Search Console Now Separates Visual Search Traffic

Until this week, a visit that began with someone pointing a phone camera at your product was indistinguishable in Search Console from a visit that began with a typed query. On September 24, 2026 Google split them. The Performance report and the Generative AI features report now carry a multimodal search type, covering Lens, Circle to Search on Android, image uploads to Google Search and Chrome's right-click image search. The data exports like any other filter.

That is the whole announcement, and it is enough to act on. What follows is what the filter counts and does not count, according to Google's post and help documentation, the three questions it can now answer for an ecommerce or image-heavy site, and a baseline routine to start this week, because the useful number is not this week's share but the change in it.

The Search Central announcement, posted by the product leads for Google Lens and for Search Console, says the new reporting is "designed to give you insights into how your content is surfaced when users search using images (such as with a smartphone camera)". The Performance report help page now lists the search-type filter as web, split into text-based and multimodal, then image, video and news.

Three boundaries matter when you read the first numbers. The filter is a subdivision of the Web search type, so a click from the Images tab is still an image-type click and is not in the multimodal row. The definition is about how the search was made, not where the result appeared: a Lens search that lands on ordinary web results counts, and one that lands in an AI Overview counts in the generative AI report. And nothing in the announcement or the help page says whether the query dimension is populated for these searches, so do not assume you will see the text Google derived from the image.

One gap is verifiable today. The Search Analytics API reference, read on September 25, lists six values for its type parameter: web, image, video, news, discover and googleNews. There is no multimodal value and no text-based value. A dashboard fed by the API will keep reporting the combined Web number until the reference changes. Our guide to Search Console's platform properties describes the same interface-first pattern for other additions.

What Google has not said, and we will not estimate, is how much traffic this is. The post gives no share, no growth figure and no example. Any number you see quoted this week comes from a third party's own properties, not from Google.

The third question is the one a marketing director asks and the one Search Console alone cannot answer. Our post on the AI performance report made the same point about AI Overview impressions: the report tells you where you appeared, and only your own data tells you what it was worth.

The filter is new, so every property's history under it starts this week. The value of the data is the comparison you can make in three months, and that comparison needs an export you took today. Our retention table lists how long Search Console keeps performance history; the short version is that it is long enough to forget to export and short enough to regret it.

A new row in a familiar report gets read with old habits. The illustrative readings below each follow from skipping one of the three boundaries in the second section. They are patterns, not reports from a client property.

Each misreading is cheap to avoid now and expensive to unwind after three months of reports. The export record in the next section is where the boundaries are written down once.

The four also show why the announcement is enough to act on without being enough to conclude from. Google has published the definition, the entry points and the reports the filter appears in; it has not published a traffic share, an API type value or a statement about the query dimension. The actions in the baseline routine use only the first set, and the misreadings each borrow something from the second.

The baseline routine asks for an export today and every Monday, and the export is only comparable later if every file carries the same fields. For each export: the date range, with the filename carrying it; the report it came from, Performance for Search results or Generative AI features, because the same Lens search can land in either; the search type, text-based and multimodal side by side; and the three tables, pages, countries and devices, for that range. Add a line recording that the Search Analytics API did not yet offer the type value on the export date, so a later reader knows why the sheet was built by hand and when it can be automated.

An illustrative weekly sheet for an image-heavy retailer reads: week ending on the stated date; Performance report, Web type split text-based and multimodal; pages table showing the product and category pages found under multimodal, with the ones absent from the text-based list marked; countries and devices tables with the mobile share noted, since Lens is a camera feature and Circle to Search is Android; Generative AI features report, the same split; API status, no multimodal type value as of this date. The monthly step joins the multimodal pages to landing-page revenue for the same dates, which is the third question the filter section says Search Console alone cannot answer.

The sheet is also the record that makes the first-week caution enforceable. When someone asks for the visual-search share in week two, the sheet shows one row of history and the pages list, and the honest answer is the list, as the callout above says. In month three the same sheet has twelve rows and the share means something.

Two habits keep the sheet comparable over the months it takes to become useful. Keep the three tables in the same column order every week, with the search type as a column rather than a separate file, so a formula can read text-based and multimodal side by side for any page. And record the date range exactly as Search Console reported it, because a sheet that mixes reported ranges with calendar-week labels can show a step that is an artefact of the export rather than a change in camera search.

When the API gains a multimodal type value, the sheet becomes the specification for the automated version: the same three tables, the same split, the same join to revenue, now scheduled. Until then it is a person on a Monday, which is cheap compared with discovering in the new year that nobody kept the first months of the row’s history.

Take the first export today with the fields above, name the file by its date range, and put the Monday repeat on one person’s calendar. The row is empty of history now; the sheet is how it fills.