Google Hands Publishers A Tag, Anthropic Cuts Fast Mode To A Third, And 44% Of Merchants Are Already Building Agentic Checkout. Pick One To Catch Up On Today.

The post I/O week looked quieter than last week's keynote barrage, but the moves underneath were just as consequential. Google rolled its Preferred Sources feature into AI Overviews and AI Mode on 27 May. 345,000 sources have already been selected by users, and Google's own data says people are twice as likely to click a link that carries the visible "Preferred" badge. Anthropic shipped Claude Opus 4.8 the next day at unchanged base pricing but a 3x cheaper Fast mode, putting frontier-class coding inside the budget of teams that previously could not run it daily. Ravelin's first Agentic Commerce and Fraud Report dropped on 29 May with the number that should set the next board meeting: 44% of enterprise merchants are already integrating agentic commerce protocols, and only 29% feel prepared for the fraud that comes with them.

Those three moves together describe the same shape. The infrastructure of AI commerce is being built, the cost curve under it is falling, and the trust layer above it is still missing. The week's other stories (OpenAI rolling out brand-customisable ChatGPT ads, Canva's 70% "soulless ads" finding, DeepSeek's permanent 75% price cut) all rhyme with the same theme. Cheap, agentic, and everywhere. The defensive moat for any brand this quarter is no longer cost or distribution. It is being the source the agent picks when it picks one.

TL;DR

Google rolled Preferred Sources into AI Overviews and AI Mode on 27 May; 345,000 sources already selected; users are 2x more likely to click a Preferred-badged link.

Anthropic shipped Claude Opus 4.8 on 28 May; Fast mode now $10/$50 per million tokens, down from $30/$150 for Opus 4.7, with SWE-bench Verified up to 88.6%.

Ravelin's first agentic commerce report: 44% of enterprise merchants already integrating agentic checkout, 32% planning within six months, only 29% feel prepared for the fraud.

OpenAI began rolling out ChatGPT ads with larger images, customisable CTAs and e-commerce layouts that show pricing and customer reviews inside the chat surface.

DeepSeek cut its flagship V4-Pro model price by 75% permanently on 23 May; the frontier-tier price war is now public and pinned.

1. Google Hands Publishers A Tag. The Source Set Becomes A Marketing Surface.

May 27, 2026 | Google rolled out Preferred Sources inside AI Overviews and AI Mode. Users designate which sites they trust most and a visible "Preferred" badge then appears next to any link from those sites inside AI responses. Google says 345,000 unique sources have already been picked by users and that people are roughly twice as likely to click a Preferred-badged link versus a normal AI Overview link. Any website publishing fresh content is eligible. Source: https://ppc.land/google-brings-preferred-sources-into-ai-overviews-and-ai-mode-today/

What I Think: This is the most under-reported move of the year for B2B and high-trust verticals. For 18 months the only AI-search lever a brand had was citation share. Preferred Sources adds a personalisation lever inside the user's own account. Every brand newsletter, every customer onboarding flow, every membership programme now has a reason to teach users to add the brand as a Preferred Source. The 2x click effect makes the ask worth it. The publishers who learn to ask now will see compounding click share before competitors notice the toggle exists.

Why It Matters: Add the ask to your email footer this week. "Add us as a Preferred Source on Google for trusted answers in your industry." For paying customers, put the ask in the welcome flow. The brands that get into 10% of their email subscribers' Preferred Sources before September will have a structural advantage that does not depend on changing Google's algorithm. Two minutes of UX work, one A/B test.

2. Anthropic Cuts Claude Opus Fast Mode To A Third. Frontier Quality Gets A Budget Lane.

May 28, 2026 | Anthropic released Claude Opus 4.8. Standard pricing unchanged at $5 per million input tokens and $25 per million output tokens. Fast mode, however, drops to $10/$50 per million tokens, down from $30/$150 for Opus 4.7. The model scores 88.6% on SWE-bench Verified (vs 87.6% for 4.7), 69.2% on SWE-bench Pro (vs 64.3%), and 74.6% on Terminal-Bench 2.1 (vs 66.1%). The release adds Dynamic Workflows that let Claude plan large tasks and run hundreds of parallel subagents inside one session. Source: https://venturebeat.com/technology/anthropics-claude-opus-4-8-is-here-with-3x-cheaper-fast-mode-and-near-mythos-level-alignment

What I Think: Fast mode at a third of the previous price is the line that matters. It means the model your engineering team rejected in March because the per-call cost did not fit is back inside budget for the same use case. The benchmark gains are real but not eye-watering. The cost cut is the actual story. Combine this with Google's Gemini 3.5 Flash from last week and DeepSeek's 75% cut on V4-Pro, and the per-token cost of frontier-tier inference fell roughly 70% across three of the top four providers in seven days. Procurement teams just had their best week in two years.

