The search console multimodal filter will not tell you which image on a page earned a click. It tells you which page did. That sounds like a downgrade, and most of the coverage published since launch stops there, usually with a screenshot of where to click. The useful question comes next: can you use this report to decide whether the pictures you commission are worth paying for, and what do you change about the next ten briefs if the answer is no.

That is what this piece is about. Not where the menu is. What the numbers justify.

What the multimodal filter actually reports

Google added a Web: multimodal search type to the Performance report in Search Console on 24 September 2026. You will find it under Search type, in the same list as Web, Image, Video and News.

It isolates results that were triggered by a search containing an image: a Google Lens capture, a Circle to Search selection, or a query where someone combined a photo or screenshot with a few words of text. Someone points a phone at a boiler control panel and asks what the fault code means. Someone screenshots a spreadsheet error and circles it. If your page turned up in that result and got a click, it lands here.

The distinction that trips people up on first look: this is not the Image search type. Image covers your pictures appearing in the image tab, where the image itself is the result. Multimodal covers a normal web result that was served in response to a visual input. Same report, different mechanics, counted separately. We will come back to why adding them together is a mistake.

One more thing before you open it. Data only exists from the launch date forward. There is no historical baseline, no like-for-like comparison with last autumn, and anyone showing you a multimodal growth percentage at the moment is comparing against a period in which the metric did not exist. Build your own baseline instead, starting now.

Find your multimodal numbers in under ten minutes

Open Performance, then Search results. Click Search type at the top of the report. Choose Web: multimodal, and here is the small decision that changes the value of everything after it: pick Compare rather than Filter, so multimodal sits next to total web traffic instead of replacing it. A figure of 40 clicks means something different on a page that gets 200 web clicks a month than on one that gets 8,000.

Then work through four moves:

• Add the Pages dimension and sort by clicks. You are looking for URLs that attract visually triggered visits, not a site total. The site total is close to meaningless at SME scale.

• Export to Sheets and log the figure weekly. One snapshot tells you nothing. A twelve-week line tells you whether a change in how you brief images did anything.

• Check the Devices breakdown. Multimodal search is camera-led, so it skews heavily to mobile. If your winning pages are slow or cramped on a phone, that is the first fix, ahead of anything to do with the images themselves.

• Glance at Search appearance. It occasionally explains which result formats the clicks came through, which matters if you are already tracking other mobile ranking behaviour and SERP features separately.

Google's own Search Console Performance report documentation is worth a read for how search types are defined and counted, particularly if you intend to put these numbers in front of someone who will question them.

Proving a specific blog image earned the traffic

You cannot prove it outright. You can build a case that is good enough to act on, which is a different standard and the one that actually matters for a content budget.

Start narrow. Filter to a single page, apply Web: multimodal, set a 28-day window, and put that figure against the same page's standard web clicks for the same period. Then run the Image search type for the same URL over the same window. Three numbers, one page. The pattern tells you whether that post earns visual attention through the image tab, through camera-led search, or neither.

What the data cannot tell you

Two hard limits. The report will not identify which image on the page was matched, so a post with seven pictures gives you one number and no attribution. And query data is largely absent, because the input was a photograph, not a string: there is often no keyword to report, since nobody typed one.

That second limit changes the method rather than ending it. You attribute at page level and corroborate elsewhere, instead of hunting for queries that were never entered. It is the same adjustment publishers had to make when reading AI Mode data in Search Console: less query detail, more inference from page behaviour.

Corroborating evidence worth gathering

Three things firm up the case. Landing page behaviour in GA4 for the same URL and date range, checking whether mobile sessions to that page moved in the same direction. The Image search type comparison above. And the simplest test of all: open the page and ask whether it contains a visual that exists nowhere else on the internet. If every picture on it came from a stock library, the multimodal clicks almost certainly came from the text, and the image did not earn anything.

A worked judgement call

Say a how-to post carrying a hand-drawn process diagram logs somewhere in the low tens of multimodal clicks over a month, while nine neighbouring posts with stock photography log none. That is a small number and you should treat it as directional. It is still a brief instruction. If a diagram you paid an hour for is the only page in that cluster pulling camera-led traffic, commissioning real diagrams for the next ten articles is a cheap bet with a visible mechanism behind it. Not a guarantee. A reasonable bet.

