This Week in SEO: Google Rewired Europe, ChatGPT Shopping Swapped Engines, And The Entry Ticket Just Changed Hands

Hi everyone, this is Neo.

Start with two numbers: 8.26% → 61.54%.

That’s not a ranking swing. It’s the share of ChatGPT Shopping product recommendations classified as coming from a merchant’s product feed, and how fast it crossed that line on July 10. In the same dataset, another number moved too: the count of distinct merchants being referenced dropped from 13,524 to 10,607 — down more than a fifth in a month.

Then last week, Google quietly shipped a new commercial layout in Europe. Comparison platforms got their own dedicated unit. Direct suppliers got one too — but theirs only appears when the comparison platform’s unit shows up.

And one more, which looks almost trivial: Google Business Profile post view counts are back, three and a half years after Google killed them.

Read separately, these are just industry news. Read together, here’s my take: in the second half of 2026, whether users ever see you is shifting from “where do you rank” to “do you hold an entry ticket.” Ranking is an outcome. Access is a qualification.

Today I’m going to connect this week’s threads, then hand you a checklist you can act on.

1. This week in one table

Story Key facts What it means for your site
Google launches aggregator + supplier units in the EEA Search Central documented both on Sept 8; live for EEA users Aggregators get the premium box; merchants get an attached slot — and have to send data
ChatGPT Shopping’s retrieval source flipped Feed-sourced recommendations went 8.26% → 61.54% on July 10 Product feeds moved from nice-to-have to admission requirement
Merchant concentration jumped Top 10 stores went 22.5% → 41.8%; distinct merchants -20% Narrower doors mean a wider gap between in and out
Business Profile post views return Rolling 18 months, Search + Maps combined, not in the API yet Content ops finally get feedback — but only “was seen,” not “was clicked”
A UK site lost 98% of news visibility after switching to .com Visibility fell from 1–2% to 0.02% in two weeks; no recovery in four months Migrations are a months-long cost, not a change of address

Three of these five are the same story: get in first, then rank.

2. Europe: Google invited comparison platforms into the results, and merchants got an attached slot

The biggest item first.

Google launched two new commercial units across the European Economic Area (EEA), documented on Google Search Central on September 8:

  • Aggregator unit — built for Vertical Search Services: OTAs, comparison shopping services, metasearch engines, and directories. It currently covers hotels, flights, long-distance trains and buses, and products.
  • Supplier unit — built for the businesses themselves, like a specific hotel or airline.

The mechanics matter more than the headline:

  1. The aggregator unit is the private room. The top-ranked aggregator is expanded by default, with its listings, prices and ratings shown inline, and clicks inside the unit go straight to that aggregator’s site. Users can switch to another participating aggregator, but only one aggregator unit shows at a time.
  2. Aggregators have to send data to get in. To participate, you need approval as a Vertical Search Service and you must supply data through direct feed integrations or real-time APIs. This is not a “build a good page” problem. It’s an engineering and partnerships problem.
  3. The supplier unit is attached. Direct suppliers don’t need to submit anything beyond what Google can crawl (feeds can improve results), but the unit only appears alongside the aggregator unit.

That last line is the one I keep coming back to. In these European commercial queries, a business’s own placement now depends on whether comparison platforms show up at all. Part of your fate sits with the aggregator.

What does Google say about its own decision?

Reuters reported the rollout the same day. Google’s SVP of knowledge and information, Nick Fox, said the changes “degrade the user experience for Europeans — boosting online intermediaries at the expense of local businesses.” Google told Reuters this is the largest reduction in Search quality in its 29-year history.

The pressure behind it is regulatory. On July 23, the European Commission fined Google €460 million for favoring its own shopping, hotel, transport and sports results over third-party services — part of an €890 million package — and gave it 60 days to comply. That window runs into late September. If the Commission decides the design falls short, periodic penalties of up to 5% of global turnover stay on the table.

Also worth writing down: Google says earlier DMA compliance changes already cut free direct booking traffic to European businesses by 30%. So this rollout lands on top of a wound that was already bleeding.

