating a deliberate shift toward using more – and more diverse – sources in each response. Google AI Mode moved in the same direction, though more modestly, with a 13% increase in source diversity over the same period.
Meanwhile, the winners and losers from ChatGPT’s citation shift are telling. For example, PRNewswire, Forbes, and Medium gained citation share. TechRadar, Tom’s Guide, and Investopedia lost it. The pattern suggests ChatGPT is actively rebalancing away from sources it was previously over-indexing on, and toward a broader mix.
Claude adds another dimension. Anthropic’s web search documentation makes clear that Claude “uses its reasoning capabilities to determine whether the web search tool would help provide a more accurate response” and can “conduct multiple progressive searches, using earlier results to inform subsequent queries.”
In practice, Claude tends to be more selective and prestige-biased in its source selection than ChatGPT – favouring authoritative primary sources over aggregator content.
For brands, this has practical consequences. The easy-win GEO tactics like paid press releases and listicles on small blogs are likely to decay fastest as recommendation logic tightens. The filtering bar will likely rise across all major models, even if each one applies it differently.
Agentic search changes the optimisation target
Until recently, GEO has been primarily about conversational search: a user asks a question, an AI generates an answer, and your content either gets cited as a supporting source or it doesn’t. The optimisation target in that world is relatively clear – be the kind of source that models trust and reference when assembling a response.
Agentic search works differently. Instead of supporting an answer, source selection starts to support a decision or an action. The AI isn’t just telling the user what it found. It’s comparing options, filtering based on criteria, and in some cases completing transactions on the user’s behalf.
This starts with so-called query fan-out, which breaks questions into subtopics, issues multiple simultaneous queries, and retrieves deeper web content than a traditional search.
Yet these are just the tip of the iceberg. The biggest strides in agentic AI are arguably being made by OpenClaw (now in the hands of OpenAI) and Anthropic’s Claude. These new agentic features fundamentally change how users interact with digital content.
Having significantly more context about their users, agents can already perform highly complex tasks autonomously, which include selecting products based on their users’ preferences, making autonomous phone calls, and even making purchases without involving the user at any point.
Moreover, Google launched Universal Commerce Protocol in January 2026, co-developed with Shopify, Etsy, Wayfair, Target, Walmart, and endorsed by more than 20 additional partners, including Visa, Mastercard, Stripe, and American Express.
UCP creates a standardised way for AI agents to browse catalogs, add items to carts, and complete checkout – directly within AI Mode and the Gemini app. In March, Google added cart management, real-time catalog access, and identity linking for loyalty programs. This is a live commerce infrastructure designed for agents to act on, and it’s likely going to intensify over the coming months.
Google also introduced Business Agent – a feature that lets shoppers chat directly with brands inside Search results, functioning as a virtual sales associate. Combined with new Merchant Center data attributes for conversational commerce, Google is building a full stack where AI agents don’t just recommend products but help buy them.
Forrester predicts this is where the market is heading. The firm expects five major U.S. or European brands to unify “agentic commerce” experiences this year, and predicts that one-third of retail marketplace projects will be abandoned as answer engines capture traffic that previously went to standalone platforms.
The implication for GEO is fundamental. In conversational search, citation supports an answer. In agentic workflows, source selection supports a decision or a transaction. “Being cited” becomes less important than “being usable” – having structured product data, machine-readable pricing, clear availability, and integration-ready business information that agents can act on.
Continue reading: What is the editorial tax (and what does it mean for GEO)?
The bottom line
Q2 2026 probably won’t bring one dramatic AI visibility revolution. Although we shouldn’t completely rule this out, there’s unlikely to be a single event that rewrites the rules overnight.
What’s more likely – and arguably harder to manage – is cumulative instability. User attention keeps fragmenting across engines. Model releases keep reshuffling citation logic. Filtering keeps getting stricter. And agentic search keeps expanding what “discoverable” actually means.
The brands that adapt fastest won’t be the ones chasing each individual shift. They’ll be the ones that treat AI visibility as a moving system – one that requires continuous monitoring, flexible strategy, and the willingness to optimise for machines that are themselves changing every quarter.
Continue reading: 5 Steps to Build External Validation for AI Recommendations
That’s the real challenge of GEO in 2026. The target doesn’t just move – it multiplies.
