Confirmed
Search Console adds a multimodal filter to Performance reporting, covering Lens and image-based search
Google Search Console began reporting on web multimodal search on September 24, in both the Search results and the Generative AI features Performance reports. The filter isolates traffic from Lens, Circle to Search on Android, image uploads to Google Search, and Chrome's right-click Search this image feature. Google said the data is rolling out globally starting today and will appear once a site receives traffic from these query types.
In plain termsMultimodal search means someone found your page by pointing a camera or uploading a photo instead of typing words, the way Google Lens or Android's Circle to Search work; Search Console can show that traffic as its own bucket, separate from typed searches.
Why it mattersSites get a visibility signal for camera- and image-triggered search, but Google's own post gives no query text with it, so the segment shows volume, not intent.
Our takeI argued in 'Search Console counts your AI impressions and cannot connect one of them to a visit' that Google ships these reports one dimension short. This is the same shape: you get a multimodal segment with no query text attached, so you can watch the volume rise and never learn what image someone searched.
Confirmed
YouTube adds AI shopping comparisons to Ask YouTube search results
YouTube's Ask YouTube AI search tool returns comparison tables organized by product attributes and categories when someone searches for product reviews, and a viewer can keep asking follow-up questions about a product directly on the video's watch page. YouTube VP Johanna Voolich announced the feature alongside other AI search updates at the Made on YouTube event, without specifying a rollout timeline or geographic limits for the shopping capability.
In plain termsAsk YouTube is a conversational search box inside YouTube itself; instead of only returning video links, it also builds a small comparison chart of products mentioned in review videos, the way a shopping-comparison site would.
Why it mattersA comparison table generated inside YouTube's own search answer is a ranking function a brand does not control, built from whatever product data YouTube's AI can find about that item.
Our takeWhether that comparison table pulls from a brand's own product data or from what a reviewer says on camera is the detail worth watching, since it decides whether accurate structured data still buys any influence over the answer.
Observed
Google tests letting business owners describe their hours in plain text and auto-generate structured hours
A screenshot shared by Ayman Ali on X shows a Google Business Profile test called "Set your hours automatically," where an owner types a plain-language description of when the business is open and Google extracts structured hours from it. Google has not confirmed the test publicly, and no rollout scope or timeline is documented.
In plain termsStructured hours are the machine-readable open and close times Google Search and Maps actually display, normally entered field by field; this test lets an owner write one sentence and has Google convert it.
Why it mattersIf Google is willing to parse free text into structured hours, the same instinct likely extends to other profile fields where owners currently leave the structured version blank.
Our takeAn AI conversion step like that only works as well as the sentence it is given, so a business with genuinely irregular hours is the harder test case, not the easy one this screenshot shows.
Observed
Google tests moving AI Overviews to a side panel for hotel searches
A test spotted by SEO professional Lluc Penycate and shared on X repositions the AI Overview for hotel searches from the top of the results page to a panel in the bottom right, underneath the hotel knowledge panel. Google has not confirmed the test, and the scope appears limited to hotel queries so far.
In plain termsAn AI Overview is the generated summary block Google sometimes shows above the normal blue links; testing it in a side panel means it competes for attention differently than a block that pushes everything else down the page.
Why it mattersMoving an AI Overview off the top of the page changes how much space and attention it commands before a reader ever scrolls, on a query type where booking sites already fight hard for position.
Our takeA side placement on hotel queries specifically suggests Google is treating travel and booking intent as a special case worth testing separately, not a template about to roll out everywhere else.
Confirmed
AWS adds generative-AI natural-language querying to Amazon SageMaker Catalog
Amazon SageMaker Catalog added the ability for consumers to query unstructured data assets, like documents, in plain English through a no-code chat agent, or programmatically through Amazon Bedrock model inference in a notebook. The capability builds on SageMaker Catalog's existing publish and subscribe model, where AI-generated metadata keywords make unstructured assets discoverable in the first place.
In plain termsUnstructured data is anything that is not already a tidy spreadsheet row, like a PDF or a support ticket; this update lets someone ask a plain-English question about a pile of those documents instead of writing a query language.
Why it mattersAnyone whose product or content data lives as unstructured documents rather than clean tables is closer to having that data usable by an internal AI agent without a separate extraction project.
Our takeThe no-code chat agent is the part worth testing before trusting, since a natural-language query over unstructured documents is exactly where a wrong but confident answer is hardest to catch.
Confirmed
Databricks ships Unity Catalog Pages in beta, a governed home for business definitions AI agents can cite
Unity Catalog Pages gives a company one governed place for the definitions of its key terms, entities, and acronyms, organized by domain next to the data those terms describe. Databricks said the same term is often defined three different ways by three different teams, and an AI agent that hits that ambiguity tends to guess confidently and incorrectly. Pages is available today in beta.
In plain termsGoverned means one definition is marked as the official one and every agent is pointed at it, instead of three teams' spreadsheets quietly disagreeing about what a term means.
Why it mattersAn agent that guesses at what active customer or net revenue means for your business is a governance gap, not a model quality gap, and this is Databricks betting that gap needs its own product.
Our takeWhether this actually gets adopted depends on someone doing the unglamorous work of writing down what every term means in one place, which no catalog product can force a company to do.