Search Console now reports Google Lens traffic separately. Here's how to read it

On 24 September 2026 Google split the Web search type in Search Console into two: Text-based and Multimodal. Two things about that date matter more than the feature itself. The multimodal numbers are newly counted rather than moved across from existing Web totals, so your totals go up without anything improving. And Google started the September 2026 spam update the same morning, so a second, unrelated force began moving the same charts on the same day.

What Google actually changed

Google announced web multimodal Search performance reporting on its Search Central blog and began rolling it out globally on 24 September 2026. In the Performance report, the search type filter's Web option now opens into two choices rather than one:

The same multimodal search type also appears in the Generative AI features report, which is where it becomes interesting for anyone tracking AI surfaces: a Lens search that resolves into an AI-generated answer is a different discovery path from a typed query, and until now neither path was separable from the other.

Note what multimodal is not. It is not the Image search type, which has existed for years and reports your images appearing in Google Images. Multimodal reports web results that a user reached by searching with an image. The direction is reversed, and conflating the two is the most common mistake we have seen in the first week of commentary.

The part most coverage has skipped: this data is additive

If the split were a simple partition, your Web total on 23 September and your Web total on 25 September would be comparable, and the only new information would be the ratio between the two halves. That is not what happened. Google has indicated that multimodal impressions and clicks were not previously included in Search Console's counts at all, as reported by Search Engine Roundtable from Google's own comments. The multimodal row is new data arriving, not existing data being relabelled.

Which means a site with meaningful image-led search exposure will see impressions, and to a lesser extent clicks, step up in the last week of September 2026 for reasons that have nothing to do with how it is performing. If you report monthly numbers to anyone, September-to-October will carry an artificial lift, and if you do not name it now, you will be asked to explain the "drop" when it does not repeat next quarter.

The 24 September confound, and how to unpick it

Here is the problem nobody is naming. Google's September 2026 spam update started rolling out on 24 September 2026 at 9:15am PDT, and Google said it could take up to two weeks to complete. It moved substantially over the weekend of 25 to 27 September, cooled, then produced what looked like a second wave around 30 September. As this publishes it has not been confirmed as complete.

So from 24 September onward, two independent forces act on your Performance report: a reporting change that can only add, and a ranking event that can add or subtract. Looking at combined Web totals across that boundary tells you almost nothing, and it fails in both directions. A site that lost ground to the spam update can be flattered back to flat by newly counted multimodal impressions. A site that gained can have its gain understated if its multimodal exposure is negligible. Either way, the chart you are looking at is a sum of two things.

The separation is straightforward once you know to do it:

1. Read ranking movement in Text-based only. Set the search type filter to Web, then Text-based, and compare against your pre-24 September baseline. Text-based is methodologically continuous with the old Web definition, so this is where spam update impact lives.
2. Read the reporting change in Multimodal only. Viewed alone, this series is the newly counted traffic. It has no "before", so do not compare it to anything; just size it against your text-based volume to decide whether it deserves work.
3. Treat combined Web as unusable across 24 September. It is fine going forward, and fine before. It is not comparable across the boundary, and no amount of date-range selection fixes that.
4. Check clicks, not just impressions. Multimodal impressions can be substantial while clicks stay small, because a Lens result that answers visually may never need a click. An impressions-only read overstates what this surface is worth to you.
5. Wait for the spam update to be confirmed complete before concluding anything about rankings. Phase-two movement around 30 September means mid-rollout readings have been reversed before.

Diagnosing multimodal traffic with no queries

The missing Queries dimension is the real analytical problem, and it is not a bug: a user who points a camera at an object has not typed anything, so there is no query string to report. Google has confirmed the dimension is unavailable for this search type.

You are not blind, though. You still have pages, countries, devices and dates, and the page that Google chose to show is itself a strong statement about what it thought the image meant. That gives you a workable substitute: infer intent from which of your pages Google selected, then act on the inference.

The pattern is the signal. You cannot know the query, but you can know which of your pages Google thought was the answer to an image, and that is close enough to act on.

Is this worth any work? Size it first

Visual search is genuinely large in absolute terms. Google disclosed in October 2024 that Lens was handling close to 20 billion visual searches a month, roughly a fifth of them shopping related, up from the 12 billion a month it reported at I/O 2023. Those remain the most recent figures Google has attached a number to, which is worth saying plainly: anything more current circulating as a percentage of all Google queries is third-party estimation, not disclosure.

Absolute scale is also the wrong question for your site. The only number that matters is your own multimodal share, which you now have. Our rule of thumb for the first read:

Give it two to four weeks of data before you act. The series started in late September, it coincided with a rolling spam update, and the first fortnight of any new Search Console report tends to settle.

