entry 002 · 14 September 2026 · covering 7–14 September

This week is unusually relevant to my client mix. The biggest development is that Google has materially changed the European search results for commercial queries, particularly travel, hospitality and other verticals. At the other end of the stack, the AI-agent conversation is shifting from capability to control. Both point toward the same thing: marketers need to understand the system around acquisition, not just optimize individual channels.

Both point towards the same analytical problem: marketers need to understand the system around acquisition, not just optimize individual channels inside it.

1. Google has changed European commercial search, and I am watching travel particularly closely

On 8 September, Google rolled out significant changes to European search results to comply with the EU’s Digital Markets Act after the European Commission’s July enforcement action.

For categories including hotels, flights and other travel searches, Google is giving greater prominence to third-party vertical search services such as comparison and booking platforms rather than directing users as prominently to businesses themselves.

According to Reuters, some of the new formats can place one vertical search service prominently in the results followed by others, while familiar Google travel features lose information such as real-time pricing.

Google says the changes represent a reduction in Search quality and claims earlier DMA changes had already reduced free direct-booking clicks to European businesses by as much as 30%. That 30% figure is Google’s own characterization.

For a travel business, this can alter the journey without anything changing on the website.

A user searching for a destination or hotel may increasingly move from Google to an aggregator/comparison service and then to a travel provider, rather than going straight from google to travel provider.

That means changes in organic sessions, conversion rate or landing-page mix this month could partly reflect a SERP architecture change, not simply rankings or demand.

For the travel businesses I work with, I am adding 8 September to the CRO and analytics change log and watching:

Google Organic sessions before/after
destination and hotel-query landing pages
Search Console impressions vs clicks
CTR changes where rankings remain stable
branded vs non-branded search
referral traffic from comparison/travel platforms
resulting conversion quality.

If impressions remain stable while clicks fall, that is a very different diagnosis from losing rankings.

This is another reason I increasingly think CRO analysis needs to begin before the landing page. For travel and hospitality clients, I’d now explicitly include a SERP Journey Review in SEO/CRO audits.

The question becomes “What does the customer have to pass through before they can reach us?” The acquisition environment is part of the conversion journey too.

2. This European change makes SEO geography more important than ever

The same business can now encounter materially different Google experiences depending on whether the customer searches from Portugal, the UK or the US. This seems particularly important because last week’s site-reputation changes already introduced another EEA-specific difference.

We’re accumulating a pattern: Search is becoming geographically different at the product and policy level, not merely at the ranking level.

A global Search Console chart can increasingly conceal structural changes. For international clients, we should avoid analyzing “Organic traffic ↓ 12%” without asking: Where?

A decline in Portugal combined with stable UK or US performance might contain information that disappears completely when the data is aggregated.

It might indicate market demand.

Or rankings.

Or competition.

But it might also indicate that Google itself is producing a different search experience in one geography. For businesses spanning EEA and non-EEA markets, I may increasingly default to country → device → query type → landing page before interpreting aggregate organic performance.

It’s a small methodological change, but the more Search diverges geographically, the more dangerous global averages become.

3. AEO tools are racing toward autonomous optimization

A launch just before this week’s reporting window illustrates where AI visibility tooling is heading.

On 4 September, MentionOS launched an autonomous AEO agent designed to monitor how brands appear across systems including ChatGPT and Gemini, identify visibility gaps and then recommend or help execute optimization work.

I wouldn’t treat the product launch itself as evidence that autonomous AEO works, but it reveals something about the direction of the category. First observe, then diagnose, recommend and finally, act.

Eventually clients are going to ask: “Why do I need an AEO consultant when this platform tells me exactly what to change?”

That’s a fair question.

Our advantage is methodological judgment.

A tool might notice that Competitor A appears more frequently and recommend creating fourteen new pages.

Before creating anything, I want to know:

Are the prompts commercially meaningful?
Is the sampling stable?
Is the competitor actually being recommended?
What sources support them?
Would creating those pages duplicate existing content?
Is the visibility gap affecting demand?
Is there a cheaper reputation or information-architecture fix?

The gap might genuinely require new content, or not.

