the assistants recommend whoever the referees rate, not whoever is biggest.

OpenAI rolled out conversion-optimised CPC bidding in beta for product-feed campaigns, dynamic URL parameters, new measurement integrations, and Automatic Advanced Matching, default on existing pixels from 17 August unless an advertiser opts out. Ads launch in Brazil and Mexico. Separately, a multi-product carousel has been seen at the foot of conversations for L.L.Bean, VistaPrint and Walmart. Confidence differs: the bidding and matching changes come from OpenAI's advertiser communications, the carousel from Digiday screenshots, with OpenAI declining to comment.

The appeals court overturned the injunction keeping Perplexity's Comet agent off Amazon. Because the agent acts only on user instruction, the court treated the user, not Perplexity, as the party accessing Amazon's servers, defeating the unauthorised-access argument. The underlying suit continues and Amazon may seek rehearing, so this is a ruling, not a settled rule.

FAQ rich results stopped showing on 7 May 2026, the search-appearance filter and Rich Results Test support went in June, and Search Console API support is removed this month. Google's documentation now says the markup is not used for rich results, while noting owners can leave it in place.

Forty consumer-phrased questions, twenty each to ChatGPT and Google AI Mode, across flights, broadband, energy, current accounts and mobile, on logged-in UK free tiers in a single day. Octopus Energy took 100% share of voice in energy, Starling and Monzo 100% in current accounts, EE, Vodafone and Three 100% in mobile, Ryanair and easyJet 75% in flights. Zen Internet, at about 2% of the broadband market, was top pick more often than BT, Sky and Virgin Media combined. Under the answers sat Which? in every category, then MoneySavingExpert, Uswitch, Ofcom, Citizens Advice, OpenSignal and Reddit. Gemini was excluded because the free tier could not complete the panel.

Judge it honestly: forty questions, one country, one day, one run each, from an agency rather than a research house. It cannot carry a claim on the site. It can illustrate, because the direction matches the Yext 86% brand-managed finding and the Semrush cited-versus-named split, and because a single-run design is exactly the failure mode our two-pass rule exists to prevent.