Google ships Gemini 3.5 Flash and Gemini Spark at I/O, AI Max for Search graduates and retires Dynamic Search Ads, Google launches Universal Commerce Protocol with Shopify and Walmart, AI Ultra subscription drops to $100

Welcome to the Daily Marketing Brief. Your daily AI news and tactics for marketers who move fast. I'm your host, Jen Bryan, and here's today's update. The most useful observation today is that Google has spent the last 24 hours collapsing the AI vendor question, the subscription question, the ads question, and the shopping question into a single distribution surface.

That is not a model launch, it is a procurement reset, and it lands at the moment Anthropic has overtaken OpenAI on US business spend, and OpenAI has tied itself to Dell on-prem. The choices a CMO ahead of paid media, an e-commerce operator or an agency lead makes in the next six weeks will shape the rest of 2026. Four stories today, and they are all Google, which is appropriate for the 48-hour window between I.O.

and Marketing Live. First, Gemini 3.5 Flash and the new model stack. Second, Gemini Spark and the Agentic Personal Assistant Story.

Third, AI Max for Search graduating from beta and dynamic search ads being formally retired. Fourth, the Universal Commerce Protocol and what it means for e-commerce distribution. Two watch list items at the end on Gemini Omni and the AI Ultra subscription restructure and the usual three actions for the week. First up, Google ships Gemini 3.

5 Flash as the default cross-search, Gemini app and developer stack. What happened? At Google I.O.

2026, yesterday evening UK time, Google launched Gemini 3.5 Flash as the new general purpose flash tier model. It replaces Gemini 3.1 Flash across the Gemini app, AI mode in Google search, and the anti-gravity developer environment.

A higher tier Gemini 3.5 Pro was announced for June. The headline benchmark is that 3.5 Flash now outperforms the previous Generation 3.

1 Pro on multiple categories Google chose to publish. Terminal bench 2.1 GDP Val AA, MCP Atlas, and Chart 5 Reasoning, while running at approximately four times the output token speed of competing frontier systems. What is confirmed?

The launch, the immediate availability inside the Gemini app and AI mode, the benchmark categories and the four times speed claim are confirmed by Google's keynote and corroborated by TechCrunch, CNBC, NGadget, and 9 to 5 Google. Pricing for direct API access has not yet been published. Gemini 3.5 Pro is confirmed for June, but the exact date is not.

Why it matters? The pre-Keynote consensus was Gemini 3.2 Flash. What Google actually shipped was a full version jump to 3.

5 Flash and Flash beating the previous Pro tier is the structural signal. Google has decided to compete on price performance against the cheap inference platforms RAMP called out in its May AI index, not on capability ceiling against Claude Mythos or GPT 5.5. If the four time speed claim survives independent verification, the unit economics of high-volume AI workloads, content generation, support classification, retrieval augmented pipelines, agentic loops have just shifted in Google's favor for the rest of 2026, what it means for business.

For any operator running AI through an API today, your procurement review for Q3 needs Gemini 3.5 flash in the candidate set, especially for high throughput tasks where pro tier capability was not necessary. The deployment scale matters more than the benchmark numbers. 3.

5 flash is the default for AI mode in search and the Gemini app, which is the largest single AI distribution surface Google has ever shipped. Expect downstream effects in AI search visibility, voice assistant quality, and workspace AI features over the next two weeks. What it means for marketers and agencies, three reads. First, AI search visibility programs need to reset their baseline.

Gemini 3.5 flash powering AI mode means the result pages your clients have been optimizing against in March and April are not the same surface they will see in June. Update your reporting templates with a volatility flag and tell clients explicitly that the next four weeks of AEO data are not comparable to the prior eight. Second, for any agency running multi-vendor model deployments, the FlashBeats Pro Framing makes a clean cost case for moving classification, summarization, and retrieval workloads from Claude or GPT 5.

5 to Gemini 3.5 Flash. Run a controlled side by side this month. Third, for engineering-led marketing operations, server-side tag automation, attribution pipelines, lifecycle triggers.

