Google Search Console adds multimodal web reporting.

Google Search Console is adding a separate view of web searches that begin with an image. The new Web: multimodal filter is rolling out globally in both the Search results Performance report and the Generative AI performance report. It can reveal which pages appear when people search with Lens, Circle to Search, an uploaded image or Chrome’s “Search this image.” It does not reveal the image a person supplied or the text of a query that was never typed.

For publishers, the useful change is a new measurement segment, not a new way to rank. Google says the metrics appear for properties receiving qualifying traffic. An absent segment or a small first-day total is not evidence that a site’s images are failing.

What the new filter separates

In Google’s Search results report definitions, Web: text-based covers text queries in the standard search bar, while Web: multimodal covers web results where an image was part of the search. The existing Image search type is a different surface. A publisher should therefore label an export by both its report and its search type, rather than put every visual-search number into an “image traffic” bucket.

Google’s dimensions documentation says the query table is unavailable for multimodal searches because they mostly use images rather than text. That makes page, device, country and date the useful dimensions for a first analysis. It also means a Search Console export cannot tell you which object appeared in a user’s photo or which visual prompt produced an impression.

A clean first export

Open Search results Performance, choose a fixed date range that does not rely on preliminary hours, select Web: multimodal, and export the page and date views. Save a second export for Web: text-based using the same property and dates. Keep the report name and filter in each file name. Google’s announcement specifically points publishers to the search-type filter and Export control.

Next, group the multimodal pages by job: product detail, tutorial, comparison, local page or editorial article. Inspect the actual visible images and the linked page for the highest-impression URLs. This is an editorial inventory, not proof that a particular alt attribute, file name or schema field caused the appearance. If the filter has no qualifying rows, record that state and repeat the export after the rollout has had time to populate. Do not fill the empty space with a synthetic screenshot or extrapolated totals.

A reporting change can also alter the way historical totals are presented. Preserve a pre-change export where available, compare like-for-like search types, and annotate September 24 as the rollout date before declaring a traffic gain or loss. Our AI visibility reporting guide explains why an impression, a source mention, a citation and a visit need separate denominators.

The AI report has a narrower outcome

Google’s Generative AI report documentation defines its metric as impressions of links to a site in supported Google Search AI features, currently AI Overviews and AI Mode. It offers a multimodal filter, but it does not provide clicks in that report. An impression there is not necessarily a recommendation, an observed citation in a saved answer, a visit or a sale.

That distinction sets the next useful test. If a page appears in the multimodal AI segment, preserve its URL and date, then separately examine server or analytics referrals and a repeatable sample of visible answers. Do not divide AI-report impressions by Search results clicks and call the result an AI click-through rate: the numerator and denominator come from different report definitions.

This article describes Google’s published reporting contract, not a completed SearchEngineAnswer property experiment. We will add a dated export and page-level observations if the property accumulates qualifying rows. Until then, the defensible action is to set up the comparison correctly and keep the new measurement boundary intact.