This is why I want my AI Visibility Audit to follow Evidence, diagnosis, business implication and then intervention.

Let software do the counting. We’ll do the reasoning. The valuable part is deciding what the count means.

4. Google’s European changes create an interesting analytics experiment

There wasn’t a consequential new GA4 product release from 7–14 September that I think warrants manufacturing an analytics story around, but Google’s European Search changes create something much more interesting.

For European businesses with meaningful traffic outside the EEA, we may have a useful quasi-experimental comparison group.

Suppose Portugal organic CTR falls after 8 September.

UK CTR does not.

Rankings and impressions remain relatively stable.

That doesn’t prove the DMA interface change caused the decline. Portugal and the UK are not identical markets and this isn’t a randomized experiment. But it creates considerably stronger evidence than “Traffic went down after Google changed something.”

Instead, we can start thinking in terms of what happened, who was exposed to it, who the comparison group it, the expected mechanism and finally the observed behavior.

The mechanism is what makes the analysis useful. If the hypothesis is that Google has inserted more intermediaries between Search and the business, then one expected outcome could be stable visibility with lower direct CTR in the exposed market

Now we have something to test.

I think this kind of reasoning is becoming more important as the systems producing our marketing data become more complex.

5. The AI workflow conversation is moving from “What can agents do?” to “Who controls them?”

Salesforce introduced its Trusted Enterprise AI Harness this week, structured around six capabilities spanning context, agency, actions, governance, security and models. Alongside it, Salesforce announced an AI Control Plane intended to help organizations understand which agents exist, what they can access and do, how they perform, what outcomes they produce and what they cost.

Much of the complete architecture is still future-facing, so I would treat this as a direction of travel rather than something every business needs to implement tomorrow, but you have to admit the direction is interesting.

OpenAI was making a related argument from a different angle this week, publicly advocating mandatory capability-based safety requirements for advanced AI systems, including independent assessments, cybersecurity requirements and incident reporting.

The common thread is that capability without governance is becoming unacceptable architecture.

5. MCP is becoming plumbing, not a novelty

A less glamorous but useful development this week is the continued packaging of marketing-data systems behind Model Context Protocol. For example, Catchr launched an MCP layer on 8 September connecting AI assistants to more than 100 marketing-data sources and reporting systems.

I find the signal is more important than the vendor, as we move from Export CSV → upload to ChatGPT → analyze toward: AI → authorized query → live marketing system → analysis.

This makes the source architecture of your – and my – AI workflow increasingly important. We need to define sources by authority.

For example:

Revenue
CRM > GA4

Website behavior
GA4 + Clarity

Search visibility
Search Console

Business/context changes
Notion project log

Paid-media spend
ad platform

Then explicitly teach the workflow what happens when sources disagree.

What I would actually do this week

The most immediate action is to annotate 8 September in reporting for relevant European clients and watch organic behaviour with the Search environment in mind.

I don’t want to reach the end of September, see a change in traffic and then attempt to reconstruct whether Google changed the route into the website halfway through the month.

Second, I am deliberately not adding another AEO tool yet. I have been testing 2 different tools, and the category is moving quickly towards automated recommendations, but I still think the difficult part is the evidence layer.

My differentiation shouldn’t be how many prompts I can monitor. It should be how I determine whether a visibility problem actually matters.

Third, I want to formalize source authority and logging inside the AI workflow pilot. We’ve spent several weeks thinking about context, skills, tools and permissions.

The next useful artifact is a small specification that defines the source, its authority and allowed actions, the required evidence, the log and lastly, human approval.

That starts turning the workflow from something useful into something trustworthy. And this week has reinforced one wider analytical principle I’m going to keep testing across SEO, CRO, analytics and AEO:

A metric is an observation produced by a system. Before deciding why that metric changed, I need to understand whether the system producing it changed first.

Google SERPs change.

AI interfaces change.

Attribution changes.

Consent changes.

Competitors change.

Agents change what they can access and what they can do.

The metric moves.

Then we tell ourselves a story about why.

I can’t help but think that the useful discipline is to resist that story for a little longer and ask: What changed around the customer?