Confirmed
Google publishes aggregator and supplier unit documentation for its EEA search redesign
Google Search Central documentation updated September 8 formally defines two new search-result components live only in the European Economic Area: an aggregator unit, a dedicated section for price-comparison and travel sites where the top-ranked listing expands by default with photos and pricing, and a supplier unit, which surfaces direct businesses like individual hotels and airlines only when an aggregator unit is present. Both apply to hotel, flight, ground-transport, and product queries under the EU's Digital Markets Act.
In plain termsAn aggregator unit works like a price-comparison shelf built into the results page itself, showing the best-ranked comparison site's own picks, while a supplier unit is a separate shelf reserved for the individual hotel or airline being compared, so a hotel's own site and a booking site can now both show up as different named shelves for the same search instead of one blended list.
Why it mattersAny EEA-facing travel or e-commerce site now competes for placement inside two newly named, separately governed result types instead of one generic organic listing.
Our takeNaming and documenting these units, rather than leaving them as an undocumented test, means Google now treats aggregator and supplier placement as a stable feature EEA sites can build around instead of a moving target. Watch whether Google extends the same two-unit structure outside the EEA once the DMA-mandated version has run for a full reporting cycle.
Confirmed
Common Crawl finds most llms.txt files are templated boilerplate, and none can actually block a crawler
Common Crawl's analysis of 584,107 llms.txt files from its July 2026 archive found 68% came from website templates or plugins, Wix alone accounting for 41%, and 22% contained no links at all. Roughly 10% of the URLs returning content at the llms.txt path were actually a site's robots.txt file misplaced there. Of 32 files that appeared to block Common Crawl's bot, none of those sites also blocked it in robots.txt, the file that actually enforces anything.
In plain termsllms.txt is a proposed plain-text file websites can publish to describe themselves to AI crawlers, similar in spirit to robots.txt, except robots.txt is the one file crawlers are actually built to obey. This study found most llms.txt files are unedited template output, and some site owners appear to believe writing 'no' in it carries the same enforcement power as robots.txt, when a crawler that already respects robots.txt was never reading llms.txt for permission in the first place.
Why it mattersFor any site treating an llms.txt file as a real access control, Common Crawl's data shows it grants nothing and blocks nothing, so robots.txt remains the only lever that works.
Our takeI measured in 'Instruction files never shrink' that AI agent configuration files grow because deleting an old rule risks a silent regression, so those files balloon over time rather than staying lean. Common Crawl's llms.txt numbers show close to the opposite failure: most of these files were never edited past their template defaults in the first place, so the risk here was never over-accumulation, it was that nobody treated the file as worth maintaining at all.
Confirmed
Shopify lets a single product variant carry up to 20 barcodes
Shopify's changelog confirms that as of September 8, product variants can now store up to 20 barcodes instead of one, covering UPC, EAN, ISBN, GTIN, ASIN, or a custom type, each format-validated on entry. Merchants can add, scan, and remove barcodes directly from the variant details page, and products become searchable by any barcode stored on a variant, not only the first one entered. The first barcode still feeds channels and integrations by default.
In plain termsA barcode field used to be one slot per variant, so a product sold under three different scannable codes, like its own UPC, a retailer's private-label EAN, and Amazon's ASIN, meant choosing which one to store in Shopify and keeping the other two somewhere else entirely. Now all three can live on the same variant record.
Why it mattersA single SKU arriving with a manufacturer UPC, a private-label EAN, and a marketplace ASIN can now carry all three natively instead of forcing a merchant to pick one and track the rest in a spreadsheet.
Our takeThis is a data-model fix arriving after multi-marketplace selling had already made a single-barcode field obsolete in practice, so the interesting question is whether Shopify's own product feed export and the Google Merchant Center integration pick up all 20 values or still forward only the default. A feed export that still collapses a 20-barcode variant record down to one field would leave the real gap exactly where it was.
Confirmed
Shopify adds Meta as an AI channel with default catalog sharing and direct checkout
Shopify's changelog confirms Meta now joins its Agentic Storefronts lineup inside admin, with a merchant's products shared to Meta by default through Shopify Catalog and no manual sync required. Merchants get separate access controls for catalog sharing and for direct checkout, plus a performance view scoped to Meta inside the agentic settings panel. The changelog describes the integration as supporting direct checkout functionality on Meta's platform for customers discovered through Meta channels.
In plain termsShopify Catalog is the single product database Shopify already uses to feed a merchant's storefront, and turning Meta into an 'AI channel' means that same database now also feeds Meta's AI shopping features automatically, the same default-on relationship the Spring '26 Catalog changes set up for other channels.
Why it mattersDefault-on catalog sharing means a merchant's product data reaches Meta's AI shopping surfaces with no opt-in step, so the access controls, not the sharing itself, are now the setting worth checking first.
Our takeI wrote in 'Shopify's catalog landgrab' that the Spring '26 Edition repositioned Catalog as the substrate underneath agentic commerce, letting even non-Shopify merchants sync in and sell through AI channels. Meta becoming a default-on AI channel is that positioning made literal, catalog sharing is on unless a merchant turns it off, which is the opposite of the deliberate build-versus-syndicate call that piece described.
Observed
SEO tool AlsoAsked reports nearly all People Also Ask answers are now AI-generated
AlsoAsked's Mark Williams-Cook reports that across a sample of 19.2 million English-language search queries, 97% of People Also Ask answers in the first week of September carried an AI-generated overview, up from 86% in August. Google has not confirmed a policy change behind the shift, and the data comes from AlsoAsked's own query-tracking tool rather than a Google-published figure. The trajectory, if it holds, points toward AI Overviews covering close to every PAA slot within weeks.
In plain termsPeople Also Ask is the expandable list of related questions inside Google's regular results, and this data says that when a searcher taps one open, they almost never see a plain text snippet anymore. They see Google's own AI-written answer instead, the same kind of box that already sits above the main results.
Why it mattersIf nearly every People Also Ask expansion now surfaces an AI-written answer instead of a linked snippet, that entire section stops functioning as a click source and starts functioning as one more AI Overview surface to track for citation, not traffic.
Our takeA jump from 86% to 97% in a single month is fast enough to confirm against a second independent tracker before treating it as a settled floor rather than a snapshot of one aggressive week. If the rate does hold near 100%, the practical shift is that PAA becomes a pure citation-visibility surface, worth the same monitoring as an AI Overview and not worth writing content aimed at winning the old snippet position.