A new Brand Accuracy Eval panel sits on the Sentiment page. It compares what AI platforms actually said about you (pulled from your scan snippets) against your brand factsheet, then surfaces what they're getting right, what's wrong, what's missing, and where AI is quietly steering certain audiences toward competitors without criticizing you directly. Each finding includes the exact AI quote behind it, and you get up to three concrete suggestions for closing the gaps.
The Sentiment page also got a structural overhaul. Prompts now collapse into a single row that expands to show each persona, then each platform under that persona — so you're reading results in the order they actually matter. The Raw button moved to the persona level where it belongs, avatars appear in both the table and the raw-response modal, and Competitor Mentions moved to its own tab now that competitors have dedicated space.
Query classification is new on the Topics page. Sentiment queries are now automatically split into two types: prompts that name your brand directly, and open-ended recommendation prompts where your brand has to earn its way into the answer. Each query is classified when saved and gets a "Brand" badge. You can filter by type or override the classification manually with an inline toggle. The fanout search tooltips on the same page now explain that sub-queries are the AI's internal research process — not additional scans or credits.
AI Overview answers now consistently come back in English. Previously, setting only the geographic parameter left the answer language up to Google, which occasionally returned Spanish or Portuguese responses for English queries. The Factsheet page also loads correctly again — the interview questions no longer stall on "Loading questions…".