Our analysts track the weekly developments reshaping organic visibility and AI discovery. Brands building search programs can partner with vetted SEO agencies for strategic implementation.
Google admitted this week that its product got worse, and it wasn't the only company forced into an uncomfortable disclosure.
- Google admitted its EU redesign hurt search quality, the worst drop it has recorded in 29 years.
- ChatGPT runs its own search index called Labrador, confirmed through antitrust testimony and job postings.
- A new survey of 131 SEO professionals ranked relevance, backlinks, and content quality as Google's three most important factors, unchanged from a decade ago.
Here's the full story behind each development and what's still just background noise.
Our Take: Who Gets to Decide What 'Search Quality' Means?
Google says its product got worse. It's also the only party measuring whether this claim holds up.
The company's own developer documentation never mentions the DMA, but an SVP gave Reuters a direct quote blaming the EU regulation.
We'd argue that the ChatGPT index story is the sturdier record, since nobody had to ask OpenAI to confirm that Labrador exists.
Antitrust testimony and job postings surfaced it regardless of what OpenAI wanted to be made public.
A disclosure forced out of a company holds up better than one that a company volunteers.
Brussels Forces Google's Ugliest Redesign
Google rolled out redesigned search results across the EEA on September 8.
Hotel, flight, and restaurant queries are now split into a dedicated aggregator unit and a separate supplier unit for direct businesses.
The aggregator unit gives one vertical search service like Booking.com or Expedia the top slot with full details.
Two rivals sit below it with less information attached. Google stripped real-time pricing from the carousel below.
Nick Fox, Google's SVP of Knowledge & Information, told Reuters that the company is making the changes to comply with DMA requirements.
The redesign, he said, "degrades the user experience for Europeans," favoring online intermediaries at the expense of local businesses.
It also strips out features that people rely on every day.
Google said internal testing with millions of European users showed high dissatisfaction. Users mainly complained about having to retype queries.
The redesign follows an €890 million ($1.03 billion) DMA fine issued by the European Commission in July.
Of that total, €460 million ($534 million) targeted self-preferencing Google's services in Search. The company had 60 days to comply.
Google's own documentation frames this as a feature rollout. Reuters got the version where the search engine blames Brussels for the damage.
ChatGPT Hides a Search Index Called Labrador
ChatGPT runs a retrieval index, internally called Labrador, according to Peec AI's research.
An A/B test spotted in mid-August, "prefer-index-over-serp-v3," ran on 8% of chats, testing that index against traditional search results.
By September 2, five separate shopping experiments were running the same comparison.
The index has been in development for years.
Nick Turley, head of ChatGPT, testified during Google's antitrust trial that OpenAI struggled with unreliable search data from partners other than Google.
What OpenAI wanted was a direct deal with Google, but it refused.
ChatGPT then started sourcing Google's results through third-party scrapers instead.
A test on a site with zero organic traffic found Google Search Console logging a traffic spike from queries the AI ran on the exact days researchers checked.
Bing shows up, too, mainly in Deep Research sessions, confirmed by Microsoft's own Web IQ platform.
Google crawling a page and ChatGPT retrieving it are two different events now. The shopping experiments are where that gap is being tested live.
Zyppy's Survey Confirms a Decade-Old Ranking
Relevance, backlinks, and content quality topped a new survey of 131 SEO professionals, Zyppy reported, the same three factors that led a similar survey a decade ago.
The 2026 edition collected 13,665 ratings across 103 possible factors.
- Relevance was named a top-three factor by 57.1% of respondents
- Backlinks by 54.8%
- Content quality by 47.6%
Meta descriptions ranked dead last, with practitioners scoring them as having no meaningful ranking impact at all.
Links now do a second job, according to SEO consultant Andrew Shotland.
"A client recently got mentioned and linked from a Reuters article. Their visibility in ChatGPT went up 4x over the next week," he said.
A single mention now feeds Google's ranking system and the data language models pull from when someone asks for a recommendation.
The survey measures what practitioners believe about Google's algorithm, Bill Hartzer noted after reviewing the results.
He reads it as the profession's collective judgment on where to spend.
Antitrust testimony and the leaked Content Warehouse API material gave the industry names for behavior it had suspected for years, he wrote.
He sees a profession that has stopped hedging. Whether the algorithm itself changed is a separate question.
If the same three factors led a decade ago and lead now, the industry's progress shows up in its evidence.
Teams building an SEO content strategy today work from proof where they once worked from inference.
SEO Industry Insights
Old, low-value pages can keep hurting a site's reputation long after they're gone, according to John Mueller’s comments on Bluesky.
He was replying to a site owner who runs a directory that tracks technology stacks across a large number of domains.
(I'm sure some others will comment here with their thoughts & links to articles they wrote on these topics - take a look, review them with a grain of salt, and see what makes sense for your specific case.)
— John Mueller (@johnmu.com) September 7, 2026 at 2:13 PM
Mueller guessed the setup amounted to programmatic SEO, generating pages by combining domain names with site attributes like technology data.
Google's systems "have possibly lost faith in your site providing good value to users based on the old pages," adding that rebuilding trust takes real time and effort.
Separately, Cyrus Shepard tested what happens when a site opts out of Google's generative AI features entirely, using the Search Console control introduced last week.
AI impressions dropped almost immediately, confirming the toggle does what it claims.
Whether that counts as a win depends on which side of AI visibility a site wants to be on.
Here's exactly what happens to your AI impressions when you completely remove a website from appearing in Google's generative AI features
— Cyrus SEO (@CyrusShepard) September 5, 2026
It's damn near instant, too pic.twitter.com/PHHAP2l6PF
Some publishers report steady traffic since AI Overviews launched.
Le Monde's audience has held up since the feature reached France, CEO Louis Dreyfus said in an interview with INfluencia.
APIG, the French press alliance Le Monde belongs to, cites Arcom estimates of 33% to 38% traffic loss in European markets where the feature already ran.
Dreyfus credits the gap partly to apps, which generated 36% of Le Monde's traffic in August. AI Overviews never touch that channel.
The machinery that ranks Google's results could be the next thing to change.
Google DeepMind published research replacing the two-step system search engines have used for years with a single AI model trained to rank documents directly.
The research is early, tested so far on academic datasets alone. It's worth watching, though planning around it is premature.
What This Means for Anyone Managing Search Visibility
Each story above complicates the narrative that the industry already tells itself.
Untangling the three stories comes down to these three checks:
- Treat Google's "quality" claims as one input among several. A company describing its regulatory compliance has a motive to make the story sound as bad as possible.
- Audit whether ChatGPT cites a page from cache or a live fetch. A site's crawl access may matter less to ChatGPT's answers than assumed.
- Check app and direct traffic before blaming AI Overviews for a decline. Le Monde's experience shows the traffic mix can matter more than the feature itself.
These checks trade a company's framing for an independent record. Search teams that skip the swap are reading press releases dressed as data.
Last week's roundup ran into the same tension between what platforms report and what site data shows.
It covered Google's new AI Search Console metrics, the citation bug in Gemini 3.8 Flash, and Duane Forrester's substitution research.
If Google says its product got worse and there's no way to check the math, whose word is worth trusting here?
These leading SEO agencies help brands separate a platform's self-reported narrative from what actually shows up in their own data.
