Google today began rolling out a multimodal search type in the Search Console Performance report, separating web results reached through an image, photo or screenshot from those reached through typed text. According to Google, the figures in the new view were not previously counted anywhere in the report.

In Short

Google has added a new option to Search Console, its free reporting tool for website owners, that counts how often a site shows up when someone searches with a picture instead of typed words, for example by pointing a phone camera at an object. Owners of sites full of products, places or anything else people photograph could not see this kind of search before, because Google says it was never counted in the report at all. You can now see which pages appeared and how often, but not what the person was looking for, since a photo arrives without a search phrase.

What changed in the Performance report

Opening the Search type filter in the Performance report for Search results now produces two choices nested under Web: Text-based and Multimodal. Image, Video and News remain separate options in the same dialog, which keeps its Filter and Compare tabs, according to screenshots published by Search Engine Roundtable, which documented the change before Google's own post went up.

Google's description of the option settles the boundary in two sentences. "Web-multimodal tracks search results triggered by a query that uses an image, photo, or screenshot. Text-only queries are tracked as Web text-based," according to the text Google attached to the filter.

The split runs along a different axis from the rest of the dialog. In Search Console, the Image, Video and News types describe where a result appeared - the Images tab, the Videos tab, the News tab. The new pair describes what the searcher supplied. A photograph of a lamp that returns a retailer's product page among standard web results counts as Web: multimodal; the same person typing a description of the lamp into the Images tab still produces an Image row. Only Web has been subdivided.

Two help documents changed alongside the interface, according to Search Engine Roundtable. The overview and basic setup page for the Performance report now lists the new type as "Web: multimodal - Includes web search results where an image was used as part of the search." The page on dimensions and data groupings gained a paragraph on queries, which turns out to be the most consequential part of the update.

Four ways in

Google's post on the Search Central blog names four routes that feed the new bucket: Lens, Circle to Search on Android, image uploads to Google Search, and the right-click "Search this image" option in Chrome. The update "is designed to give you insights into how your content is surfaced when users search using images (such as with a smartphone camera)," according to Google. The post carries the names of Harsh Kharbanda, listed as Product Manager Lead for Google Lens, and Moshe Samet, Product Manager Lead for Search Console.

Both names recur in the product's recent history. Samet co-authored the June 3, 2026 post that gave generative AI features their own Search Console reports. Kharbanda described how Circle to Search breaks an outfit into separately searchable items on a Made by Google podcast published on April 27, 2026.

The list of four is notable for what it omits. Search Live, which lets people point a phone camera at something and talk to Search about it, reached more than 200 countries in March 2026. Image prompts inside AI Mode have existed since Google opened the product to image-based searches in April 2025. Neither is named. Whether those sessions fall under "image uploads to Google Search", sit elsewhere in the report or are excluded altogether, the material published today does not say.

A search type without search terms

The dimensions document states the limit plainly. "Because multimodal searches mostly use images rather than text, specific text query data isn't available for this traffic. As a result, the queries dimension isn't available when this search type is selected," according to Google.

John Mueller of Google put it more casually on Bluesky. Search Console "now reports on 'multimodal' queries like when you take a photo w/Lens," he wrote, adding that the photo itself stays hidden, "but you can see which of your pages showed up & how often."

Two phrases in Google's documentation carry more weight than they first appear to. The overview page says an image was used "as part of the search", and the dimensions page says multimodal searches "mostly" use images. Both leave room for searches that pair a picture with typed words. On a plain reading of the tooltip, only text-only queries count as text-based, so an image combined with a typed refinement belongs in the multimodal bucket - and the typed portion is not reported, because the queries dimension is switched off for the whole search type.

That gap differs in kind from the one site owners already work around. In the text-based report, Search Console withholds anonymised queries - terms issued by no more than a few dozen users across a two-to-three month window - and caps each interface table at 1,000 rows, so the visible rows routinely add up to less than the headline total. Those are partial omissions. The multimodal view removes the dimension entirely. Google names queries as unavailable and mentions no other restriction; nothing in the published material says clicks, impressions, position, pages, countries or devices are withheld in the main report.