The four times speed claim is the practical lever, not the benchmark numbers. My take. The skip from 3.2 to 3.

5 is the most interesting bit. Google had a flash refresh in market and chose to leapfrog the version it had been leaking inside iOS Gemini App Strings. That tells you Google rebuilt the model post-Mythos and post-GPT 5.5 to compete specifically on cost and speed, and that the version numbering is now marketing as much as engineering.

Take the benchmark claims with the usual skepticism, these are Google's own numbers, but the deployment scale is the harder fact. The 48-hour window where Gemini 3.5 Flash becomes the default Gemini experience for billions of users is a structural event, whatever the benchmarks say. Confidence level, high on the launch and the immediate availability, medium on the four times speed and benchmark leadership claims until independent third-party benchmarks land.

Speculative on whether 3.5 Flash holds the price performance lead once OpenAI and Anthropic respond. Next up, Google launches Gemini Spark, a 24-7 personal AI agent that runs in cloud VMs and works while devices are off. What happened?

Google launched Gemini Spark at I.O.2026, described on stage as a personal AI agent that helps users navigate their digital life and take action on their behalf under their direction. Spark is built on Gemini 3.

5 and operates inside the anti-gravity development environment Google previewed earlier this spring. The agent runs on dedicated virtual machines inside Google Cloud rather than locally, which is why Google can claim it continues working when the user's phone or laptop is off. Spark integrates with Gmail, Calendar, Docs, Sheets, and Slides at launch, with third-party Gemini connected apps expanding to include Canva, Open Table, and Instacart. Rollout is to trusted testers this week and US AI Ultra subscribers next week.

What is confirmed? The launch, the Gemini 3.5 base model, the anti-gravity harness, the Cloud VM architecture, the integrations list, and the AI Ultra Distribution Path are confirmed by Google's Keynote and TechCrunch, Yahoo Tech, CNBC, and Crypto Briefing. UK and EU availability dates are not confirmed.

Pricing outside the AI Ultra bundle is not disclosed. Why it matters? Spark is Google's most serious attempt at the personal agent category, and the architectural choice is the substantive bit. Running the agent on dedicated virtual machines in Google Cloud rather than on device or inside the Gemini app gives Google control over the long-running task surface that ChatGPT agent and Anthropics Claude Skills have been competing for.

It also positions Gmail, Calendar, and the Workspace Stack as the primary distribution mode for personal agents. Google did not have a credible cross-app agent before yesterday. It does now what it means for business. Three commercial implications.

First, for any business with a workspace tenancy, expect employee spark usage from June onwards, regardless of whether the IT team has formally approved it. Update your acceptable use policy and your DLP rules before the rollout reaches your domain. Second, for SOS vendors whose products live inside the productivity stack, Spark integration is the new distribution play. The companies that landed in the I.

O. integration list, Canva, Open Table, Instacart, got a positioning gift. Companies that did not need to be calling their Google Partner Manager this week. Third, for B2B brands selling into knowledge worker buyers, expect inbound research and procurement workflows to be brokered by Spark on a meaningful share of accounts by Q4.

Your sales collateral comparison pages and pricing structure need to be readable by an agent. What it means for marketers and agencies. The lifecycle implications are large and underweighted. If a meaningful share of a customer's email triage, calendar booking, and document drafting is happening through Spark, your email open rates, your CRM journey triggers, and your retention nudges are reaching the agent first, not the human.

Treat that as the working assumption for Q3 planning. Practical actions. Audit your top 10 customer-facing emails for whether they would still trigger the intended action when summarized by an agent. Redesign the landing page to conversion path on the assumption that a fraction of arrivals are agent initiated and brief your CRM team on what Gemini integration looks like for your stack.

