Search Console’s Multimodal Filter (September 24, 2026): What Michigan Parts and Equipment Sellers Should Read in Their Lens Data

On September 24, 2026, Google split the “Web” search type in Search Console’s Performance report into two views: Text-based and Multimodal. The Multimodal view counts impressions and clicks from searches that start with an image — Google Lens, Circle to Search on Android, image uploads to Google Search, and Chrome’s right-click “Search this image.”

It landed the same morning Google began rolling out the September 2026 spam update, a day after Search Engine Roundtable flagged unconfirmed ranking volatility. For Michigan businesses selling things people photograph — trailer parts, furnace components, outboard props, industrial fittings — three causes can move your numbers at once, and only one is new data.

What is the Multimodal filter in Google Search Console?

It is a new search-type option that isolates traffic from image-based searches. Google’s definition, as reported by Search Engine Roundtable: “Web-multimodal tracks search results triggered by a query that uses an image, photo, or screenshot.”

You get impressions, clicks, average position, pages, countries, devices, and dates. You do not get queries. Google’s explanation, per Search Engine Journal, is that “multimodal searches mostly use images rather than text, so specific text query data isn’t available.” Google’s John Mueller summarized it on Bluesky: the photo itself stays hidden, but you can see which of your pages showed up and how often. According to Search Engine Land, the multimodal data also appears in the Generative AI features report, which still offers no click or query data.

The filter appears only for properties with multimodal traffic and, per Search Engine Journal, wasn’t in every property on launch day. Coverage hasn’t mentioned API or bulk export support, so confirm before building a dashboard on it.

Why did my Search Console numbers change in late September 2026?

Probably for one of three overlapping reasons, and you need to separate them before reacting. The events are:

1. New data in the Web totals. Mueller confirmed the team checked and that “the data wasn’t previously in the counts.” Multimodal clicks are being added to what you see, not carved out of it.
2. The September 2026 spam update. Google announced it on September 24 as a global, all-language rollout that “may take up to two weeks to complete.” Search Engine Land counts it as the fourth spam update of 2026, following August’s.
3. Unconfirmed volatility on September 23–24. Search Engine Roundtable reported heavy ranking-tool movement that Google has not tied to any named update.

One loose end: SEO consultant Dave Smart posted on Bluesky that he saw “corresponding drops in impressions and clicks starting the 10th on a few properties.” Search Engine Roundtable read that as an artifact of the new breakdown rather than lost traffic, but Google hasn’t explained a September 10 date, so treat any shift that starts there as unresolved rather than as a penalty.

What do the symptoms in your Multimodal data actually mean?

Read the report by pattern, not by total. Here is how common patterns map to causes:

• Web is up, Text-based is flat or down. Your organic search performance did not improve. Multimodal volume is covering a text decline — possibly the spam update, possibly the September 23–24 movement. Report Text-based to leadership until you understand it.
• Multimodal impressions concentrate on a handful of product or part pages. Those are the objects people are holding in their hands and trying to identify. That is a demand signal you cannot get from keyword tools.
• High multimodal impressions, near-zero clicks. Lens is showing your image, but the result card isn’t winning the tap — often because another listing shows a price, stock status, or a closer visual match.
• Multimodal traffic lands on blog posts, not product pages. Your best photos live in the wrong place. The image matched, but the page can’t finish the sale.

How does Google Lens decide which page to show?

Visual similarity gets a page into consideration; the text around the image decides what the result says and whether someone taps it. Google hasn’t published Lens ranking factors, but the logic follows from what the search contains.

A Lens search has no typed words. Google works from the pixels to find indexed images that resemble the photo, then uses each page’s filename, alt text, surrounding copy, and product data to identify the object. Google’s image SEO guidance stresses the same inputs.

That leads to a prediction you can test on your own catalog. People photograph objects as they actually appear: installed, dirty, at an angle. A page showing only the manufacturer’s white-background render ties with every reseller using that render. An original real-world photo labeled with the part number is more likely to be the closest match. Sell the same SKU as 40 competitors with the same stock image, and Lens has no reason to prefer you.

Does multimodal data actually matter for Michigan businesses?

Yes for sellers of physical, fitment-critical products. It matters much less for pure service firms. Google says the data will help you “optimize your visual content.” The real question is whether your customers ever point a camera at something they need to replace.

