What Google I/O 2026 Means for Brands: AI Mode, Search Agents, Universal Cart

Last week's Google I/O is being treated like another feature drop. It is not. Three of the announcements - Gemini 3.5 Flash as the default in AI Mode, Search Agents this summer, and Universal Cart - change who you compete with in search and what 'visible' means.

What Google actually shipped

The three things that matter for brand visibility:

1. Gemini 3.5 Flash is now the default model in AI Mode, globally. Google says it outperforms 3.1 Pro on agentic benchmarks at Flash speed. Faster answers means more follow-ups per session - your content has to survive turn 2 and turn 3, not just turn 1. Source: blog.google.
2. Search Agents. Google calls it 'the biggest upgrade to Search in 25+ years'. Information Agents launch this summer for AI Pro & Ultra subscribers - users build and customize agents inside Search. The query stops being a string and becomes a brief.
3. Universal Cart. Agentic checkout inside Search, plugged into the agent layer. The shopper of late 2026 is not browsing PDPs - the agent is. Your product page has to convince a model, not a human.

Gemini Omni and Android XR got the keynote applause. They are not where your traffic problem lives this quarter. Source: 100 things announced at I/O 2026.

How people actually use AI Mode

The week after I/O, Google published its first usage report for AI Mode in the US. Aleyda Solis' read on X is the one I would trust over the press recap - directional insights, not a complete picture, because the underlying data is sampled Google Trends. Useful all the same.

What the report shows people actually doing in AI Mode (the 10 topic categories from Google's slide):

• Creative / Content Generation. Writing, brainstorming, image and media editing.
• Media / Entertainment. Film, TV, gaming, music, ticketing.
• Education / Knowledge. Factual information, study help, quizzes.
• Personal / Professional Development. Career, finance ('What do I do with my 401k if I am changing jobs?').
• Fashion / Beauty. Product recommendations + style advice.
• Productivity / Practical Tasks. DIY, how-tos, translation.
• Food / Drinks. Restaurant, bar, cafe recommendations.
• Technology / Coding. Setups, reviews, learning resources.
• Health / Wellness. Routines, nutrition, medical info ('I hate cardio - give me a routine that avoids it but still works').
• Travel. Trip planning, hidden gems, local guides.

The behavioral shift behind these categories matters more than the categories. Three patterns from Aleyda's read:

1. Queries are 3x longer than classic Search and conversational. Keyword research has to be paired with prompt, task, constraint and scenario research. 'Help me find a romantic restaurant for date night this Saturday with intimate seating, candlelight, pescatarian and vegetarian options' is one query, not five.
2. Follow-up queries grew 40%+ per month. The unit of analysis is the journey, not the query. A brand can be mentioned at turn 1, dropped at turn 2, compared at turn 3, misrepresented at turn 4. Track all of it.
3. AI Mode decides, not just discovers. 'Which' queries grew 40% faster than AI Mode overall. Planning queries grew 80% faster. Top retail attributes the model looks for: price, location, color, brand, availability, size, material, style, type, quality. Local/availability intent is loud - 'near me', 'in stock', 'replacement parts', 'car dealerships with financing'.

The new reality, in one paragraph

The 7 moves to ship in the next 10 days

Ordered by leverage. If you only have time for three, do 1, 2, and 5.

1. Audit robots.txt for the current bot names. Open yourdomain.com/robots.txt and confirm you are allowing: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended. The deprecated anthropic-ai line does nothing. GEOlikeaPro's Visibility Vitals checker flags missing bots in 60 seconds.
2. Rewrite your top 10 H2s as buyer questions. Pull your 10 highest-traffic pages. If an H2 reads 'Why Choose Us', kill it. Replace with the actual question a buyer types or speaks - 'How long does shipping take to Germany?'. Each H2 gets a 40-80 word direct answer below it, then the supporting prose. Biggest passage-retrieval win I've seen in the last year of audits.
3. Validate Product schema against the Universal Cart playbook. Open Google's Rich Results Test and validator.schema.org. Required fields for agentic shopping: name, image, brand, offers.price, offers.priceCurrency, offers.availability, aggregateRating, shippingDetails, hasMerchantReturnPolicy. Missing any of the last three and the agent passes you over for a competitor whose feed is clean.
4. Reconcile your price, stock, and policy across site + schema + Merchant Center. The agent cross-checks. If your PDP says $49 and your Merchant Center feed says $52, you get filtered out for inconsistency, not for price. This is a data-quality problem dressed as an SEO problem - audit weekly until it stops drifting.
5. Add visible dates and real author bylines to your top 20 pages. 'Last updated: May 2026' under the H1. Real human author name linking to a real bio with credentials. Stale 2023 stat blocks get filtered out of AI Mode answers - I have watched it happen in our 50+ brand audit. Free, takes one afternoon.
6. Run your 5 most-cited claims through a corroboration check. Pick 5 claims from your homepage and best-selling PDPs. Google each one in quotes. If only your own domain returns, you have a credibility problem the model will detect. Pitch a quote to 2 trade publications, get 3 customer reviews referencing the specific claim, add an external citation to a primary source. AI models reward triangulation.
7. Get your brand entity into Wikidata + 3 expert round-ups. When a user spins up a personal Information Agent for 'best running shoes for flat feet under $150', the agent runs against trusted entities first. If your brand is not in Wikidata, not in 3 expert round-ups, not in 2 forum threads with your name as the answer, you are not in the candidate set. Slowest move on the list. Start now.

How to audit yourself in 30 minutes

Two free audit checklists:

• GEOlikeaPro's Visibility Vitals. Our own free checker - paste a URL, get bot-allowance + schema + extractability + freshness flags in 60 seconds. Built for the post-I/O bot list.
• Aleyda Solis' 5-shift measurement framework. Her post-I/O take on how AI Mode changes what you measure: (1) keywords -> prompts, tasks, constraints; (2) rankings -> presence, citations, representation accuracy; (3) single queries -> follow-up journeys; (4) content-only optimization -> entity, product, local and feed-level readiness; (5) observed traffic -> a more nuanced view of visibility and influence. Source: Aleyda Solis on X.

What I am watching next

1. How Google labels Information Agent citations differently from current AI Overviews. If the citation card is bigger, the click-through case for being cited gets stronger fast.
2. Whether Universal Cart pulls from Product schema only, or also from OfferCatalog / Merchant Center / MCP product feeds. That answer decides who needs to ship a feed by Q3.
3. Whether Search Agents respect category-level brand entity preferences (the 'I trust this brand for running shoes') or treat every query stateless. Different brand strategies depending on the answer.

GEOlikeaPro tracks your brand across ChatGPT, Perplexity, Claude, Gemini, and Google AI Mode in one dashboard - scored on bot access, schema completeness, citation share, and entity recognition. See where you stand - free.