And it’s likely to spread. A Google Actions Center page says dining, services, and things-to-do searches are being updated with the same two units. No launch date, and Search Central doesn’t list local queries for either unit yet.

Neo’s take

If you sell into Europe, there are three layers of practical impact:

One: your data now forks by region. The same query returns a different layout in EEA countries than in the US or UK. Over the next few weeks, expect “unexplainable” declines in your rank tracker and in country-filtered Search Console data. That may not be a performance problem at all — it’s a layout change. Don’t do aggressive content or structural surgery on your European pages right now. First figure out whether the layout moved or you did.

Two: product queries are in scope. The docs list products explicitly. Ecommerce sites targeting Europe should check whether an aggregator unit now appears for their money keywords. If it does, your competition is no longer a peer site — it’s a comparison platform plus a peer site.

Three, and this is the opportunity: aggregators need data, so feed them. The aggregator unit is populated by feeds or APIs. That means you can plug in as a supply side. Get your inventory and pricing into a comparison platform and you have a shot at the unit that’s expanded by default. Most people will consider this too much hassle. That’s exactly why it’s a moat.

One more note: back in March I wrote about Google’s big EU search changes when they were still in testing, starting with lodging queries. Now they’re live and documented. This isn’t finished either — watch that late-September compliance window.

3. ChatGPT Shopping: feeds went from 8% to 62%, and concentration followed

Second item, and the most practical one for site owners this week.

Profound published a shopping dataset in August (SEJ covered it on September 9). The headline facts:

  • The share of product recommendations it classifies as feed-integrated jumped from 8.26% to 61.54% on the single day of July 10.
  • That analysis covers 1,757,723 tracked prompt runs in July.
  • By September 3, feed retrieval accounted for roughly 65% of the product recommendations it tracks — meaning feed retrieval overtook web search retrieval in August.

Now the uncomfortable part. Across 687 customers in that sample:

  • 450 saw Shopping visibility fall by at least a third (comparing July 7–9 with July 10–12);
  • 67 gained at least a third;
  • A statistical model built on lost web-search retrieval and gained feed retrieval explained 83% of the variation in those visibility changes. In other words, that two-day swing was mostly this source flip.

And here’s the number that says “narrow doors”:

Metric Before the flip After the flip
Share of references going to the top 10 stores 22.5% 41.8%
Distinct merchants referenced 13,524 10,607

Feed retrieval pulls from a narrower pool of sources, so fewer merchants get referenced and the head gets more concentrated. For smaller sellers, that matters far more than the “62%” headline: the tickets are being handed to a subset.

The prime suspect on timing is clear. OpenAI released GPT-5.6 on July 9, saying the rollout would finish within 24 hours, and the report ties most of the July changes to that release. OpenAI hasn’t confirmed the connection, and its ChatGPT release notes contain no shopping updates on July 9 or 10. So this is a strongly correlated, officially unclaimed conclusion — I’m keeping that discount in the text on purpose.

What OpenAI’s own pages say:

  • ChatGPT selects products using structured product data like price and description, sourced from data providers and from stores themselves;
  • OpenAI states product results are chosen independently and are “not ads, nor influenced by any OpenAI partnerships”;
  • Shopify stores are already connected via Shopify Catalog with no extra steps; Etsy catalogs are connected too;
  • Other retailers can request direct feed access, but OpenAI’s merchant page says applicants are on a waitlist;
  • Feeds run through the Agentic Commerce Protocol (extended to product discovery on March 24), and can also come from supported providers such as Salesforce and Stripe;
  • Shopping is currently live for US users only, with OpenAI saying it will expand and that a self-serve feed platform is planned for later this year.

Now here’s the counterintuitive datapoint you should not skip.

An earlier Profound breakdown (~1 million product offers, ~201,000 prompt runs, June 18–25) found:

  • Of citations derived from direct product feeds, about 99.9% appeared in the first offer position — feeds are the strongest lever for the top slot;
  • Yet 88.29% of all offer instances still came from web PDPs, and even among merchants already feed-integrated, the figure stayed at 75.81%.

The conclusion isn’t “PDPs are dead.” It’s this: feeds decide whether you get the first slot; PDPs decide how many times you show up at all. Both matter, and they’re different jobs.