Why It Matters: Re-open every AI vendor invoice for the last quarter and run the math on Opus 4.8 Fast. If your engineering team is paying for high-frequency code review, test generation or summarisation on Opus 4.7, the same workload on 4.8 Fast costs a third. Use the savings to fund the visibility audit you have been deferring, or fund the second customer-facing assistant. Either way, the bill is the lever.

3. OpenAI Turns ChatGPT Ads Into An E-Commerce Surface.

May 29, 2026 | OpenAI began rolling out richer ad formats inside ChatGPT. Brands can now customise their ad creative with larger images and call-to-action buttons. A dedicated e-commerce layout shows the product, price band and customer reviews directly inside the chat surface. The updates are framed as preparing ChatGPT to scale as an ad platform, with industry expectation of an OpenAI listing later this year. Source: https://digiday.com/marketing/brands-can-now-customize-how-their-ads-appear-in-chatgpt/

What I Think: This is the moment ChatGPT stops being a "linked AI surface" and becomes an end-to-end purchase surface. The price tag, the reviews and the CTA all live inside the same response. The funnel that used to be search to clickthrough to product page to add-to-cart now compresses to one chat turn. For brands, the question moves from "how do I get cited?" to "how do I get cited with a buy button next to my name?"

Why It Matters: If you are a consumer brand, make sure your product data feed is ready for ChatGPT ad ingestion. Pricing, availability, reviews, hero image, returns terms. The same metadata that powered Google Shopping for fifteen years now decides whether your ad in ChatGPT renders correctly. If you are a B2B brand, watch the lead-generation ad format roll-out. Calendly links, demo buttons and lead capture inside chat are the next two product releases away. Have the brief ready.

4. Ravelin Sizes Agentic Commerce. 44% In, 29% Ready For The Fraud.

May 29, 2026 | Ravelin published the first Agentic Commerce and Fraud Report, surveying 1,504 fraud and payments professionals across 10 countries, all working for enterprise merchants with over $50M annual revenue or 450+ employees. 44% of enterprise merchants are already integrating agentic commerce protocols. Another 32% plan to within six months. Merchants expect AI shopping agents to account for 6% to 30% of transactions within three years. Only 29% said they feel highly prepared for the fraud and security risks that come with agent-driven checkout. Source: https://pages.ravelin.com/agentic-commerce-fraud-report/

What I Think: The 44% number is the one to memorise. Agentic commerce is not a 2027 conversation. It is a 76% of enterprise merchants moving on it inside twelve months conversation. The 29% prepared-for-fraud number is the other side of the same shape. The merchants building the buy buttons are not the merchants building the trust layer to defend them. Expect the first six months of agentic checkout to look like the first six months of mobile commerce: lots of revenue, lots of chargebacks, a few public incidents that reshape policy. The brands that ship policy now will not be the ones answering the regulator later.

Why It Matters: Two memos to write this week. One for the product team: which checkout flows go agent-first in Q3, and what is the fall-back when the agent gets it wrong? One for the fraud and trust team: how do we tell a legitimate Operator-class agent from a malicious one impersonating it? Ravelin's framing flags agent hijacking and impersonation as the top two emerging vectors. Have the answer before the incident.

5. Canva Quantifies The "Soulless Ad" Problem. 70% Can Tell.

May 29, 2026 | Canva published its "State of Marketing and AI 2026" report. The headline finding: 70% of consumers say they can usually recognise AI-generated advertising because it feels "robotic, emotionally flat, or missing its soul." The data does not directly measure ad performance impact, but it sets the bar for consumer detection, which is the first input into any future trust or backlash metric. Source: https://martech.org/consumers-want-ai-ads-with-a-human-touch/

What I Think: The number to watch is not the 70% directly, it is what happens next when "AI ad detection" becomes a TikTok genre. Brands have already lived through the "obviously CGI" moment in cinema and the "obviously photoshopped" moment in magazines. We are in the same opening act for AI ads. The brands that get caught publishing soul-less creative will be screenshotted, mocked and lose a small chunk of brand equity each time. The brands that use AI for ideation and humans for emotional craft will look like they are not using AI at all, which is the desired outcome.

Why It Matters: Audit the last three AI-assisted campaigns your team shipped. Were they signed off by a copywriter who could ad-lib the same scene at a dinner party? If not, send them back. The market is still rewarding "couldn't tell it was AI" creative. The window before "obvious AI" becomes a brand-safety line is six to nine months. Use it.

