Search Console now separates camera searches. What to do with the data

On 24 September 2026 Google split the Web search type in Search Console into text-based and multimodal. For most sites this will look like a new filter that returns a small number, which is roughly the right reaction. The part worth your attention is quieter: a figure you have been reporting for years just turned out to be two figures, and the API you pull it through still cannot tell them apart.

What Google actually shipped

Google began rolling out a new search type globally on 24 September 2026. Where the Performance report previously offered Web, Image, Video and News, Web is now divided into text-based and multimodal. Multimodal means the search started with an image rather than typed words, and Google lists four surfaces that feed it: Google Lens, Circle to Search on Android, an image uploaded directly to Google Search, and Chrome's right-click "Search this image" action.

The filter appears in two places. In the Search results performance report you get impressions and clicks for multimodal traffic. In the Generative AI features report, which Google introduced earlier in 2026 for AI Overviews and AI Mode, you can apply the same filter to see impressions from image-led searches that ended up in an AI surface.

One dimension is missing, deliberately. There are no queries for multimodal traffic, because there was no query, somebody pointed a camera at an object. Google has switched the Queries dimension off for this search type. You still get Pages, Countries, Devices and Dates.

The part nobody is mentioning: your API still blends them

This is the detail that matters most if you report to clients or run dashboards, and it is easy to miss because the UI change is so visible.

The Search Analytics API documents its report types as web, image, video, news, Discover and Google News. There is no multimodal value. So an API request for type: web returns text-based and multimodal added together, exactly as it did before 24 September. Every Looker Studio dashboard, every rank tracker that pulls Search Console, and every automated client report built on that API is still showing you the blended number while the interface shows you two.

That produces a specific and avoidable failure: somebody opens the UI, filters to text-based, sees a figure lower than the dashboard, and spends an afternoon looking for a tracking bug. The difference is the multimodal slice. Take a manual export of both search types now, dated, and keep it as the baseline your dashboards will eventually need to reconcile against.

How to read multimodal data when there are no queries

Losing the Queries dimension sounds crippling, and for keyword research it is. But a camera search carries its intent in a different place: the page Google chose to answer it. If Lens matched a photograph to your product page, the implied question was not a mystery. It was one of four things.

This is the framework we use to turn a page list into a decision, in the absence of query data.

Two things follow from that table. First, the fix for almost every row is something you would want anyway: clearer above-the-fold answers, accurate structured data, real photography. Multimodal search rewards the same discipline as conversion work, which is unusual for an SEO signal. Second, the homepage row is a genuine "do nothing", a burst of multimodal impressions on a homepage is often Google failing to match rather than an opportunity.

Does this apply to your business at all?

Most of what has been written about the new filter assumes the answer is yes. For a large share of businesses it is not, and finding that out in ten minutes is a better outcome than a programme nobody needed. The variable is simple: does a physical object or place stand in for the query?

The B2B row is the one most often skipped and the one we would look at first. When a maintenance engineer photographs a nameplate to find a replacement part, that is a high-intent commercial query with no keyword attached to it, and the page that wins it is usually the one that shows the same part photographed under normal light with the model number written out as text.

What "optimising for camera search" actually means

There is no separate ranking system to game here, and anyone selling you a multimodal SEO package is ahead of the evidence. What exists is a set of prerequisites for being matchable at all, most of which are ordinary technical hygiene that a lot of sites skip.

1. Use real photography of the real thing. Stock imagery is, by definition, shared with everyone who licensed it. A visual match to a generic image is a match to nothing in particular.
2. Serve images as crawlable files. An image applied as a CSS background or injected by JavaScript after load is not reliably available as an image. This is the single most common reason a product catalogue is invisible to image-led search.
3. Shoot more than one angle. A camera searcher photographs an object from wherever they happen to be standing, which is rarely the marketing three-quarter view.
4. Write filenames and alt text that describe the object. Not for the keyword, but because they are the only text attached to the file itself.
5. Keep the files fast. Modern formats and sensible dimensions, per our image optimisation guide. A slow page that wins the match still loses the visit.
6. Put structured data on the page the image lives on. Product, Offer or LocalBusiness markup that agrees with the visible content, which is also what schema markup is for in ordinary search.
7. Answer the four questions above the fold. What it is, what it costs, whether you have it, and how to get it.