What, then, does a multimodal row tell its reader? That a page was returned for a picture, how often, where and on which device. Whether the picture showed a product, a landmark or a screenshot of a text message, the row cannot say.

New data, or old data moved?

The opening of Search Engine Roundtable's story described the multimodal figures as brand new, never folded into the web totals. That reading was tested within hours. Dave Smart asked on Bluesky whether the impressions had previously sat under Web, equal to the sum of the two new sub-types, and reported corresponding declines in impressions and clicks on a few properties starting on the 10th, which he said lined up with the new data.

Mueller's first reply hedged. "I'll double-check, I thought these were not reported at all before," he wrote. The second was firmer: "checking with the team, the data wasn't previously in the counts, the report has new data."

That fixes Google's position. It does not explain Smart's observation. If multimodal impressions were never part of the web totals, their arrival cannot have subtracted anything from text-based figures, and the declines he saw would need another cause. The material published today does not identify one, nor does it say from which date the multimodal series begins. A second question is also open: whether selecting Web at the top level now returns text-based and multimodal combined. If it does, sites with meaningful camera-driven exposure would see Web totals step up at the point the new counts begin, an artificial break in any year-on-year comparison.

The calendar adds noise. Search Engine Roundtable's listings for the day also record ranking volatility on September 23 and 24 and the start of a September 2026 spam update, so movement in text-based figures this week has several candidate explanations.

Search Console has a record of such breaks. AI Mode clicks and impressions were merged into Web totals from June 17, 2025, with no way to separate them. A logging error inflated impressions from May 13, 2025 until Google acknowledged it on April 3, 2026. Impressions and clicks fell again in mid-August 2026, which Google annotated as a logging error dated August 13. Each of those events changed what the word Web contained for a period. Today's change is the first to subdivide it in the interface.

The generative AI view gets the same split

The change is not confined to the classic report. Google's post says multimodal reporting appears "in the report for performance on Search results as well as in the report for Generative AI features."

That second report has a short history. It arrived on June 3, 2026, covering impressions, pages, countries, devices and dates for AI Overviews, AI Mode and generative features in Discover, and reached every site worldwide on August 31, 2026 without click or query data. Search Engine Roundtable noted today that clicks and queries are still absent there. Layered together, the two limits mean the multimodal slice of the generative AI report reduces to impressions by page, country, device and date. It can show that a camera search led to a page appearing inside an AI response. It cannot show what the camera saw, or whether anyone clicked.

Regulators have asked for more than that on the generative side. The UK Competition and Markets Authority's publisher conduct requirement, imposed on June 3, 2026, calls for impressions, click-throughs and click-through rate for generative AI features, disaggregated from organic results, through a platform such as Search Console. The multimodal filter adds a way to segment the data. It does not supply the click figures that requirement describes.

How large the camera channel has become

Google has published scattered figures on visual search, none of them broken down by the result type the new filter isolates. When Shopping ads arrived in Lens on October 3, 2024, Google put visual searches through the tool at nearly 20 billion a month, around 20 percent of them shopping-related. Its first-quarter 2025 figures added five billion monthly Lens searches since that October and nearly 40 percent growth in Circle to Search usage, and by July 9, 2025 Circle to Search was available on more than 300 million Android devices. Robby Stein, vice president of product for Google Search, cited a 65 percent year-over-year increase in visual searching during a podcast on July 23, 2025.

The entry points have multiplied since. Lens in desktop Chrome gained a route into AI Mode on July 29, 2025, and Circle to Search began analysing the entire image rather than only the circled region on February 26, 2026.

All of those are platform-level volumes. None indicates how many camera searches end on a third-party website rather than inside a Google surface, which is the quantity the multimodal filter begins to expose, property by property. The overlap with commerce is where that matters most. A fifth of Lens activity was shopping-related on Google's own 2024 figures, and the retail pages capturing that activity through organic web results had, until today, no separate line in Search Console.

One input, three treatments

The same photograph is now handled three different ways, depending on who is reading the data.

