Confirmed
Bing Webmaster Tools Groups AI Search Queries Into Topic Clusters
Bing announced on October 13, 2026 a beta Topics Portfolio feature in Bing Webmaster Tools' AI Performance report, which groups a site's AI-answer grounding queries into thematic clusters (for example, taxes or football) rather than one flat list. Each topic cluster reports Topic Size, Sites Cited, and Query count, and the report highlights Top Opportunity Queries, shown as ranges rather than exact numbers, where competitors are being cited but the publisher's site is not.
In plain termsAI Performance is the Bing Webmaster Tools report showing which search queries caused Bing's AI answers to cite your site, and this update sorts those queries into named subject groups instead of one long list, plus flags whole topics where rivals get cited and you don't.
Why it mattersTopic-level grouping turns Bing's AI-citation data into content-strategy signal, showing which whole subject areas a site is losing to competitors in AI answers rather than one query at a time.
Our takeGrouping citation data by topic rather than query is the direction every AI-visibility report needs to go, since a practitioner managing hundreds of queries can act on a content gap but not on a single missed citation. Watch whether Google Search Console ships an equivalent grouping, since Bing moving first on structured AI-performance reporting puts pressure on Google's still-flat data.
Confirmed
Shopify's Sidekick AI Assistant Now Configures Checkout And Customer Accounts
Shopify announced on October 13, 2026 that its built-in AI assistant, Sidekick, can now help merchants configure checkout and customer-account settings directly inside the editor. In the checkout and accounts editor, Sidekick answers configuration questions, recommends apps for specific needs, and searches the Shopify Help Center; inside the Checkout Blocks app specifically, it explains what each block and function does, reviews a merchant's existing setup, and points to the relevant settings to change.
In plain termsSidekick is the chat-based assistant built into the Shopify admin, and this update lets a merchant ask it directly about checkout and account settings instead of searching Shopify's help docs or guessing at what a setting does.
Why it mattersCheckout configuration is one of the highest-stakes surfaces on a Shopify store, since a broken setting there blocks every sale, so an assistant that can review and explain an existing setup before a merchant changes it lowers the risk of self-inflicted checkout errors.
Our takeShopify keeps routing configuration work through its own AI assistant rather than through app-marketplace documentation, which means Sidekick's in-editor recommendations become another channel deciding which checkout apps a merchant discovers. Watch whether those app suggestions favor Shopify's own first-party tools over third-party checkout extensions.
Confirmed
Google Adds Experimental Soft-Navigation Metrics To CrUX
Google announced on October 13, 2026 that CrUX, the Chrome User Experience Report, now publishes experimental versions of all three Core Web Vitals that separately measure soft navigations, in-app page changes that skip a full reload. Only 7 to 11 percent of origins send majority soft-navigation traffic, but 35 to 42 percent of the top 1,000 sites on Android and desktop do. The metrics graduate around May 2027, when Search Console adopts the new data.
In plain termsCrUX is the real-world speed and stability data Google pulls from actual Chrome users rather than a lab test, and this update makes it start counting what happens when a page updates without a full reload, the way a lot of modern JavaScript-heavy sites work, instead of only counting the very first load.
Why it mattersCore Web Vitals on a high-traffic single-page app have measured only the first load and missed everything after, so once this graduates into Search Console, SPA-heavy sites should expect their reported Vitals numbers to move, for better or worse.
Our takeA single-page app has been getting Core Web Vitals credit only for its first load and a free pass on everything after, so this closes a measurement gap rather than changing a ranking signal today. The runway to May 2027 gives SPA-heavy sites time to fix soft-navigation performance before Search Console starts grading it.
Confirmed
Google Adds Gemini 4 Argon To AI Mode For Pro And Ultra Subscribers
Google's October 13, 2026 announcement introducing Gemini 4 Argon says the model is now selectable inside AI Mode in Search for Google AI Pro and Ultra subscribers, alongside the Gemini app. Ultra subscribers also get it in Google Antigravity and AI Studio; Pro subscribers get only the Gemini app and AI Mode. Google calls Argon its frontier reasoning model, ranking first on the Vals Index across finance, coding, legal, and tax tasks.
In plain termsAI Mode is the chat-style tab in Google Search that answers a question directly instead of listing links, and this update lets paying subscribers switch it to a more capable, newly released model instead of whatever Google runs for everyone else by default.
Why it mattersA new model option inside AI Mode changes what answer a paying subscriber actually sees for the same query, so any AI Mode visibility tracking needs to account for which model tier generated a given response, not just that AI Mode fired.
Our takeModel choice inside AI Mode is now a paid-tier feature, so the same search query can produce a materially different answer depending on subscription level alone, not just prompt or context. That is a new variable for anyone reasoning about what AI Mode cites or says, since the free and paid versions no longer run the same model.
Model & AI releases
AI model launches, model API price changes and major assistant features, kept separate from the search news above. Every model story Pulse has covered is collected under LLM Models & Agents, and the monthly model comparison shows how they stack up.
Confirmed
Liquid AI Releases LFM2.5-24B-A2B For On-Device Agents
Liquid AI released LFM2.5-24B-A2B on October 13, 2026, an open-weight reasoning model with 24 billion total parameters but only 2 billion active per token, built for on-device, tool-calling agents. Liquid AI reports it scores 48.09 on LiveClaw against 38.19 for Qwen3.6-35B-A3B, leads on instruction-following and tool-use benchmarks, and runs at 184 tokens per second on an Apple M5 Max and 21,000 tokens per second on an Nvidia B300, with a 131,000-token context window.
In plain termsActive parameters are the portion of a large model that actually does the work on any given answer, rather than the full stated size, and this model switches on only about a twelfth of its total size per request, which is why it can run fast enough for a phone or laptop instead of needing a data-center GPU.
Why it mattersA 2-billion-active-parameter model beating a 35-billion model on tool use is a mixture-of-experts architecture doing the work, so anyone evaluating on-device or edge agent deployments should treat active-parameter count, not total parameter count, as the real cost comparison.
Our takeOpen-weight models claiming to beat much larger closed competitors on narrow benchmarks are common enough that the benchmark choice matters more than the score, and LiveClaw is not yet one most practitioners can sanity-check against their own workload. The tokens-per-second figures on consumer hardware are the more concrete claim here, and worth verifying directly before counting on it for an on-device build.
Confirmed
Meta Releases Muse Wonder And Muse Wonder Full
Meta announced Muse Wonder and Muse Wonder Full on October 12, 2026, frontier models for personal and professional agent tasks, built around Active Memory, which retains context across conversations, and Active Focus, which sustains effort toward a goal while adapting mid-task. Both reach users through Meta's Model API and Muse Code, with Muse Wonder Full reserved for subscribers. Meta claims state-of-the-art results on Harvey's Legal Agent Benchmark, FocusBench, and SWE-Marathon, with no scores disclosed.
In plain termsAgentic tasks means a model working through a multi-step job on its own, like booking a trip or writing and testing code across several turns, rather than answering one question at a time, and these models are built to hold onto context and stay on-goal through that whole process.
Why it mattersMeta naming Active Memory and Active Focus as the headline features, rather than a raw benchmark jump, signals it is positioning Muse Wonder against long-running agent workflows rather than chat, a different buying decision than picking a model for one-shot tasks.
Our takeMeta reporting benchmark wins without disclosed scores is a weaker claim than it is presented as, since state-of-the-art with no number attached cannot be checked against a competitor's own reported result. The two named capabilities, memory that persists across a conversation and effort that holds steady on one goal, are the more concrete signal of what Meta is building toward: long-running agent work rather than one-shot chat.