Confirmed
Google Confirms JPEG XL Is Unsupported In Images And Discover Previews
Google Search Advocate John Mueller said on Bluesky on October 15, 2026, that JPEG XL is currently not supported for images shown in Google Images or as preview thumbnails in Discover. He linked directly to Google's own Image SEO documentation, whose supported-format list names BMP, GIF, JPEG, PNG, WebP, SVG, and AVIF with no JPEG XL entry. Sites serving JPEG XL should add a fallback image in a supported format to keep Images and Discover visibility.
In plain termsJPEG XL is a newer image format built for smaller file sizes than regular JPEG at the same quality; Google Images and Discover simply cannot read it yet.
Why it mattersA site that switched a product or article's hero image to JPEG XL for its compression gains gets no visual presence in Google Images or Discover unless a supported-format fallback is also served.
Our takeThis is a gap worth checking before a format migration, not after: confirm the fallback renders correctly in both channels rather than finding out from a drop in image-search traffic.
Confirmed
Google Consolidates Its Search Data-Control Documentation Into One Page
Google updated its crawling-and-indexing help documentation on October 21, 2026, to centralize every existing method for controlling what content Search uses: removing content, password-protecting files, noindex, robots.txt, the robots meta tag, data-nosnippet, X-Robots-Tag, and the Search generative AI control that covers inclusion in AI Overviews and AI Mode. Every option listed already existed; Google states the goal is to keep the page current with the full set of options available to site owners, not to add new controls.
In plain termsThe Search generative AI control is the opt-out that keeps a site's content out of AI Overviews and AI Mode specifically, separate from blocking it from regular search results.
Why it mattersA single reference page that lines up noindex, nosnippet, and the generative-AI opt-out side by side makes it easier to audit which control actually does what a site owner wants, since those three are commonly confused for each other.
Our takeWorth a bookmark rather than an action: nothing changed about what the controls do, only where they are documented, so this is reference hygiene rather than a new lever to pull.
Confirmed
Cloudflare Says Its Network Signals Helped ChatGPT Search Index Hundreds Of Millions Of New Pages
Cloudflare and OpenAI published results on October 15, 2026, from a pilot that started in July 2026 across about 1% of Cloudflare's total network. The pilot shared traffic-pattern signals, change-detection signals showing when a page is worth re-crawling, and safety signals flagging compromised pages, rather than page content itself. Cloudflare says the signals helped OpenAI discover and index hundreds of millions of additional web pages, and that a signal grants discovery only, not crawling permission, which robots.txt and crawler settings still control.
In plain termsA signal here is metadata about a page, such as how recently it changed or how much real traffic it gets, not the page's actual content; OpenAI still has to request and be allowed to crawl the page itself.
Why it mattersA metadata partnership that surfaces low-link-equity pages to an AI crawler is a new discovery path independent of backlinks, so a page with weak link signals but real traffic or recent updates may now get found and cited where it wasn't before.
Our takeExpanded discovery is not expanded ranking: a page surfaced this way still competes on OpenAI's own relevance evaluation, so treat this as a wider net rather than a guaranteed citation.
Confirmed
Google Merchant Center Sets Country-Specific Day Thresholds For Fast-Shipping Badges
Google Merchant Center help documentation now specifies the maximum delivery time that qualifies a product listing for fast-shipping annotations and badging, such as "Free 3-day" or "Get it by [date]," by country: 1 day for Japan and New Zealand, 2 days for India, 3 days for Australia, Canada, the United Kingdom, and the United States, and 4 days for Brazil. The documentation does not state when these thresholds took effect or whether they replace different prior figures.
In plain termsA fast-shipping badge is the small label Google shows next to a Shopping listing promising delivery within a set number of days; missing the country's threshold removes the badge, not just the specific day count shown.
Why it mattersA retailer whose delivery promise sits one day over its country's threshold loses the fast-shipping badge entirely rather than losing it gradually, so this is worth checking against actual carrier delivery windows per market rather than assuming last year's promise still qualifies.
Our takeSkipped: no stated effective date, so this reads as documentation catching up to an existing eligibility rule rather than a change retailers need to react to this week.
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
Mistral Launches Voxtral 3, A Real-Time Voice Agent SDK
Mistral launched Voxtral 3, a voice-agent SDK combining transcription, text-to-speech, and an agent framework into one real-time pipeline that listens, reasons, calls tools, and speaks. Its batch transcription model covers 50-plus languages; its realtime transcription model delivers sub-500ms latency across 40-plus languages. Text-to-speech covers 35-plus languages and 50-plus dialects with voice cloning from 30-plus minutes of reference audio. Pricing runs $0.0035 to $0.008 per minute for transcription and $0.04 to $0.08 per minute for the full voice-agent API.
In plain termsA voice-agent SDK bundles the listening, reasoning, and talking parts of a voice assistant into one connected pipeline, instead of a developer stitching together three separate vendor tools.
Why it mattersA single SDK pricing listening, reasoning, and speaking together changes the build-vs-buy math for a voice agent that previously stitched together separate transcription, LLM, and TTS vendors.
Our takeThe latency and pricing figures are Mistral's own, so they are the part worth independently re-testing before swapping an existing voice stack onto this SDK.