by SeanEvery hot take after Google I/O 2026 has landed in roughly the same place. AI Mode is the future. Universal Cart kills retail SEO. Marketers need a new acronym to survive.
That summary is neat, dramatic, and incomplete.
Google’s actual announcements point at something more useful. Search is becoming more multimodal, more agentic and in some cases more transactional, but Google is still grounding those experiences in its core Search systems and still serving classic results alongside them. Google’s own wording: “You’ll continue to get a range of results from Search.”
Google’s own answer to whether SEO still matters is “still SEO.” The harder problem is that AI Overviews and AI Mode can now surface your brand, summarise your content, compare your products and in some cases move shoppers towards checkout before your analytics stack sees a familiar site visit.
Search is getting harder to observe with old reporting habits.
One caveat upfront. Google’s May 2026 core update started rolling out on 21 May and is still active as of 28 May. Any ranking or traffic movement you’ve seen this week is sitting inside the overlap between I/O announcements and a live core update.
What’s actually live this week
The first useful thing we can do is separate what shipped at I/O from what’s rolling out, what’s geo-limited and what’s still “in the coming months.” Hot takes flatten that distinction. Here’s how it actually breaks down.
| Topic | Status as of 28 May 2026 | Implication |
|---|---|---|
| AI Mode + new AI Search box | Live, rolling out where AI Mode is available | Treat as a real behaviour shift. Blue links survive |
| Generative UI, simulations, mini apps in Search | Announced for summer; some features subscriber-gated first | Use cautious language; don’t write as if the SERP has already been replaced |
| Universal Cart + direct checkout | Rolling out in the U.S.; direct checkout limited to eligible merchants and certain product types | Strong for ecommerce strategy; do not overstate availability |
| UCP geographic expansion | Canada and Australia first; U.K. later | Plan for staged readiness |
| Business Agent + Merchant Center AI insights | Rolling out, largely U.S. and market-limited first | Excellent proof that brand and product data observability is becoming more important |
What Google actually launched
Three things actually matter. Each one extends a pattern Google has been running for a year.
AI Mode is the new layer on Search
Google said AI Mode has surpassed 1 billion monthly users and that queries have been more than doubling every quarter since launch. Gemini 3.5 Flash is now the default model in AI Mode globally. The Search box itself has been rebuilt as an AI interface that accepts text, voice, images, files, videos and Chrome tabs. Antigravity sits behind it, generating dashboards, simulations, trackers and mini apps inside the result page. Google was explicit that these experiences sit alongside traditional results.
The blue-link SERP is still there. The dominant interface for new search behaviour has changed.
Shopping is moving closer to checkout
Universal Cart now works across Search, Gemini, YouTube and Gmail. It can use Google Wallet, offer data, loyalty signals and compatibility logic to help users choose. But this is where the hot takes get it wrong. Direct checkout via Google is available with eligible U.S. retailers, with restrictions for some product categories, and the merchant remains the seller of record. UCP geographic expansion is planned for Canada and Australia first; the U.K. is later. The direction is clear. “Universal” is still more roadmap than reality.
Merchant Center is becoming a discovery surface
Google has been adding new Merchant Center attributes designed for discovery in conversational commerce, including product Q&A-style details, compatible accessories and substitutes. On 20 May 2026, Google announced AI performance insights that will show a retailer’s share of voice on AI surfaces versus similar brands in Australia, Canada, India, New Zealand and the U.S. in the coming months. Business Agent extends merchant-side levers into conversational interactions. SEO-adjacent work is shifting closer to product feeds, structured data and merchandising operations every quarter.
None of this is new
Sophia, our head of content, had this to say…
“Brand authority stopped being a shortcut. A strong, well-known site can no longer carry thin or generic pages on reputation alone. Pages were scored on their own merits, and what held up was depth, specificity and a tight match between the URL, the content and the search term. Broad category hubs and loosely-matched pages slipped, even on sites that ranked comfortably before.”
“The range of results widened. Editorial articles, videos, community threads, comparison sites and smaller specialist retailers entered commercial searches that used to be held by a handful of established names. The best answer wins the slot now, whatever type of source it comes from.”
I/O extended that direction across more surfaces.
Why the real disruption is measurement
Two reasons.
The first sits in Google’s own documentation. AI Overviews and AI Mode sit inside Search Console’s overall Web performance reporting. The separate AI report doesn’t exist. If your brand appears in AI features you may see the downstream click, but you will not get a native first-class view of how often AI surfaces mentioned you, summarised you or preferred a competitor before the click happened.
Independent research is starting to put numbers on the gap. One paper accepted at SIGIR 2026 found AI Overviews on 51.5% of an 11,500-query representative benchmark, with less than 0.2 average Jaccard similarity between classic Google results and AI Overview cited sources. A separate preprint covering 55,393 trending queries found 13.7% AIO activation overall, rising to 64.7% on question-form queries, and 11.0% of atomic claims unsupported by the cited pages. Both studies are early. Both should be read carefully. Both point the same direction. AI surfaces do not mirror classic top-10 results, and your standard rank-tracker does not see them.
The second reason is that Google itself is acknowledging the gap by shipping new merchant-side observability. Merchant Center AI performance insights launching in Australia, Canada, India, New Zealand and the United States “in the coming months” will show share of voice on AI surfaces versus similar brands. Retailers are getting a partial fix on the commerce side. Other industries are still waiting.
If standard reporting were enough, Google wouldn’t be building parallel observability tooling.
Welcome back to 1980s marketing.
The internet briefly let marketers feel like physicists. Track every click. Model every funnel. Attribute every conversion to its proximate cause. AI mediation has just stripped that. The discovery layer is moving onto Google-mediated surfaces that don’t give you a session, a referrer or a query string. Your dashboard sees less every quarter.
