Confirmed
300 French newspapers file an antitrust complaint over Google's AI Overviews
The APIG federation of nearly 300 French newspapers filed a complaint with France's competition authority on August 11, alleging Google violated a 2022 compensation agreement by launching AI Overviews in France in late July without publisher consent. APIG president Marc Feuillee said publishers want the value of their content shared, not innovation halted. The same authority already fined Google 250 million euros in 2024 for breaching parts of that deal.
In plain termsAPIG is a French trade association representing daily newspapers, and the 2022 compensation agreement is a deal where Google agreed to pay French publishers for using snippets of their content, the specific commitment this complaint says Google just broke.
Why it mattersA formal regulatory complaint over AI Overviews' effect on referral traffic gives SEO and publisher-side practitioners a live case study in how measurable a traffic decline needs to be before regulators act.
Our takeI argued in Google's billions-of-clicks claim that Google's own AI-traffic numbers can't be checked against any individual site's own data, and this complaint is publishers trying to force exactly that kind of accounting through a regulator instead. APIG's 2022 deal gives France a specific, checkable commitment to point to, which is firmer ground to stand on than Cloudflare or Wikipedia's aggregate traffic data offers publishers anywhere else.
Deep dive: Google's billions-of-clicks claim: why it can't be checked against your own site data
Confirmed
OpenAI expands ChatGPT ad tests to the UK, Mexico, Brazil, Japan, and South Korea
OpenAI said August 11 that ChatGPT Ads, running as a test since February for logged-in US Free and Go tier users, has now launched in five more countries: the UK, Mexico, Brazil, Japan, and South Korea. Ads are matched to conversation topic and chat history, always labeled sponsored, and excluded from accounts flagged as under 18 or near sensitive topics like health and politics. Paid tiers stay ad-free, and OpenAI says the ads never change ChatGPT's answers.
In plain termsThe Free and Go tiers are ChatGPT's no-cost and low-cost subscription levels, the ones OpenAI is testing ads on first, while paid Plus and Pro subscribers keep an ad-free experience.
Why it mattersAn AI answer surface used by hundreds of millions of people now carries paid placements matched to conversation context, a competitive surface practitioners tracking AI visibility need to watch alongside organic citations.
Our takeFive new markets in six months is a fast expansion for a test OpenAI still frames as learning-phase, and the real signal to watch is whether dismissal rates and trust metrics hold as the free-tier population scales past English-speaking markets. Ads matched to chat history rather than a single query are a different targeting model than search ads, and how well that holds up outside the US pilot is still unproven.
Observed
OpenAI's ChatGPT search index treats sites with no licensing deal the same as its partners, new data finds
French SEO consultancy Resoneo read 1,249 ChatGPT answers in July and found no difference in format, length, or freshness between how OpenAI's in-house search index, internally named labrador, served licensed partner sites versus sites with no OpenAI content deal. The finding backs a correction SEO researcher Suganthan Mohanadasan published in July, after he initially misread the same index as a closed allowlist of major publishers. Resoneo also found the index typically preserves a cited page's H1 heading in the snippet it stores.
In plain termsLabrador is the nickname researchers gave OpenAI's own search index, separate from a live Google scrape, and a licensing deal is a paid content-sharing agreement some publishers, but not most sites, have signed with OpenAI.
Why it mattersA site's absence from OpenAI's publisher-deal list does not appear to gate whether ChatGPT's free-tier answers can cite it, so the practical lever for AI visibility here is page structure, not licensing status.
Our takeI covered this same fragility in ChatGPT's runner-up layer: OpenAI rewrote the exact server-side field a public teardown had just documented, eight days after it published, on July 21 and 22. Resoneo's data traces back to that same tagging field, the one both investigations depended on before OpenAI pulled it, so this finding is durable only until OpenAI's next unannounced edit.
Deep dive: ChatGPT's source selection lives in an editable prompt OpenAI can rewrite overnight
Observed
Google's AI Overviews cite a single self-promotional listicle as the source for an entire local business list, SEO consultant finds
Local SEO consultant Joy Hawkins posted screenshots on X showing a Google AI Overview local result citing one lawyer's self-promotional listicle as the primary source for every law firm in a list, linking to that listicle instead of the individual firms' own websites. Search Engine Roundtable's Barry Schwartz reproduced the behavior across several browsers on August 12. Google generates these AI summaries for local-pack results only intermittently, and there is no indication Google has acknowledged the pattern.
In plain termsA listicle is a list-format article, often written to promote its own author, and being the AI Overview's cited source here means Google links to that list instead of to the actual local businesses it describes.
Why it mattersA local business with a fully optimized site can still lose its own link in an AI Overview to a third-party listicle that happens to mention it, a citation-source risk local SEO work doesn't yet have a lever to fix.
Our takeGoogle generating these summaries only intermittently, and for some queries but not others, makes the failure mode hard to monitor systematically, since a business can't tell in advance when it's about to lose its own citation to a listicle. Watch whether Google treats this as a ranking-quality bug worth fixing, given the gap between a stated goal of helpful AI answers and quoting self-promotional content as the definitive source.