Notice that none of this is new work invented by a September 2026 announcement. That is the honest read on this change: it did not create a new channel, it gave you a way to measure one that was already there and previously hidden inside a larger number.

A measurement routine that fits in fifteen minutes a month

Because there are no queries and no published benchmarks for what good looks like yet, the only useful comparison is against yourself. Establish that first, then watch it.

1. Baseline. Performance report, last three months, export once filtered to text-based and once to multimodal. Record multimodal clicks as a percentage of the two combined.
2. Segment by device. Multimodal is a phone behaviour, so a desktop-heavy multimodal profile is a signal something unusual is happening, usually Chrome's "Search this image".
3. Take the top ten pages and classify each against the intent table above. That list is your work queue, in order.
4. Check the Generative AI report with the same filter, to see how much of your image-led visibility is being absorbed into AI Overviews and AI Mode rather than sent to you as a click. Note that the generative report has its own shorter history and, in its current version, does not include clicks.
5. Re-check monthly, act quarterly. One month of a new report is not a trend. Two consecutive months of the multimodal share rising on commercial pages is.

The thing to be sceptical about

A new Search Console filter reliably produces a wave of content declaring a new discipline, and this one arrived in the same month as the September 2026 spam update, so attribution is going to be messy for a few weeks. Keep two cautions in mind. Multimodal impressions are not necessarily incremental, some of them are searches that would previously have been typed, now performed with a camera because the camera was faster. And an impression from a Lens match is not a commercial signal on its own; somebody identifying a houseplant is not a buyer. Until you can see your own click-through and conversion behaviour on that segment, treat the volume as interesting rather than valuable, and let the pages that actually convert decide where the effort goes.

Frequently asked questions

What is the multimodal search type in Google Search Console?

It is a search type filter Google began rolling out globally on 24 September 2026 that separates image-led searches from typed ones in the Performance report. Multimodal covers Google Lens, Circle to Search on Android, images uploaded to Google Search and Chrome's "Search this image" action. Web is now split into text-based and multimodal rather than reported as a single figure.

Why does Search Console show no queries for multimodal search?

Because the input was an image rather than a string of text, so there is no query to report. Google has switched the Queries dimension off for this search type. Pages, countries, devices and dates are all still available, which is enough to work with even though it changes how you diagnose the traffic.

Does my old Search Console data change now that Web is split?

Your totals do not change, but the composition of the Web figure you have been reporting is now visible for the first time. Anything you previously described as web search traffic already included image-led searches. The split does not retroactively remove them, it just lets you see the share, so a flat Web line can now be decomposed into two lines moving in different directions.

Can I pull multimodal data through the Search Console API?

Not as a separate type at the time of writing. The Search Analytics API documents web, image, video, news, Discover and Google News as report types, with no multimodal value, so an API request for type: web returns text-based and multimodal combined. Any dashboard, rank tracker or client report built on that API is still blending the two, which is a reason to take a manual export as your baseline.

How do I optimise a page for Google Lens and camera search?

Treat the image as the entry point rather than a decoration. In practice that means original photography of the actual product or place rather than stock, images large enough to be matched and crawlable as files rather than CSS backgrounds, descriptive alt text and filenames, product or local business structured data on the same page, and the answer a camera searcher wants, what is this, what does it cost, can I buy it, where is it, visible near the image rather than three scrolls down.

Which businesses get the most multimodal search traffic?

Anything where a physical object or place is the query: retail and e-commerce with visually distinctive products, home and interiors, fashion, plants and hardware, car parts, food, and local businesses with photographed storefronts. Service businesses with no physical artefact typically see very little, which is a legitimate reason to check the report once, confirm the number is small and move on rather than build a programme around it.

Is multimodal search traffic worth optimising for?

Check your own share before deciding. Filter the Performance report to the multimodal search type for the last three months, compare the click total with your text-based total, and look at which pages appear. If multimodal is a rounding error against text, it is a monitoring task. If it is a visible share on your product or location pages, the work it implies is image quality, structured data and above-the-fold answers, all of which pay off in ordinary search too.

The takeaway

The useful version of this change is not "optimise for Lens". It is that a number you have reported for years was a composite, you can now see the parts, and the tooling most people report through cannot. Export both slices this week so you have a baseline. Then classify your top multimodal pages by what a camera searcher was asking, and you will find the work queue is the same one your product pages already needed: real photographs, accurate structured data, and the price and the buy button above the fold.