It’s a return to the discipline most of the marketing canon has been built around.
Binet and Field, Byron Sharp, the IPA Effectiveness Awards. Twenty years of evidence has been pointing at the same thing. Brand-building drives long-term share. Mental availability beats funnel optimisation. Marketing mix modelling, econometric attribution and incrementality testing are the right tools for any business serious about growth. The performance-marketing fashion that put last-click ROAS at the centre of every dashboard was a temporary blip enabled by digital media’s exceptional measurability.
AI mediation is just removing the exceptional measurability. The toolkit that replaces it is the one we had before digital made us forget.
Contribution beats attribution.
The right starting point is the philosophical reframe. Stop asking which click gets credit. Start asking what your marketing contributed to the decision.
Attribution is a backward-looking exercise that argues over a fixed pool of credit. Contribution is a forward-looking exercise that measures effect at the margin. The difference matters because attribution depends on observable touchpoints, and the observable surface is shrinking. Contribution measures lift against a counterfactual: what would have happened if this marketing didn’t run? You can answer that question even when the customer journey is mostly invisible.
Marketing as a skill again. Analytics is what you do when the data behaves like physics. Marketing is what you do when the data is thinner than that.
Pre-cognitive analytics: the layer the dashboard never saw.
Consumers were already making most of their decisions before any conscious search action. Awareness, recall, salience, peripheral exposure. The iceberg below the click that Sutherland, Sharp and the IPA effectiveness canon have been pointing at for years.
Digital analytics never measured any of that. It measured the click. The fact that the click correlated with sales, for a particular period in time, when discovery still funnelled through a query box, was the convenient illusion that propped up the entire performance-marketing dashboard.
AI Mode, Personal Intelligence, Antigravity-generated answers and Universal Cart all move more of the customer journey into that pre-cognitive layer. The dashboard’s view of it shrinks proportionally. The honest measurement frame from here is one that takes the iceberg seriously. Brand search volume. Aided and unaided awareness. Share of voice. Salience. Behavioural-economic indicators that have always been the leading signal but rarely got dashboard real estate because the click was easier to count.
Place your bets. Measure the effect.
The operational shape that replaces last-click reporting is brand-marketing-as-portfolio. Make a bet. A campaign, a brand investment, a content piece, a PR play, an offer architecture. Measure the lift in branded demand, in incrementality-tested conversion volume, in mental availability scores. Iterate the bets that lift. Park the ones that flatline.
This is what marketing measurement looked like before the internet. It’s what marketing measurement looks like for any organisation big enough to run an MMM or sophisticated enough to run incrementality holdouts. The tools are mature: Robyn, Meridian, Lightweight MMM, the vendor stack from Measured and similar, established incrementality methods across Meta, Google and TikTok ads, POAS layers where margin maths is uneven. The work is putting the old discipline back at the centre of the dashboard.
Close enough counts in horseshoes. Marketing measurement was always horseshoes. Digital gave it a false-precision skin. AI’s stripping the skin off.
Uplift-based measurement.
Uplift modelling is the cleanest form of contribution thinking we have. It asks one question. How much more did our outcomes move with the marketing than they would have without?
Holdout groups, propensity-matched controls, geo-experiments, switchback tests, incrementality vendors. The methods range from elegant to expensive. The principle is the same. Measure the lift.
For in-house teams operating under tightening AI mediation, uplift is the metric the CFO actually wants. Not “we contributed to 47% of conversions” because that’s attribution dressed up. Uplift is “we increased qualified revenue by 14% over the holdout.” Defensible. Repeatable. Comparable across channels. And, crucially, invariant to which surfaces Google adds next.
Sense check from our SEO team
Dave, our Head of SEO, gave me his thoughts:
“Google’s AI features rely on the same index and ranking signals as traditional search, which means the groundwork already in place matters more than ever. Ensuring your site is properly crawlable, that Google can interpret your content through additional methods like structured data, and that your brand demonstrates genuine expertise and trustworthiness, that’s precisely what positions a business well for this new landscape. It’s a validation of the right approach.”
One thing to stop
Investing in content that answers questions Google can now handle itself. Basic FAQs, generic explainer pages, thin articles. AI surfaces all of that without sending a visitor to your site. The E-E-A-T work matters now precisely because Google wants to know whether that brand is a credible source worth citing.
One thing to start
Treat your own knowledge and experience as your most valuable SEO asset. Google’s AI has to link to sources it can’t reproduce, so original insight, real expertise, and content only your business could publish is what earns a click in this environment. This thinking extends to commerce too. Google’s Shopping experience is already evolving towards AI-driven results that pull product data, pricing and availability directly into search. Ensuring that product information is accurate, structured and complete is how you show up where customers are increasingly finding things.
Where the industry is getting it wrong
SEO isn’t dead. Google’s own line, posted on the @NewsFromGoogle X account earlier this week: “AI Mode is not the default experience in Search. Our new search box helps you describe exactly what you’re looking for, but using it does not mean that you will only get AI features. You’ll continue to get a range of results on Search.” Traditional results remain. The bar for earning those clicks is rising, and the businesses that have been investing in quality, trust and technical soundness are best placed to clear it.
The right conclusion
Google I/O 2026 has reshaped SEO. It has introduced more AI-mediated ways for people to discover brands, compare products and in some cases buy without leaving Google’s environment as early in the journey as they used to. Google’s own guidance still points back to the same fundamentals. Recent research suggests AI surfaces pull from a wider and different source set than classic results. Google’s own product roadmap is adding new merchant observability tools because the old ones are incomplete.
SEO survives.