For agencies, the product e Sable offer is a workspace agent readiness audit, pricing, governance, integration's lifecycle implication sold to enterprise clients in May and June before the broader rollout begins. My take the most useful question to ask about Spark is not whether it works well at launch, but whether the Cloud VM architecture becomes the standard pattern. Running personal agents in dedicated cloud environments solves the always-on problem and the multi-device synchronization problem, but it concentrates a substantial amount of personal data in one provider stack.

That is a real privacy story over the next 12 months and a real procurement question for any regulated business. Expect Apple to counter with a private cloud compute pitch at WWDC on June 8th, and expect Anthropic to ship a more bounded equivalent inside Cloud Skills inside the quarter. The category is open. Google moved first, but the lock-in is not yet decided.

Confidence level high on the product launch, medium on the trusted tester rollout schedule, speculative on the share of knowledge worker time that moves to Spark inside the next six months. Too early to model with credibility. Next up, AI Max for search graduates from beta with a 7% conversion uplift claim, and dynamic search ads are retired from September 2026. What happened?

Ahead of Google Marketing Live, which opens its keynote at 4.45 p.m. UK time today, Google's Director of Product Management for Ads, Brandon Irvin, confirmed that AI Max for Search has formally graduated from beta after 11 months.

Google stated benchmark is an average 7% uplift in conversions or conversion value at similar CPA or ROAS when the full feature suite, search term matching, text customization, and final URL expansion is enabled. In parallel, Google announced that dynamic search ads automatically created assets and campaign level broadmatch are being retired. From September 2026, all eligible campaigns running those formats will be auto-migrated to AI Max for search. What is confirmed?

The graduation, the 7% benchmark, the September 2026 forced migration date, and the deprecated formats are confirmed by Google's official blog, Search Engine Land, Marketing Dive, and Yahoo Finance. The current 7% figure is lower than the 14% Google cited at AI Max's original launch in May 2025. Also confirmed, Google has published an upgrade tool that preserves campaign history, settings, and asset data when moving DSA campaigns into AI Max standard ad groups before the auto migration window.

Why it matters. Dynamic search ads have been the production workhorse of long tail and discovery paid search for 15 years. Retiring DSA inside one quarter is the largest force change to Google Ads campaign structure since smart bidding became the default. Most agency account books still have meaningful spend running through DSA, automatically created assets or campaign level broadmatch.

Those line items have a hard sunset date of September. The 7% uplift number is the second signal. It is materially lower than the original launch claim, which suggests the easy gains are gone and that performance now depends on landing page quality, asset library depth, and feed hygiene rather than on the AI doing the heavy lifting. What it means for business, for any brand running paid search at scale, three operational decisions need to be made before August.

First, whether to upgrade existing DSA campaigns proactively using Google's upgrade tool or wait for the forced migration. Proactive upgrade preserves more control. Second, whether the asset library landing page set and product feed are AI max ready. Text customization and final URL expansion both depend on it.

Third, whether your reporting layer can isolate AI max performance from the legacy campaign structures inside the same account. The 7% benchmark is achievable only when the full feature suite is enabled, and most agency accounts today have at least one of those three features turned off. What it means for marketers and agencies. This is the single most billable story of the week.

The DSA retirement is a defined scope, time-bound project across every paid search client on the books. Three productizations open. First, a DSA to AI Max migration audit, scoped as a four-week engagement, priced in the £4,000 to £8,000 range per client, covering campaign inventory, asset library readiness, feed hygiene, and account structure. Second, an AI Max performance accelerator in the landing page asset and feed work, scoped as a 90-day retainer add-on.

Third, a quarterly Google Ads platform briefing as a retained service for clients who want a single source of truth on the AI Max Performance Max and YouTube performance era reset over H2. For in-house teams, the equivalent project is a Q3 plan that completes proactive migration before August, allowing a clean read on AI Max performance from September onwards. My take. The 7% figure is the part to interrogate.

At launch, Google cited 14% uplift on the full feature suite. Eleven months later, with the same feature set, the working benchmark is half of that. The honest interpretation is that the AI now performs better than DSA on average, but not dramatically better. And that part of the original headline number was sample selection or early mover effect.