Confirmed
Google Merchant Center splits YouTube affiliate traffic out of organic reporting and adds a Network dimension, starting August 24
Google will change four things in Merchant Center performance reports starting August 24. YouTube affiliate traffic, where creators earn commissions for featuring a merchant's products, moves out of the Organic value into its own category, with matching updates to organic click and impression definitions for YouTube. Ads product-level reporting expands to cover Performance Max, Video, App, and Demand Gen formats, and a new Network dimension lets merchants compare performance across Google's ad networks.
In plain termsThe Organic reporting value in Merchant Center counts free product listings that show up without an ad campaign behind them, and this change carves YouTube's creator-driven affiliate sales out of that count so they stop inflating the organic number.
Why it mattersMerchants who rely on Merchant Center's organic-versus-paid split to judge free-listing performance need to re-baseline that comparison after August 24, since YouTube affiliate clicks that used to count as organic traffic move into their own bucket.
Our takeSplitting YouTube affiliate clicks out of the organic bucket is a genuine measurement fix, but it also means any merchant benchmarking this month's organic traffic against last month's needs to know the retroactive July 1 baseline before trusting the comparison. Watch for the same alignment logic to eventually extend to other free-listing surfaces Google has bundled into Organic without much definitional scrutiny.
Confirmed
Shopify ends Delivered Duty Unpaid support in Managed Markets, moving international checkout duties to the point of sale
Shopify will discontinue Delivered Duty Unpaid (DDU) support in Managed Markets on August 24, automatically converting any market currently using DDU to Delivered Duty Paid (DDP) wherever Managed Markets supports it. Under DDP, international customers pay duties and taxes at checkout rather than facing a surprise charge on delivery. Merchants who want to keep the pay-on-delivery model must disable Shopify Managed Markets before the deadline; everyone else needs to take no action.
In plain termsDelivered Duty Unpaid means the shopper pays customs duties when the package arrives, often as a surprise fee, while Delivered Duty Paid bakes that cost into the price shown at checkout so there's no charge later.
Why it mattersMerchants running Managed Markets who haven't opted out will see their international checkout pricing and customer duty experience change automatically on August 24, whether or not they've reviewed the DDP conversion.
Our takeThis closes the gap between Shopify's July 2026 change baking duties into displayed prices and DDU, the checkout-time-only version of that idea that still relied on charging the shopper later. Any merchant who wants to keep charging duty on delivery needs to actively disable Managed Markets before August 24, since the default here is an automatic, no-action-required switch to duty-paid pricing.
Confirmed
Google wires Looker's governed semantic layer into Gemini Enterprise so its AI agent runs the same metric logic every dashboard user gets
Google integrated Looker's governed semantic layer, business-approved metric definitions, table relationships, and row- and column-level access controls, directly into Gemini Enterprise, its conversational AI platform for enterprise data. Instead of letting the AI model guess at what a metric like Revenue means, Gemini Enterprise now generates deterministic SQL from Looker's existing, version-controlled definitions and enforces Looker's own permissions. The two systems connect over Google's Agent-to-Agent protocol, and Looker data is queried live rather than copied into Gemini Enterprise.
In plain termsA semantic layer is a rulebook that defines exactly how a business calculates a metric like Revenue, so different tools querying the same underlying data always agree on the answer instead of each guessing independently.
Why it mattersTeams already governing their metrics in Looker get that same governance automatically enforced on any AI-generated query, closing a gap where a conversational AI layer could otherwise return a different number for Revenue than the dashboard everyone already trusts.
Our takeGenerating deterministic SQL from a codified semantic layer is Google's answer to the most common AI-hallucination complaint in enterprise BI, an agent inventing its own definition of a metric that already has an official one. Whether this holds up depends on how many organizations have a Looker semantic layer clean enough to hand to an agent today, since the integration only governs queries that pass through Looker's existing rules.
Confirmed
Amazon OpenSearch Service ships GPU-accelerated vector indexing, builds a billion-vector index in under five hours
AWS made GPU-accelerated vector index building generally available on Amazon OpenSearch Service and OpenSearch Serverless, using NVIDIA's open-source cuVS library to offload the compute-heavy part of building a k-NN index to GPU workers. In a published benchmark, the service indexed one billion 1024-dimensional vectors in 274 minutes, with build time scaling linearly with data volume instead of growing disproportionately as datasets scale. The system routes each indexing job to GPU or CPU automatically, and customers pay only for active GPU build time.
In plain termsA vector index is the data structure that lets a system quickly find semantically similar items, the technology behind AI-powered search and recommendations, and building one for a billion items used to be slow and expensive enough that most teams avoided that scale.
Why it mattersTeams building retrieval-augmented AI search or recommendation systems on OpenSearch can now build billion-scale vector indexes in hours instead of days, without provisioning standing GPU infrastructure.
Our takeLinear scaling matters more than the headline billion-vector number, since it means the cost of a much bigger vector search deployment stays predictable instead of degrading as data grows. Automatic GPU-or-CPU routing based on segment size also means teams don't have to manually decide when a workload justifies GPU cost, usually the harder operational problem than indexing speed itself.