Neo’s take

I’ve never liked framing GEO or AI search as mysticism, because the moment data shows up it becomes an engineering problem. And this dataset describes a very familiar commercial structure: the platform standardizes the supply side, then rewards whoever standardized first.

For independent site owners:

1) If you’re on Shopify, you’re already in the room — but only in the room. Default integration is not default victory. The quality of the structured fields in that feed — pricing, availability, attributes, variants — is what started separating winners in July. Profound predicts the next round of competition shifts from whether you have a feed to which fields your feed carries.

2) If you’re not on Shopify (custom builds, WooCommerce, BigCommerce), pin down your path now: direct application (waitlist), a supported feed provider like Salesforce or Stripe, or OpenAI’s self-serve platform. Don’t wait for that platform to launch. Field specs and data cleanup can start today and depend on nobody’s approval.

3) Treat feed-to-PDP consistency as a project. If price, stock, title or attributes disagree between the two, you’re handing the machine two contradictory resumes. I made the same point in the piece about conflicting brand information: the problem isn’t that AI lacks information about you — it’s that your information contradicts itself.

4) This concentration jump is an early warning. Top-10 share going from 22.5% to 41.8% in a month doesn’t mean un-integrated merchants got slightly less traffic. It means they largely disappeared from recommendations. 450 stores losing a third or more isn’t a gradual shift. It’s a cliff.

4. An old friend that came back — sort of: Business Profile post views

This one is small, but any site with a local footprint should pay attention.

In September’s Small Business Bulletin (posted to the Business Profile Help community by Google’s Lisa Landsman), Google confirmed that post view counts are rolling out again globally.

The details:

  • Counts show on each post card in the dashboard, covering the past 18 months on a rolling basis;
  • They combine views from Search and Maps;
  • They are not in the API yet;
  • The previous post metrics were retired on February 20, 2023, leaving a three-and-a-half-year blind spot.

One warning: this is views, not clicks — and Google retired clicks.

So the number answers exactly one question: was the post seen. It cannot tell you whether it drove action. If your team adopts it as a KPI, you’ll get a metric game where people post for impressions.

My suggestion: pair post views with directional signals — profile-page search volume after a post, call clicks, direction requests. Use views as a visibility check-up and the rest as an intent check-up.

5. What Mueller actually said: cleanup and proving value are two different jobs

Google’s John Mueller answered a familiar programmatic-page problem this week. Someone had auto-generated pages from combinations of domains, technologies and attributes, then cleaned them up — and the site still wasn’t recovering.

His reply, in essence: Google’s systems had “possibly lost faith in your site providing good value to users.”

I read that line twice, because it separates two tasks people constantly merge:

  • Cleaning up pages — deleting, merging, updating low-value content. Technical work you control.
  • Re-proving value — getting the system to believe again that your site solves real problems. That comes from other people (users, citations, clicks, dwell), not from you, and it accrues slowly.

So the answer to “why hasn’t my site recovered after I deleted everything” is: because these are two jobs and you finished one. Recovery timelines are measured in months, and the clock starts when your core pages begin generating real usage signals again.

For programmatic SEO sellers, the practical read is blunt: the window for combination-generated pages is far narrower than it was two years ago. If you still do it, do the version with a real data source, real search intent, and genuine differentiation between pages — not a template with swapped variables.

6. A UK site lost 98% of its news visibility after moving to .com

Last item, and the only complete post-mortem this week.

NewzDash founder John Shehata shared an anonymized case on LinkedIn: a UK news site moved from .co.uk to .com — they wanted one global brand domain, and they wanted ad systems to stop classifying them as a UK-only site. The result:

  • For the 14 months before the move, it held a steady 1–2% Search Visibility on the news queries NewzDash tracks (comparable sites usually sit at 3–6%);
  • Within two weeks of the move, visibility fell to 0.02%;
  • Four months later it still hadn’t recovered, with the new domain showing only isolated appearances.