Why most blog images never show up in this report

Because they are not distinctive enough to match on. The multimodal filter rewards visual uniqueness, not volume. A hero image of a smiling person at a laptop sits on thousands of other pages, so there is nothing for a visual search to anchor to, and no reason to serve your page over anyone else's. Teams generating images from their own palette, logo and font tend to see visually triggered traffic before teams buying from free stock libraries, for exactly that reason.

The other common causes are duller and easier to fix:

• Missing or decorative alt text. "image1"and empty alt attributes describe nothing. Write what is in the picture, plainly.

• Unreadable file names. DSC_4471.jpg tells Google nothing. vaillant-ecotec-fault-f28-panel.jpg tells it quite a lot.

• Lazy-loading implemented outside parsable markup. If the image only exists after a script runs, assume it may not be seen.

• Over-compression. A screenshot squeezed until the text is mush cannot be resolved by Lens. Legibility beats file size for anything containing words, numbers or a label.

The formats that give visual search something to grip are consistent across clients: product shots, labelled diagrams, charts with real figures, signage, packaging, premises photography, and before-and-after job photographs. A UK installer's blog is a good illustration. The only genuinely unique visuals on most of them are the job photos and the occasional labelled kit diagram, and those are the pages most likely to surface when someone photographs the same unit on their own wall.

Turning multimodal data into editorial decisions

The honest use of this report is editorial, not technical. Rank your pages by multimodal clicks, then look for the shared trait across the top handful. Is it screenshots? Diagrams? Photographs of physical kit? That trait goes into the next batch of briefs.

Pair it with competitor work. If rivals own the text result for a term and the gap looks expensive to close, a distinctive visual is often the cheaper route in, because almost nobody is competing on that axis yet. Our approach to finding content gaps through competitor keyword tracking works the same way: find the opening nobody has bothered to defend.

Two habits make the difference. Keep images on-brand, using a real palette, logo and typeface, so a Lens match reads as your business rather than as a free stock library. And write the image brief inside the article brief, before drafting starts, rather than bolting a picture on at publish time because the template has a slot for one. We plan briefs against the live top 10 and generate images from the client's actual brand assets for precisely this reason, which is part of how we write and research every piece.

The traps that make this report look worse than it is

Small numbers invite overreaction. Treat anything under roughly 20 to 30 clicks a month as directional only. That is our working rule, not a Google threshold, and it exists to stop a team rebuilding its image strategy around eleven clicks.

Do not retro-fit alt text across an entire archive in one afternoon. Site-wide overnight changes look like manipulation, and more practically, they teach you nothing about cause, because you cannot tell which edit did what. Fix a defined set of ten or twenty pages, wait a full reporting cycle, then compare. Gradual beats spiky, in image work as in publishing cadence generally.

Never add multimodal and Image clicks together. They overlap conceptually, they are counted separately, and summing them produces a number that means nothing to anyone.

Finally, decide the cadence now. Monthly is enough for most SMEs, with a weekly export running quietly in the background to build the trend line. Otherwise the search console multimodal filter becomes one more panel nobody opens, which is a shame, because for once the data points at something you can actually change: the pictures.

Frequently Asked Questions

What does the multimodal filter in Search Console actually measure?

It measures clicks and impressions for your pages in web results that were triggered by a search containing an image, such as Google Lens, Circle to Search, or a photo combined with a text question. It is a search type filter in the Performance report, added on 24 September 2026. The numbers are page-level, not image-level.

How is Web: multimodal different from the Image search type?

Image covers your pictures appearing as results in the image tab, where the image is the thing being shown. Web: multimodal covers ordinary web results returned in response to a visual input. Different triggers, counted separately, so compare them against each other rather than adding them up.

Why are there no queries shown for my multimodal clicks?

Because the input was often an image rather than typed text, so there is frequently no query string to report. Attribute at page level instead and corroborate with landing page behaviour in GA4 and the Image search type for the same URL.

Can I tell which image on a page triggered a Lens result?

No. Search Console does not report image-level attribution for multimodal results at the time of writing. The practical workaround is to publish one genuinely distinctive visual per page for a period, then watch whether that page's multimodal clicks move.

How often should I check the multimodal filter in Search Console?

Export weekly so you accumulate a baseline, but only review and act monthly. Volumes at SME scale are small enough that weekly interpretation produces noise, and since data only exists from launch, the trend line is the only part worth reading for now. If you want a second opinion on what your current pages are earning, our free content marketing snapshot is a reasonable starting point.