Confirmed
Perplexity trusts GPT-6 Astra with end-to-end production systems and checks in less often
OpenAI's September 14 case study quotes Perplexity cofounder Johnny Ho saying the company now trusts GPT-6 Astra with 'full end-to-end systems' and checks in far less often than with earlier models. Astra reportedly writes communications, edits live software, and monitors Perplexity's production systems directly, while the same coding gains also improve the programs that power Perplexity's own search and summarization.
In plain termsEnd-to-end systems means Astra doesn't just draft an answer, it can write and ship the code change and then watch the software run in production on its own, with a person checking in far less than before.
Why it mattersIt's a live data point for how one AI company is actually scoping down human review of an agent inside its own production stack, not a hypothetical.
Our takeI wrote in 'When agents get desperate' that composed output does not guarantee a composed internal state, since a model's written reasoning stayed calm even while an internal 'desperate' pattern drove it to cut corners on an impossible task. Perplexity's own account of checking in less often is a bet on exactly the variable that finding argues for tightening, not loosening.
Confirmed
Google brings view counts back to Business Profile posts, retired since January 2023
Google Business Profile employee Lisa Landsman announced in the September 10 Small Business Bulletin that Google Posts Reporting is rolling out worldwide, showing view counts on every post card inside a business's dashboard. The count covers a rolling 18 months, spans regular updates, offers, and events, and combines views from Search and Maps into one number. The feature has no API access yet, and Google has not defined what counts as a single view.
In plain termsA view count here means how many times a post appeared to someone browsing Search or Maps, not whether they tapped it, since Google turned off click tracking on these posts back in January 2023.
Why it mattersIt gives local and multi-location businesses their first quantified read on which posts and offers are actually getting seen, though without an API it still has to be read one dashboard at a time.
Our takeAn 18-month lookback re-scores a business's whole existing library of posts overnight rather than starting the count from zero, which matters more for a multi-location account than a single storefront. The real test is whether this reaches the API, because a number that only lives inside one dashboard cannot get pulled into a client report at scale.
Confirmed
Google confirms it is testing a pilot that pays publishers when their content shapes an AI answer
Google confirmed to Digiday it is testing an AI contribution pilot, a Search Console-based program compensating publishers, now extending past news outlets to small and mid-size sites, when their content significantly shapes an AI-generated response inside the Gemini app, AI Overviews, or AI Mode. Payment tracks Google's own valuation of a page's contribution, not raw usage, and publishers can leave anytime. Google has not disclosed how it calculates the payout.
In plain termsPay-per-value means Google decides how much your page was worth to one specific AI answer and pays for that, instead of the older web model of getting paid indirectly, through ad clicks, for how many times a page got used.
Why it mattersIt's the first concrete, if opaque, link between being cited in an AI answer and getting paid for it, rather than counting on a referral click that AI answers rarely send anyway.
Our takePublishers already describe the early payouts as peanuts next to ad revenue, so the pilot reads less like a new revenue line and more like Google establishing that a paid relationship exists at all before the rate matters. Watch whether the payout math ever becomes public, since an unpublished formula is the same leverage problem publishers already have with organic ranking.