Meta wants to decide where your ads run.
Microsoft wants to decide which searches count, what the ad says and where the click lands.
Google will now build the video—and is testing whether the landing page needs to be involved at all.
This sounds like another week in which the machines took over more of media buying. It was. But the more interesting story is what moved in the opposite direction.
Responsibility.
The platforms are taking more execution off your hands. They are not taking responsibility for weak product data, bad creative inputs, broken fulfilment logic, misleading reports or channels that look efficient only because you left half the costs out.
That is where the operator work is moving.
Meta Is Testing the End of Placement Exclusions

Some Meta advertisers are seeing an in-product notice that says excluding placements, platforms, devices and operating systems will no longer be available at the ad-set level. Meta ads specialist Jon Loomer reports that this is currently a test, appears limited to Sales and Leads objectives and has no public date for a broader rollout. Meta’s own public Reels guidance still describes manual placements, which is a useful reminder that this is not yet a clean platform-wide change. (Jon Loomer, Meta)
That distinction matters. Do not tell a client Meta has killed manual placements everywhere because somebody posted a screenshot from 1 account.
But do take the direction seriously.
If the test expands, the control does not disappear completely. It moves. Loomer identifies account-level placement controls and value rules as the remaining levers. Instead of simply switching a placement off inside an ad set, an advertiser may need to define a business-wide restriction or adjust how Meta values the inventory. (Jon Loomer)
For an operator, the practical move is an audit: where are you excluding placements now, and why? Separate genuine brand-safety or lead-quality rules from inherited account folklore. Then make sure every eligible placement has creative that actually belongs there.
The machine can choose inventory. It cannot turn a lazy crop into native Reels creative.
Google Changed the Report Before It Changed the Business

On August 24, Google changed Merchant Center performance reporting. YouTube affiliate traffic now gets its own category instead of sitting inside “Organic,” and Google changed how YouTube organic clicks and impressions are defined. Those 2 changes also restate historical data back to July 1. Google warns that merchants may see a 1-time drop in reported organic traffic even when actual demand has not changed. (Google Merchant Center)
Paid product-level reporting is expanding too. Merchant Center will include product performance from all ad channels and formats, including Performance Max networks plus Video, App and Demand Gen campaigns. That can create a 1-time increase in reported impressions and clicks. (Google Merchant Center)
This is not a glamorous update. It is exactly the kind that creates a bad Monday meeting.
Organic appears to fall. Paid product exposure appears to rise. Somebody diagnoses a channel problem, changes a feed or moves budget, and only later discovers that Google moved the labels.
Annotate the reporting change. Preserve a pre-change baseline. Break YouTube affiliate out from free listings. Before “fixing” anything, ask whether performance changed or the accounting did.
Microsoft’s AI Max Is Global—and Defaults Matter

Microsoft Advertising made AI Max available to all Search advertisers on August 27. The optional suite combines search-term expansion, AI-generated text and final URL expansion, which can route a click to the page Microsoft believes best matches the query. Advertisers retain negative keywords, budgets, bidding, brand controls and reporting on generated assets and landing pages. (Microsoft Advertising)
Microsoft’s early data deserves both attention and restraint. Across 44 advertiser-run A/B experiments, it reports a spend-weighted conversion uplift of about 13.6% with AI Max enabled. But only 10 experiments produced statistically significant positive results on conversions or conversion value. This is vendor data, not a universal forecast for your account. (Microsoft Advertising)
Here is where it gets interesting: AI Max is enabled by default for newly created Microsoft Search campaigns, though it can be turned off. Google imports can also carry AI Max settings into corresponding Microsoft campaigns. Existing auto-generated asset settings continue as text customization without automatically opting the campaign into the full suite. (Microsoft Advertising)
So “optional” does not mean “nothing changes unless I deliberately test it.”
Check new-campaign defaults and scheduled imports. If you test AI Max, avoid turning on broader matching, generated copy and URL expansion and then calling the combined result a clean diagnosis. Final URL expansion deserves particular scrutiny for DTC stores with out-of-stock products, weak collection pages, subscription paths or promotional URLs that should not receive paid traffic.
Google Can Build the Video. It Is Also Testing the Conversation.

Google’s August Demand Gen update made Multimodal Video Creation in Asset Studio generally available. Advertisers can move from storyboarding to horizontal and vertical video assets in the same workflow. Google is also testing a different path: letting YouTube viewers start a conversation with a brand on a messaging app directly from a Demand Gen ad. (Google Ads)
Those are 2 different maturity levels. Video creation is generally available. Messaging is a test.
The first lowers production friction. It does not eliminate creative judgement. Generated variations still need human review for product accuracy, offer language, visual oddities and whether the opening earns attention.
The second could create a lower-funnel route that bypasses the standard landing page. That is useful only if somebody owns the conversation. Before testing messaging ads, define response-time expectations, qualification, consent, CRM capture and the event that counts as success.
A new destination is not automatically a shorter funnel. Sometimes it is just a different place to lose the lead.
Shopify Is Making Fulfilment a Checkout Variable

Shopify announced 3 connected fulfilment changes on August 24: shipping options organised by market, mixed shipping and pickup inside 1 checkout, and clearer default split-shipment information. The rollout is staged through July 1, 2027 rather than arriving as a single switch. (Shopify)
The most immediate piece is ship-and-pickup in 1 order, available now to Plus and Enterprise merchants through Test Drive. Market-based shipping opt-in begins October 1, 2026. Shopify says the new fulfilment setup, mixed delivery options and qualified split-shipping defaults will be broadly enabled by July 1, 2027. (Shopify)
This is a conversion story disguised as an operations story.
A customer should not need 2 orders because 1 item can be collected and another must ship. International buyers should see a rate that reflects their market rather than a fight between delivery profiles. A split order should not become a post-purchase surprise.
The caveat is app compatibility. Shopify says custom and third-party apps that read, create or modify delivery profiles may need updates. Stores using delivery, fulfilment or order-management apps should test mixed carts, pickup inventory, international rates and split shipments before holiday volume makes the edge cases expensive. (Shopify)
ChatGPT Ads Reached 31 More Markets. Measurement Still Has to Earn the Budget.