The right test is your own A slash B inside a representative account, not Google's published claim. The harder, less negotiable fact is the forced migration. DSA is gone after September. The work is going to happen whether or not the uplift number holds.

Front load the project before client budgets tighten in autumn. Confidence level high on the graduation, the migration timeline, and the deprecated formats. Medium on the 7% uplift holding at account level once volume normalizes. Speculative on the long tail performance implications of the migration.

The data does not exist yet at scale. Lastly, Google launches Universal Commerce Protocol backed by Shopify, Etsy, Walmart, Target, Visa, and MasterCard, and demos Universal Cart. What happened? Google launched the Universal Commerce Protocol UCP as an open standard for Agentic Commerce designed to let AI agents act across the entire shopping journey from discovery to post-purchase support.

UCP is interoperable with the agent payments protocol, the agent-to-agent protocol, and the model context protocol. Initial endorsing partners include Shopify, Etsy, Wayfair, Target, Walmart, ADIN, American Express, Best Buy, Flipkart, Macy's, MasterCard, Stripe, The Home Depot, Visa, and Zalando. A consumer side Universal Cart was demonstrated at I.O.

as the buyer-facing hub for agent-led purchases with direct buying enabled inside AI mode in Google Search and Inside Gemini. What is confirmed? The protocol launch, the open standard framing, the interoperability with AP2, A2A, and MCP, the partner list and the Universal Cart demo are confirmed by Google's blog Search Engine Land, PIMT's Constellation Research and Shopify's own engineering post. Google has published the developer documentation on developers.

google.com slash merchant slash UCP. What is not confirmed is the timeline for general availability of agent-led checkout at scale or the commercial terms for retailers participating in UCP. Why it matters.

This is the most significant agentic commerce announcement of 2026 to date, and the partner list is the structural point. Shopify, Etsy, and Walmart's involvement gives UCP coverage across roughly the entire English-speaking e-commerce buyer journey, not a slice of it. The payment side participation visa MasterCard, Addy and Stripe, American Express, turns UCP into a credible standard rather than a Google product. And the interoperability with Anthropic-led MCP and Google's own AP2 and A2A means UCP, can compete on openness with whatever Anthropic and OpenAI ship in the same category.

For the first time, the Agentic Shopping Conversation has a defined protocol stack and a defined set of retailer endpoints. What it means for business, for any e-commerce brand selling through Shopify on Walmart, on Etsy, or directly through Google Shopping, the next product feed update needs to be done assuming UCP readable attributes. The Universal Cart demo at I.O.

is the consumer side endpoint, but the harder commercial implication is on the discovery side. A meaningful share of product comparisons, specification reading, and basket building will move into Gemini and AI mode over the next two quarters. Brands that have already invested in structured product data, review velocity, and price history credibility will compound. Brands relying on hero brand copy and atmospheric category pages will quietly lose share to better structured competitors.

What it means for marketers and agencies. Three actions this month. First, audit your top 50 SKU against the UCP attribute schema in Google's merchant documentation. Identify where your feed is incomplete, where prices change without history, where review counts are thin.

Second, brief client retail leadership on UCP this week with a one-page summary, the partner list, the interoperability with anthropic-led standards, and the timing. Third, if you sell SEO or e-commerce services, productize a UCP readiness audit alongside the existing AEO baseline. The two questions sit next to each other on the client agenda from now on. For brands negotiating Shopify renewal or commerce platform decisions in H2, UCP support is now an evaluation criterion.

My take. The Universal Commerce Protocol is the part of yesterday's announcements that has the longest commercial tail. Models will get faster and personal agents will get more capable. But the protocol layer that lets agents transact across retailers is the durable infrastructure shift.

The fact that Shopify worked on this with Google and that Anthropic-led MCP is one of the interoperability targets, tells you the agentic commerce category is settling around open standards rather than a single dominant vendor. The risk for operators is overinvesting in the consumer side universal cart demo and underinvesting in the structured data work that makes UCP work for them. Treat UCP as a product data hygiene project, not a creative campaign. Confidence level, high on the protocol launch and the partner list, medium on the consumer side rollout timing, speculative on the share of shopping volume that moves through UCP mediated agents by year end.

