# Pulse ยท October 28, 2026 > Google confirms a Discover clicks bug and updates its favicon crawl docs, Shopify adds auto-synced translations, and Liquid AI ships an on-device model family. Canonical: https://brandonlazovic.dev/pulse/2026-10-28/ Author: Brandon Lazovic Published: 2026-10-28 ## Google Confirms Disabled Discover Source Clicks Were An Experiment Bug Status: Confirmed | Topics: ai-overviews-ai-mode Google's VP of Engineering for Search, Rajan Patel, said on X on October 27 that taps on source links inside Discover's AI-generated summaries had stopped working because of a bug in an active experiment, not a deliberate design choice. Patel said the experiment is now off and the links work as before, though he did not disclose what the experiment was testing or how long it ran. In plain terms: Discover is the personalized feed of articles Google shows in the Google app and on Android home screens, and its AI summaries sometimes carry small links back to the source article. Those links briefly stopped being clickable. Why it matters: Publishers watching Discover referral traffic now have a name for an unexplained dip, and a reminder that an unannounced Google experiment can silently break a click path site owners depend on. Our take: A VP confirming a bug within a day, with no account of what the experiment measured, is damage control for the click numbers rather than transparency about the test itself. Any read of Discover referral trends from the affected window should discount the dip as experiment noise rather than a ranking or policy change. - [Search Engine Journal: Google: Disabled Discover Clicks Were A Bug In An Experiment](https://www.searchenginejournal.com/rajan-patel-disabled-discover-source-clicks-bug/594579/) ## Google Updates Favicon Documentation To Require Redirect-Following Status: Confirmed | Topics: crawling-indexing-rendering Google revised its favicon documentation on October 27 to state that Googlebot must crawl a site's homepage, follow any redirects from there to wherever the favicon is declared, and that Googlebot-Image must be able to crawl the favicon file itself. The prior version covered crawling the homepage and not blocking the favicon file, but never addressed a redirect in between. John Mueller said the clarification responds to confusion, including from AI systems, about troubleshooting a missing favicon. In plain terms: A favicon is the small icon shown next to a site's name in a browser tab or in Google's search results, and 'redirect-following' means Googlebot has to trace a chain of forwarding addresses rather than stop at the first page it loads. Why it matters: A site that routes its homepage through an intermediate redirect before reaching the favicon reference now has a documented, checkable reason its icon might be missing from search results. Our take: A documentation fix framed as useful for AI systems as well as human developers signals Google is now writing crawl guidance with a machine reader in mind. Check any redirect between your homepage and your declared favicon file before assuming a missing icon is a caching problem. - [Google Search Central: Favicon in Search Results](https://developers.google.com/search/docs/appearance/favicon-in-search) - [Search Engine Roundtable: Googlebot Must Follow Redirects From Homepage To Favicon, Updated Doc Says](https://www.seroundtable.com/googlebot-redirects-from-homepage-to-favicon-42396.html) ## Shopify Adds Automatic Weekly Translation Syncing For Up To Eight Languages Status: Confirmed | Topics: platform-ecommerce Shopify shipped an Auto-sync option inside its Translate & Adapt app on October 27 for US and Canada merchants, letting a store automatically retranslate new or edited content every week across up to eight published languages. Auto-sync is off by default and never overwrites a merchant's manual edits. Shopify also renamed the prior 'Auto-translate' button to 'Translate now' for immediate, on-demand translation outside the weekly cycle. In plain terms: Translate & Adapt is Shopify's built-in app for turning a store's product pages and policies into other languages, and 'Auto-sync' means it now checks weekly for anything a merchant edited and updates the translation on its own instead of waiting to be asked. Why it matters: Merchants running localized storefronts no longer have to manually re-trigger a translation after every product or policy edit, closing a gap where international buyers were landing on outdated-language pages. Our take: A weekly sync cadence is a deliberate middle ground between translation cost and freshness, cheap enough to run on every published language without charging a merchant per edit. Any store relying on Translate & Adapt should confirm which languages have Auto-sync turned on, since the feature ships off by default and a stale translation stays stale until someone flips it. - [Shopify Changelog: Translate & Adapt US/CAN Update: Auto-sync translations for up to 8 languages](https://changelog.shopify.com/posts/translate-adapt-us-can-update-auto-sync-translations-for-up-to-8-languages) ## Google Tests A Fan POV Button That Opens A Customized AI Mode View For Sports Searches Status: Observed | Topics: ai-overviews-ai-mode Search Engine Roundtable reported on October 28 that Google is testing a 'Fan POV' button on sports queries such as the World Series, first spotted by Mordy Oberstein on X. Clicking the button, which carries a sport-specific emoji, opens a dialog for picking a game and a team, then returns an AI Mode answer written from that team's fan perspective rather than a neutral recap. Google has not confirmed the test or its scope. In plain terms: AI Mode is Google's conversational search experience that answers a query in paragraphs instead of a list of links, and 'Fan POV' is personalization applied to tone rather than to facts: the same event gets rewritten to sound like it is rooting for whichever team a reader picks. Why it matters: A results page that branches into partisan framing based on which team a reader picks is a new kind of AI Mode output sports and local-team content strategies have not had to account for yet. Our take: Fan-perspective framing is a search feature that is honest about being biased by design, a different trust question than a neutral AI summary getting