If it is material, fix these in this order

There is no "multimodal optimisation" discipline to learn, and anyone selling one this month is ahead of the evidence. What makes a page the destination for an image-led search is the same work that has always made images findable, prioritised by the pages the report actually names.

1. Confirm your images are crawlable and indexable. Images behind lazy-loading that never resolves without interaction, hidden behind scripts, or blocked in robots.txt cannot be matched. This is the single most common cause of the "spread thin, low clicks" pattern.
2. Serve images at a resolution a matching system can use. Aggressive compression that looks acceptable to a human can strip the detail a matcher needs. Our broader guidance on image optimisation still applies; just stop treating the smallest acceptable file as automatically the right one on product pages.
3. Use unique photography on pages you want selected. Manufacturer stock images put you in a pool with every other retailer carrying the item. If Google must choose one page for an identified object, identical images give it no reason to choose yours.
4. Surround the image with text that names the object precisely. Model name, variant, material, size. Alt text matters here, and so does the visible copy next to the image; our alt text guide covers the mechanics.
5. Keep product data accurate in structured data. Price, availability and variants. A shopping-intent visual search that lands on a page with stale availability converts worse than no visit at all.
6. Check mobile performance on the landing pages named in the report. Multimodal traffic skews mobile by construction, because the camera is on the phone. A page that is fine on desktop and slow on a mid-range Android is losing the clicks it earned.

None of that is new work. What is new is knowing which pages to do it to, which is the actual value of the report. If image-led discovery turns out to be doing real acquisition work for you, that belongs in the brief for the build rather than in a clean-up pass eighteen months later, which is how we treat it on website design and development projects.

Who should ignore this entirely

Most B2B service businesses, software companies and professional practices. If nothing you sell exists as a physical object someone could point a camera at, your multimodal row will be close to noise, and the only thing you need from this change is the annotation on your totals so nobody misreads September. The teams that should care are the ones selling things that exist in the world: retail and e-commerce product pages, parts and components, fashion, food and drink, furniture, plants, and any service diagnosed from a photograph of a problem.

The wider point is about how Google's reporting now works. Search Console has added AI Mode traffic, generative AI reports and now multimodal within eighteen months, each time changing what a number means without changing its name. If you benchmark against your own history, which you should, then every one of those additions needs a dated annotation. We have written separately about optimising for AI search surfaces, and the measurement discipline is identical: know what the metric counted before, and when that changed.

Frequently asked questions

What is the multimodal search type in Google Search Console?

Multimodal is a search type filter under Web in the Performance report, introduced by Google on 24 September 2026. It reports web search results where an image formed part of the search, covering Google Lens, Circle to Search on Android, image uploads to Google Search and Chrome's right-click "Search this image". The Web option in the search type filter now splits into Text-based and Multimodal.

Does multimodal data increase my Search Console totals?

Yes. Google has indicated that multimodal impressions and clicks were not previously included in Search Console counts, so this is additional data rather than a reallocation of existing Web totals. Sites with meaningful image-led search traffic should expect totals to step up from late September 2026, and that step is a reporting change, not a performance improvement.

Why are there no queries for multimodal traffic?

Because most multimodal searches start with an image rather than typed text, there is often no text query to report. Google has confirmed the Queries dimension is unavailable when the multimodal search type is selected. Pages, countries, devices and dates remain available, as does export.

How do I separate the multimodal reporting change from the September 2026 spam update?

Filter to Web then Text-based and compare that series with your pre-24 September baseline. Text-based is the like-for-like continuation of what Web used to mean, so ranking movement from the September 2026 spam update, which began on 24 September, shows up there. Multimodal viewed on its own shows the newly counted traffic. Comparing combined Web totals across 24 September mixes the two and will mislead you in both directions.

How big is visual search?

Google disclosed in October 2024 that Lens handled close to 20 billion visual searches a month, roughly a fifth of them shopping related, up from 12 billion a month reported at I/O 2023. Those are the most recent figures Google has put a number to. Your own multimodal row in Search Console is the only reliable guide to what share of your traffic it represents.

What should I change on my site because of multimodal search?

Nothing, until you have read your own multimodal numbers for two to four weeks. If multimodal is a negligible share of impressions, treat it as a reporting curiosity. If it is material, the work is conventional image and page quality work prioritised by the pages the report names: images that are crawlable, indexable, high resolution and unique, surrounded by text that identifies the product or object, with product data where it applies.

The takeaway

Do three things this week. Annotate 24 September 2026 in whatever you report search performance from, because your totals changed definition on that date and nobody will remember why in January. Switch your ranking analysis to the Text-based filter, so the September spam update can be read without newly counted multimodal impressions papering over it. And size your own multimodal share before you let anyone sell you a visual search strategy: for most service businesses it will be close to nothing, and for retailers of physical goods it may be the most under-managed acquisition path on the site. The report is genuinely useful. It is useful because it names the pages, not because it is new.