Some necessary caveats, because I don’t want this turning into “migrating a site kills it”:

  • The site was already sliding through several core updates and had internal technical problems, so the data can’t prove the migration was the only cause;
  • The 98% figure is NewzDash’s own Search Visibility score, which tracks trending news queries (Top Stories plus organic), estimates click share, and predicts clicks rather than actual traffic. Standard SEO tools typically estimate a 60–70% drop in this situation;
  • The case doesn’t name the site or publish the data, so it can’t be independently verified.

What’s genuinely useful is Shehata’s explanation next to Google’s own documentation. He coined the term “Google Reset”: drastic changes like switching CMS, redesigning, or changing domains trigger Google to reevaluate the site entirely. His core argument is that a 301 passes link equity but “may not carry that history” — and news surfaces weigh hostname-level signals like publication history and editorial trust.

Google’s official site-move guide, meanwhile, advises: map old URLs to new ones, set up redirects, submit Change of Address in Search Console, and monitor. For large sites it recommends testing one section first, and it advises against combining a domain move with a CMS change or redesign.

Line the two up and this case hit both of Google’s recommendations: a full cutover, done overnight, with ad-system and branding motives tangled into the decision.

My advice: if you’re weighing a move to a single .com, budget the recovery in months, migrate section by section, and freeze other big changes during the move. And tape Mueller’s June quote somewhere visible: “A site move is always a big deal, even if you take the time to do it right. It’s a lot of work, and the outcome is impossible to know fully ahead of time.”

7. Connecting it: the tickets are changing hands

Look back at all five stories and you’ll see three faces of one thing:

Situation The old qualification The new qualification Who hands out the ticket
EEA commercial queries Good pages, strong rankings Approval as an aggregator / supplying feed or API data Google (approval) + aggregators (data)
ChatGPT Shopping PDP optimization + organic traffic Product feed integration (default on Shopify; waitlist or provider otherwise) OpenAI + feed providers
Local visibility via GBP Whether posts got clicks Whether posts got seen — and only views Google (Search + Maps combined)
Site recovery / migration Clean up pages, wait for the rebound Re-prove value + rebuild hostname history User behavior and algorithmic confidence

Read that table and you arrive at a counterintuitive conclusion: “content is king” was never wrong, but content now only decides whether you deserve to be shown. Access decides whether you get a chance to be shown. Those used to be one pipeline (good content → crawl → index → rank). Now they’ve split into two.

Practically, that means the people on your team who understand data pipes and the people who understand content need to start sitting at the same table.

8. Seven things you can start this week

Don’t just read the news. Do this:

  1. Map your product feed path. Shopify (already integrated), Etsy (connected), or everyone else (waitlist or a supported provider like Salesforce, Stripe). Write down where you stand, and if you’re unsure, check OpenAI’s merchant page directly.
  2. Pull a feed field completeness report. Title, price, availability, variants, attributes, images — note every gap. The next round is about field quality, not mere presence.
  3. Run a feed-vs-PDP consistency audit. Sample 20 SKUs. Compare price, stock, title and key attributes. Any mismatch is a bug.
  4. Build a separate Search Console view for Europe. Isolate EEA countries and watch hotel, flight and product queries. If numbers dip, check for a layout change before touching content.
  5. Pick one or two vertical comparison platforms and evaluate integration. Aggregator units need feeds or APIs — right now that’s an opening on the supply side.
  6. Redefine your GBP post KPIs. Views as a visibility check-up, plus profile search, call clicks and direction requests as intent signals.
  7. If you’re planning a migration or redesign — go section by section, do it alone, freeze everything else, and budget 3–6 months of recovery.

9. Wrapping up

Five stories this week: a European redesign, ChatGPT swapping engines, GBP adding a number back, Mueller sounding a warning, and a UK site losing 98%.

Underneath all of them is one sentence: platforms are standardizing how the supply side plugs in, and the standardized tickets go to whoever prepared first.

I’ve been building and running independent sites for a long time, and I’ve watched far too many “I wish I’d started earlier” moments. Feed work costs you a few hundred hours of field cleanup today. Next year, it could cost you your entire presence in AI shopping recommendations.

So my one piece of advice this week: check whether you’re holding a ticket. If you’re not, get in line now.

This is Neo. See you in the next one.