a fact wrong. Sports and local-team publishers are worth watching here first, since a feature this specific to fandom is unlikely to generalize across AI Mode broadly. - [Search Engine Roundtable: Google Search Fan POV (Point Of View) Button](https://www.seroundtable.com/google-search-fan-pov-42401.html) ## Google Business Profiles Starts Adding Facebook Messenger Links Automatically Status: Observed | Topics: platform-ecommerce Search Engine Roundtable reported on October 28 that Google is emailing businesses to tell them it has automatically added a Facebook Messenger m.me link, Meta's short URL that opens a direct chat with a Page, to their Google Business Profile. The report rests on a forwarded email and social posts rather than a Google help-center update, and does not say which businesses are affected or whether the link can be removed. In plain terms: A Google Business Profile is the free listing showing a business's hours, reviews and contact options in Google Search and Maps, and m.me is Meta's shortcut link format that opens a Messenger chat window when clicked instead of placing a call. Why it matters: A business that doesn't actively staff Messenger may now be directing searchers to an unmonitored chat channel it never opted into. Our take: An automatic add with no opt-in and no visible setting to remove it is worth checking even though Google has not documented it anywhere. A business owner who doesn't staff a Messenger inbox should check their own profile directly rather than wait for a help-center page that may never arrive. - [Search Engine Roundtable: Google Business Profiles Adds Facebook Messenger M.me Links](https://www.seroundtable.com/google-facebook-messenger-mme-links-42377.html) ## Google Tests Moving The Image Save Button Onto The Search Results Overlay Status: Observed | Topics: organic-search-core Search Engine Roundtable reported on October 28 that Google is testing a version of Image Search where the Save button appears directly on an image when a searcher hovers over it in the results grid, rather than only after opening the image's preview window as today. The change was spotted by Damien Andell in a video posted to X, and Google has not confirmed the test or said how widely it is running. In plain terms: The 'overlay' is the hover state Google Images shows over a thumbnail before a searcher clicks it, and moving the Save button there means a user could save an image to a collection without ever opening it. Why it matters: A lower-friction save action could shift how much time a searcher spends on the results grid itself instead of clicking through to a publisher's image preview or source page. Our take: Making it easier to save an image without opening it works against the click-through behavior publishers depend on from image search. Sites that get meaningful referral traffic from Google Images are worth watching for a dip tied to this test's rollout rather than a ranking change. - [Search Engine Roundtable: Google Image Search Tests Save Button On Image Overlay](https://www.seroundtable.com/google-images-save-button-overlay-42394.html) ## Bing Tests A Scan-To-Call QR Code Extension In Sponsored Local Ads Status: Observed | Topics: ads-paid Search Engine Roundtable reported on October 28 that Bing is testing a phone icon inside sponsored ads in desktop local pack results that, when clicked, reveals a QR code a searcher can scan to call the business instead of dialing from the page. The sighting came from a single screenshot shared by Khushal Bherwani on X, the article's author could not reproduce the test, and Microsoft has not confirmed it. Google shipped a similar scan-to-call QR code in ads back in 2023. In plain terms: A 'local pack' is the small map-plus-listings block Bing or Google shows above regular results for a 'near me' or business-name search, and this test is an extra tappable element inside a paid ad placed in that block, not a change to the free organic listings. Why it matters: Even an unconfirmed, low-visibility test is worth flagging for clients running local paid search, since a QR-code call path changes which device completes the call and what gets logged as a conversion in an ad account. Our take: A feature this unreplicated by the outlet reporting it belongs in next month's recheck, not in a client recommendation today. If it ships broadly, a local paid-search account should expect calls placed by a second device to complicate call-attribution counts. - [Search Engine Roundtable: Bing Tests Get Phone Number & QR Code To Scan](https://www.seroundtable.com/bing-get-phone-number-qr-code-42397.html) ## Model & AI releases ### Liquid AI Ships LFM3, An Open-Weight Family Of On-Device Multimodal Models Status: Confirmed | Topics: llm-models-agents Liquid AI released LFM3 on October 27, an open-weight family of on-device models spanning a 1.7B-parameter text model with a 256K-token context window, plus 2B-parameter vision, audio and all-modality variants. Liquid AI reports its LFM3-1.7B text model outperforms its own prior LFM2.5-2.6B while using 35% fewer parameters, and scored 38.42 against 3.02 for LFM2.5-2.6B on the WebArena-Verified agentic-browsing benchmark. Weights, fine-tuning notebooks and integration guides are live on Hugging Face now. In plain terms: 'Open-weight' means the trained model file is published for anyone to download and run, unlike a closed model reachable only through an API, and 'on-device' means a phone or laptop can run it locally rather than sending data to a server. Why it matters: A team choosing a model to run on a device rather than through an API now has an open-weight option sized for that constraint, with a stated edge over its own predecessor on agentic-browsing tasks worth a benchmark re-test against whatever on-device model is in production today. Our take: Liquid AI is positioning on-device size as the differentiator rather than raw benchmark leadership, since its own comparison point is its prior generation rather than a larger frontier model. The WebArena-Verified score is Liquid AI's own number, so treat the 38.42-versus-3.02 gap as a claim to re-run rather than a settled result before swapping in a production on-device model. - [Hugging Face Blog: Build Specialized Multimodal Models With LFM3](https://huggingface.co/blog/LiquidAI/lfm3)