On the paid side, Google revised its ad group prioritisation documentation in May 2026 so that searches from Lens, AI Mode, AI Overviews and autocomplete show advertisers Google's best approximation of intent in the search terms report, rather than a literal string. An advertiser whose ad serves against a Lens search sees an inferred term.

In Brussels, the European Commission's Digital Markets Act decision on search data sharing, adopted on July 16, 2026, treats Lens and Circle to Search inputs as queries, because the legal definition of a query extends to non-text input such as images.

In Search Console, the site owner receives no query at all.

Each position has its own logic. Advertisers need a term to bid, exclude and report against. Rival search engines under the DMA are entitled to data on those inputs in some form. And, as Google's documentation puts it, the searches mostly use images rather than text. The result, though, leaves the organic reporting surface as the only one of the three in which the searcher's input is not represented.

Rollout and first sightings

"The integration of multimodal search data is rolling out globally starting today," according to Google, and the metrics appear only where a site "is receiving traffic from these queries." The post points to the Export button for downloading the data. It does not mention the Search Console API or the bulk data export, and nothing published today says whether either carries the new type.

The filter was visible before the post. Tauqueer Aziz shared a screenshot of it on X ahead of Google's statement, and Search Engine Roundtable's story was updated afterwards to record that Google posted at 8:40am Eastern Time. One commenter on that story reported finding the filter, with data, on all of roughly ten websites checked; another noted at the time that no official statement existed yet. An account replying to Barry Schwartz of Search Engine Roundtable on X pointed out that multimodal web search carried no query data, and Schwartz replied that he was seeing the same.

A staggered arrival matches recent Search Console practice. The generative AI reports took from the first week of June to August 31 to reach every site, and platform properties for Instagram, TikTok, X and YouTube moved from a gradual start on July 7 to full availability on July 29, 2026.

What the split changes for measurement

When AI Mode data entered Search Console in June 2025, Glenn Gabe objected that ten blue links, featured snippets, AI Overviews and AI Mode now sat together under Web Search. Today's filter cuts into that bucket directly, rather than building a parallel report beside it as Google did for generative AI in June.

Because the new cut is by input and the June report is by surface, the two overlap. A camera search can produce an AI Overview; a generative AI impression can be text-based or multimodal. Search Console now carries two independent axes - where a result appeared and what triggered it. At their intersection, a camera search that surfaces a page inside an AI response, the report shows neither the words behind the search nor a click.

For publishers, relevance depends on subject matter. Mueller pitched the change at image-heavy websites, and the sites most exposed are likely those built around objects people point cameras at: retail catalogues, food, home and garden, travel. For agencies reporting organic performance, the first consequence is definitional. Until Google clarifies how the top-level Web option now aggregates, a line labelled Web in a client dashboard may not mean what it meant last month.

Timeline

Summary

Who: Google, through a Search Central blog post carrying the names of Harsh Kharbanda, Product Manager Lead for Google Lens, and Moshe Samet, Product Manager Lead for Search Console, with public comment from John Mueller of Google. The change affects website owners, publishers, retailers and agencies using Search Console; Search Engine Roundtable documented it ahead of Google's post, with early sightings from Tauqueer Aziz and others.

What: The Search type filter in the Search Console Performance report now splits Web into Text-based and Multimodal. Multimodal covers web results triggered by searches that use an image, photo or screenshot, through Lens, Circle to Search on Android, image uploads to Google Search and Chrome's "Search this image" option. The queries dimension is unavailable for the new type, and the split also applies to the generative AI performance report, which still lacks clicks and queries. Google says the data was not previously in the counts.

When: Today, with Google's post at 8:40am Eastern Time and a global rollout starting the same day. Screenshots circulated on X before the post.

Where: Globally, inside the Performance report for Search results and the generative AI performance report in Search Console, for sites that receive traffic from image-based searches.

Why: Visual search has grown into a channel measured in billions of monthly searches, yet its traffic to third-party websites had no line of its own in Search Console. The filter gives site owners page-level visibility into that traffic but no query information, even though Google Ads reporting and the EU's data-sharing decision both treat the same inputs as queries. How the top-level Web figure now aggregates remains unclear.