ChatGPT Ads went live across 31 European markets on August 24. At launch, advertisers access the channel through OpenAI’s Ads Solutions team, agency partners and technology partners; self-service was still to follow. Ads appear on Free and Go plans, while Plus, Pro and Enterprise remain ad-free. (OpenAI)
OpenAI says the platform now supports conversion optimisation, geo-targeting, custom audiences, an OpenAI Pixel, Conversions API and third-party measurement integrations. It also says ads remain labelled and separate from answers, with conversations kept private from advertisers. (OpenAI)
This is a real advertising channel now, not a speculative slide about where search might go.
It is also young. An ad beside a high-intent conversation may be valuable, but a new surface should not inherit budget merely because the surrounding product is fashionable. Early operators should decide what ChatGPT Ads is supposed to do: create demand, intercept active evaluation or close a decision already formed elsewhere. Those are different jobs and need different tests.
I would be reluctant to compare its CPC with Meta or Google before examining downstream conversion quality, new-customer mix and incrementality. A cheaper click in a novel interface can still be an expensive sale. A costly click can still be useful if it enters unusually late in a considered purchase.
The Industry Is Trying to Measure a Journey With No Clicks

The IAB is developing an AI-era attribution framework for classifying and crediting conversions influenced by AI. Its own initiative page describes categories such as agent-initiated, agent-recommended and signal-triggered outcomes. Digiday reports that the work is still being negotiated and that publishers, platforms and measurement companies disagree about what deserves credit and which signals can be trusted. (IAB, Digiday)
The problem is simple to describe and hard to solve.
An AI system can read a review, compare products, recommend a shortlist and influence a purchase without producing the familiar chain of ad impression, site visit and checkout. Even a crawler visiting a page does not prove that the sponsored information was processed, surfaced to a user or changed the decision. Digiday’s sources also warn that platforms may retain the black box by controlling the signals outside measurement firms need. (Digiday)
The framework is not a finished standard. Until definitions settle, operators should resist dashboards that turn “AI visibility” into a precise-looking revenue claim. Track AI referrals where they exist. Ask customers how they discovered the brand. Watch branded search, direct traffic and assisted conversions. Use incrementality where the spend justifies it.
The absence of a click does not mean the absence of influence. It also does not give every influence vendor permission to invent certainty.
BK Beauty’s “Organic” Channel Now Needs 1,000 Samples a Month

This surprised me.
BK Beauty co-founder Paul Jauregui told Practical Ecommerce that the brand pulled back from TikTok Shop after its early organic period became more pay-to-play. Conversions on other channels then slowed, and the company returned in 2026 because it sees TikTok Shop as a discovery engine whose effects carry into the rest of the business. (Practical Ecommerce)
The return is not cheap. Jauregui says GMV Max is now essential to acquisition, the system needs a heavy volume of content, and BK Beauty increased creator samples from roughly 200 per month to 1,000. The brand’s reported channel mix is about 50% on its own site, 20% TikTok Shop, 20% Amazon and 10% Ulta. These are founder-reported figures from an interview, not audited results. (Practical Ecommerce)
There is a useful operating lesson here. Social commerce is not “organic” merely because a creator made the post. The economic stack can include GMV Max spend, samples, commission, creator retention, content management, marketplace fees and internal labour.
BK Beauty also reported selling more than 20,000 units in 24 hours during a 50%-off Ulta promotion, after the marketplace gave the brand premium placement during its 21 Days of Beauty event. That shows the other side of distribution: margin can buy relevance, and retail presence can restore share of voice during a tent-pole moment. (Practical Ecommerce)
The right question is not whether TikTok Shop or Ulta “works.” It is whether the total contribution and cross-channel lift justify the full cost of being visible there.
The Bigger Picture
The visible theme this week was automation.
The deeper theme was control moving upstream.
Meta may remove a placement checkbox, but somebody still has to define brand-safety rules and make creative for the inventory. Microsoft may expand queries, write copy and select a URL, but somebody still has to govern the site it reads. Google may generate the video and start the message thread, but somebody still has to make the claim accurate and answer the customer. Shopify may coordinate fulfilment at checkout, but somebody still has to make the inventory, rate logic and apps agree.
Meanwhile, ChatGPT Ads and AI-mediated discovery are making the path to purchase harder to observe just as platforms ask for more trust in their automation.
That changes the job of a strong growth operator. The edge is less about knowing where every switch lives. It is more about giving the system better inputs and building guardrails around the decisions it can make:
- clean product and conversion data;
- creative built for the actual placement;
- landing pages the machine is allowed to choose;
- fulfilment promises the operation can keep;
- contribution economics that include the whole channel cost; and
- measurement that admits what it cannot prove.
The machines are getting more freedom.
Your advantage is knowing what they should never be trusted to decide alone.
About Phill J. Smith

Phill J. Smith is the founder and creative strategist behind Frozen Lemon Media. He writes about marketing, media and technology through the lens of human behaviour, drawing on 13+ years across media buying, copywriting and creative strategy.
All writing & ideas