6. DeepSeek Cuts V4-Pro By 75% Permanently. The Frontier Price Floor Just Dropped.

May 23, 2026 | DeepSeek announced a permanent 75% price reduction on its flagship V4-Pro AI model. The cut is not a promotional period or a usage-based discount. It is a baseline pricing change. Coming a week after Google's Gemini 3.5 Flash launch and five days before Anthropic's Opus 4.8 Fast mode price cut, it confirms the frontier-tier price war is now structural rather than reactive. Source: https://www.reuters.com/world/china/chinas-deepseek-make-permanent-75-price-cut-flagship-v4pro-ai-model-2026-05-23/

What I Think: When three frontier-tier vendors cut prices inside seven days, that is not a coincidence. It is the end of the era when frontier inference was rationed to the customers who could afford it. The CFO conversation about "which model to bet on" gets replaced by "which set of models to mix per workload." The vendor lock-in pitch is now structurally harder to make. Multi-model routing, which was a 2027 architectural debate, just became a 2026 procurement requirement.

Why It Matters: If your AI architecture today is single-vendor, add a router pattern this quarter. Even if you stay on the same provider for sensitive workloads, the optionality is worth the engineering hours. If your team is paying premium frontier prices for workloads that could run on cheaper tiers (retrieval, summarisation, extraction, simple classification), reroute them inside 30 days. The savings fund the next visibility audit.

Travel Desk

For travel, the news this week was the ongoing ripple from last week's Google I/O hotel-booking announcement. Google's six launch partners (Booking.com, Expedia, Marriott International, IHG, Choice Hotels and Wyndham) are now actively integrating against AI Mode booking, with Wyndham named as the first to ship a direct-booking rollout in Q2 alongside its earlier native ChatGPT app for around 8,400 hotels. Against that backdrop, the Skift x Curacity number from 11 May still hangs in the air: 94% of hotels are effectively invisible in AI search, and on Hyatt-specific queries NerdWallet pulls 13.6% of AI citations versus Hyatt.com's 10.3%. This week's Geotravel daily LinkedIn was paused while the agentic shape settles, but the through-line is unchanged: citation share decides booking share inside Google's new stack, and the brands that built the listicle format and the structured property data this winter are the ones the agent will pick when AI Mode bookings go live. Full travel breakdown in this week's GeoTravel newsletter: Geotravel.ai.

The Strategic Reality Check

The post I/O week clarifies the shape of the next two quarters across every industry. Last week's headline was AI Mode hitting a billion users. This week's headlines are what the system around that billion-user surface looks like in practice. A new Preferred Sources tag that 345,000 users have already configured. A 3x cost cut on frontier-class Anthropic. A 75% permanent price cut on DeepSeek's flagship. A 44% enterprise adoption number for agentic checkout. A 70% consumer recognition rate for AI ads. None of these are coincidental. They are the same shift expressed five different ways: the AI commerce stack is being built faster, cheaper and more visibly than the trust, brand and policy layers under it.

For CMOs the strategic posture for Q3 should be obvious by now. The cost of building inside the AI stack just fell. The benefit of being inside the source set just rose. The penalty for ignoring the trust layer is starting to compound. The brands that thread those three together (cheap inference, hard-earned source set position, defended trust) will own the agentic commerce flow when it lands at scale.

Three truths every CMO should write down this morning:

1. AI inference is now a commodity input. Re-open every vendor contract this quarter. The renegotiation is the easiest budget win of the year.

2. Source-set membership is the new domain authority. Preferred Sources, citation share, structured data, fan-out coverage. The brands that win these compound. The ones that do not, do not.

3. The trust layer is now a strategic moat, not a compliance overhead. Agentic checkout is moving without it. The brands that ship policy now will sell against the brands answering regulators next year.

What You Should Do This Week

Add a "Set us as a Preferred Source on Google" link to your email footer and welcome flow today. Two minutes of work, 2x click multiplier, no algorithm to chase. Track the number of customers who add the tag and report it monthly.

Re-open every AI vendor invoice from Q1 and Q2. Recompute on Opus 4.8 Fast, Gemini 3.5 Flash, and DeepSeek V4-Pro. The provider you negotiated with three months ago just gave the procurement team a new floor price. Use it before renewal.

Brief your fraud and trust team on the Ravelin agentic checkout findings. If 44% of enterprise merchants are already moving and only 29% feel ready, your number is probably in between. Decide which agentic checkout vectors you support in Q3 and which you defer until the trust pattern is settled.

Audit the last five AI-assisted creative pieces shipped by your team. If a copywriter cannot defend the emotional arc in plain English at a dinner party, send them back. The "obvious AI" detection meme is forming this quarter. Do not be the brand that powers it.

Pull your AI Assistant traffic report from Google Analytics for the last 30 days and compare to your direct traffic baseline. If you cannot account for the difference, you have your Q3 visibility budget brief written for you.

The frontier just got cheaper and the agentic stack just got more visible. The brand work to be the source the agent picks did not get any easier, but the budget to do it just did.

Think Preferred Sources is just another publisher toggle? Wait for your competitor's email footer to land in your inbox first. Hit reply when it does.

Want to audit your AI visibility before Q3 board meetings? Let us talk: https://calendly.com/vincent-getinference/

See you next week.

Vincent Chief AI Enthusiast, SimplyAI: AI SEO (GEO)