Heading into the weekend, an AI Max study found nearly three in ten exact match impressions are already algorithmic broadening, right as Google prepares to switch the feature on by default. Stack Overflow’s new question volume finished July down 99.3 percent from its 2014 peak, and a donor database logged 117,000 daily requests from an Amazon crawler that sends nothing back.
Exact Match Erodes: Google Widens Its Own Discretion
AI Max already reclassifies nearly three in ten exact match impressions as algorithmic broadening, and Google is applying the same quiet-discretion logic to ad delivery across every surface it sells.
- AI Max Already Broadens 29% of Exact Match Traffic, Smec Finds Smec’s Mike Ryan pulled 383 million ecommerce Search impressions from AI Max accounts and found nearly 29 percent of what shows up in exact match reporting is now algorithmic broadening, up from zero at the start of last year. The finding lands three days after Google confirmed it will switch AI Max on automatically for eligible Search campaigns starting September 1, with no advertiser opt-in required.
- Google Broadens Limited Ad Serving Across Every Ad Surface Google is extending Limited Ad Serving beyond its earlier scope to every Ads product, rolling out in stages through 2028. An account judged unqualified simply gets fewer impressions with no disapproval notice and no policy strike attached, so a throttled account and a slow week look identical on the dashboard.
The Question Well: Drying Up as the Answers Take Over
Stack Overflow’s new question volume has collapsed 99.3 percent from its 2014 peak, and Google’s own former chief scientist says what matters now is not the model but what feeds it, a system built to compensate for exactly that kind of disappearing supply.
- Stack Overflow’s July Questions Fall 99.3% From 2014 Peak Stack Overflow logged just 1,442 new questions in July 2026, against a March 2014 peak of 207,204 a month, according to the site’s own Stack Exchange Data Explorer. The human written questions and answers that trained the models now answering those same questions instead have largely stopped being created.
- Jeff Dean Says Context Engineering Matters More Than Model Choice Jeff Dean, who left Google’s chief scientist role after 27 years, told Y Combinator partner Diana Hu that AI progress now comes less from bigger models and more from the retrieval, tools and memory built around them, a combination he groups under context engineering. He did not address how a system chooses which outside sources qualify for that retrieval in the first place.
Extraction Without Return: Crawlers, Monopolies, Same Logic
One donor database operator clocked Amazon’s crawler at 117,000 requests a day with zero referral traffic in return, and a new amicus brief argues that same extraction logic, left unchecked, is what built Google’s AI lead in the first place.
- Donor Database Blocks Amazon Crawler After 117,000 Daily Requests PatronView founder Nick Gray blocked Amazon’s Amzn-SearchBot at his firewall on August 5 after it reached roughly 117,000 requests a day against his 1.5 million page site, with bots making up over 99 percent of his traffic. Gray says he would rather meter and sell access to the crawler than block it, but no market mechanism for that exists yet.
- Amicus Brief Says Google’s AI Lead Came From Its Search Monopoly A 34 page amicus brief filed August 4 in the D.C. Circuit by Joel L. Thayer and Asad Ramzanali argues generative AI competition cannot discipline Google’s search monopoly, tracing Google’s AI buildout back to resources its search dominance funded. It is an argument from outside parties, not a ruling, filed in the consolidated appeal of Judge Amit Mehta’s 2024 search monopoly decision.
ChatGPT Ads Grows Up: From Measurement to Full Auction
OpenAI spent the summer catching ChatGPT Ads up on measurement. This wave builds the auction itself.
- ChatGPT Ads Adds Conversion Bidding and Product Carousels OpenAI added a beta conversion optimized bidding mode for ChatGPT Ads product feed campaigns, began testing a carousel that shows several products in one ad, added dynamic URL parameters for attribution, and confirmed Brazil and Mexico get the ad platform this week.
Today’s Quick Hits
- Googlebot Also Sends PATCH, DELETE, PUT, OPTIONS, HEAD Calls Gary Illyes confirmed Google’s crawlers issue HEAD, OPTIONS, PUT, PATCH and DELETE requests beyond standard GET and POST calls, combining for under 1.5 percent of all crawler requests. He attributed the pattern to JavaScript firing those calls during rendering, not scraping abuse.
- Google Ads Tests “Strongest Match” and “Strong Match” Badges Google Ads is live testing Strongest match and Strong match badges on search ads, the first sighting since Google pre-announced them in mid June. The labeling draws on signals already used in ranking, so nothing changes in the auction, and there is no advertiser reporting tied to it yet.
- Microsoft Ads Adds Bulk Editing for Disapproved Assets Microsoft Advertising now lets advertisers bulk edit and appeal up to 5,000 disapproved text and image assets at once, ending the need to challenge each rejection one at a time.
- Google Ads Adds Pure Measurement for New Customers A new Report on new customers acquired setting adds New Customers and New Customer Value columns as pure measurement with no bidding impact, replacing the one cent bid modifier workaround advertisers used to fake the split.