The demand side is not yet measurable. Two things for the watch list. One, the Gemini Spark Trusted tester window runs over the next 10 days with USAI Ultra Availability next week. Watch for the first independent reviews of how Spark handles long-running tasks, calendar conflicts, and email triage.

The early reviews will set the procurement bar for the entire personal agent category for the rest of the year. UK and EU availability is the related signal. If Google holds back launch in Europe pending Digital Markets Act assessment that tells you the regulatory window is narrower than the marketing window. Two, Gemini OmniFlash is now live for AI Plus subscribers and rolls out free inside YouTube Shorts and YouTube Create later this week.

10 second clip cap, synth ID, watermarking on all outputs. The Omni Pro tier was teased without a date. Worth watching as the lead competitor to OpenAI Sora and ByteDance C Dance for creative production budgets. The question for marketers is the cost per finished asset against VO Nano Banana and the existing multimodal video creation stack inside YouTube ads.

What matters most? The pattern across today's four stories is consolidation. Google has used the 48-hour window between I.O.

and Marketing Live to collapse the model question, the subscription question, the ads question, and the shopping question into a single distribution stack. AI Max retires 15 years of DSA. Gemini 3.5 Flash becomes the default on the largest single AI surface in the West.

Universal Commerce Protocol turns agentic checkout from a roadmap item into a partner program. Gemini Spark gives the productivity stack a credible cross app aid. Agent for the first time. Signal, anything that affects the procurement, the account structure, or the feed and asset library you actually operate.

Noise, any model leaderboard published this week, including Google's own. What I would do if I were running this account, brand, or agency this week. One, if you run paid search, kick off the DSA to AI max migration this week, not in August. Inventory, every campaign using DSA automatically created assets or campaign level broadmatch.

Estimate the migration effort in hours by client price the engagement at four to eight thousand pounds per account depending on complexity, and book the work into June and July. Two, if you run AEO or AI search visibility, treat the next four weeks of data as non-comparable to the previous quarter. Reset your baseline, communicate the reset to clients in writing, and rerun the top 20 branded query audit against AI mode powered by Gemini 3.5 Flash.

3. If you run e-commerce or sell into e-commerce, schedule a feed and structured data audit against the UCP attribute schema for your top 50 SKUs. The work overlaps with shopping feed hygiene, most accounts already need. 4.

If you run a knowledge work team inside a workspace tenancy, brief the IT and security leads on Gemini Spark this week. Get the acceptable use policy updated and the DLP rules tested before US Ultra rollout reaches your domain. Build a list of three to five workflows where Spark would deliver measurable hours back. Pilot one in June.

Five, if you advise B2B clients, write a one-page Spark readiness note covering sales, collateral readability, pricing page, agent comprehension, and inbound research workflow design. Send it to the top 10 accounts by Friday. What I'd tell a client today: one, dynamic search, ads are gone in September. The migration is a four to twelve week project.

Start it now or pay a premium for rushed work in August. Two, Gemini 3.5 Flash now powers AI mode in Google search. Your AI search visibility baseline needs to be reset, and the next four weeks of data are not comparable to March or April.

Three, the Universal Commerce Protocol turns agentic shopping into a real distribution channel. Your product feed quality is now a strategic question, not an operational one. The story of the week is not which Google product launched. It is that Google has used its event window to compress the model, the subscription, the ads platform, and the shopping protocol into a single procurement decision.

The brands and agencies that treat the next eight weeks as a procurement and account structure project, DSA Migration, AEO baseline reset, UCP feed audit, spark readiness, will end the quarter in better shape than the ones who treat it as keynote viewing. Briefing back tomorrow with the post marketing live debrief and the first independent benchmarks on Gemini 3.5 Flash. That's been today's episode.

See you tomorrow.