That describes a large part of the Michigan economy. An aftermarket shop in Warren sells hitch couplers that customers can’t name but can photograph. A marine dealer near Grand Rapids sells props and impellers for customers who are winterizing now. An HVAC supply house in Livonia sells igniters and flame sensors to homeowners doing the first furnace start-up in October. A Sterling Heights distributor sells fittings whose part numbers have worn off. In each case the search is the object itself.

Keep the scale in proportion. Google said in October 2024 that Lens handled nearly 20 billion searches a month, about 20% of them shopping-related. Still, on most small sites multimodal clicks will be a sliver of text clicks, and a page with eight a week swings on noise alone. Judge pages over at least four weeks. An Ann Arbor home-services firm with no catalog can check monthly and move on.

The real financial risk is a reporting error, not a Lens problem. Picture a Rochester Hills parts distributor whose September report shows Web clicks up 6%. The owner holds the content budget flat. Meanwhile Text-based clicks fell 9%, and nobody checked because the headline looked healthy. By the time the decline shows up in quote requests, Q4 is half over.

How should a small team run the multimodal audit this week?

Plan on about three hours, split between a marketer and whoever manages product photos.

1. Re-baseline your reporting. In Performance, set Search type to Web → Text-based and compare the last 28 days with the prior period and last year. Report that line, not total Web, this quarter.
2. Annotate September 10, 23, and 24. Mark them in your reporting sheet so nobody later attributes a shift to the wrong cause.
3. Export the Multimodal Pages tab. Sort by impressions. The top 20 pages are your “photographed products” list.
4. Check each top page’s images. Mark every page whose only images are manufacturer renders. Those are your highest-leverage fixes.
5. Shoot real-world photos for the top 10. A phone and a shop bench are enough: the part installed or in hand. Name the file with the part number; write alt text stating what it is and what it fits.
6. Put price and stock near the image. Keep Product structured data accurate so a Lens result can show the details that win the tap.
7. Move stranded images. If multimodal traffic lands on a blog post, link that image to the matching product page or reuse the photo there.

If your catalog also runs through Merchant Center, the same photo audit pairs well with the product video mapping audit for Michigan catalog sellers. Manufacturers whose key product details are buried in PDFs should also read our breakdown of Google’s PDF drop-off and what it means for spec sheets, because a Lens match can’t send traffic to an image that’s only in a PDF.

What dates matter over the next few weeks?

• September 10, 2026: The start date of the unexplained shifts Dave Smart reported. Unconfirmed; annotate it and watch it.
• September 23–24, 2026: Unconfirmed ranking volatility, reported by Search Engine Roundtable.
• September 24, 2026: The Multimodal filter launches, and the September 2026 spam update begins.
• About October 8, 2026: The end of the two-week window Google gave for the spam update. Don’t judge spam-update impact before then.
• Late October 2026: Four weeks of multimodal data — enough to prioritize new photography before holiday and winter-parts demand peaks.

Frequently asked questions about Search Console’s Multimodal filter

Can I see what photo someone searched with?

No. Google shows neither the image nor any text query for multimodal searches. You see which pages appeared, impressions, clicks, and average position, by country, device, and date. Intent has to be inferred from the landing page, which makes the Pages tab the most useful part of the report. Export it regularly while data builds.

Did adding multimodal data lower my existing Web numbers?

According to John Mueller, no: the multimodal data “wasn’t previously in the counts,” so it is being added. That means total Web figures can rise without any change in your text rankings. If you see declines starting around September 10, don’t assume they’re a penalty. That pattern was reported by a practitioner, not explained by Google.

Is this the same as the Image search type?

No. The Image search type covers Google Images results, where the search usually starts with typed text. Multimodal sits under Web and covers searches where an image is the query itself, via Lens, Circle to Search, uploads, or Chrome’s “Search this image.” Report them separately: one measures browsing, the other object identification.

Should service businesses without products care?

Only a little. A dental group or law firm will see little multimodal traffic. The exception is trades that install identifiable equipment, such as HVAC, plumbing, and appliance repair, where homeowners photograph the unit or failed part. For them, a model-specific page with a real photo can capture searches competitors don’t even know are happening.