Your Search Console impressions just jumped. Here is what the multimodal filter changed

If your Search Console impressions stepped up in the last week of September without a matching rise in clicks, nothing happened to your rankings. Google added a category of search it had never reported before, and the new number landed in your existing totals.

What Google actually added

The Search type filter in the Performance report used to offer Web, Image, Video and News. Web has now been split into two options: Web: text-based and Web: multimodal. Selecting multimodal filters the report down to search results that were triggered by an image rather than typed words.

Four entry points count as multimodal:

1. Google Lens, a photo taken in the Google app or the Lens camera.
2. Circle to Search on Android, circling or highlighting something already on screen.
3. An image uploaded to Google Search directly.
4. Chrome's "Search this image" right-click action on any image on any page.

The filter works in two places. In Performance → Search results you get clicks, impressions, CTR and average position. In Performance → Generative AI you get impressions from AI Overviews and AI Mode where an image formed part of the query. The Pages, Countries, Devices and Dates tabs all work normally underneath.

One dimension does not: Queries. Google's dimensions documentation explains why in plain terms, "Because multimodal searches mostly use images rather than text, specific text query data isn't available for this traffic." There is no keyword list for a photograph.

The detail most coverage skipped: this is new data, not moved data

This is the part that changes what you can and cannot say about your own numbers, and it turns on one sentence. Asked whether the multimodal figures were being carved out of the existing Web total, John Mueller confirmed that "the data wasn't previously in the counts, the report has new data."

Read that as an accounting change rather than a search change. Before 24 September 2026, a person pointing their camera at your product and landing on your page was, as far as Search Console was concerned, invisible. Now they are counted. Your site did not gain visibility on 24 September; your report gained a column.

Which means the number you compare against matters enormously for the next twelve months. The table below is the practical version.

Practical consequence: Web: text-based becomes your reporting default and Web (all) becomes your diagnostic view. If you send a monthly search report to anyone, change the saved filter this week, before someone celebrates a 6% impressions rise that is entirely a definition change. Clean year-on-year comparisons on the unfiltered Web total do not return until late September 2027, and Search Console only retains 16 months of data, so the boundary will sit inside your comparison window for the whole of the coming year.

One thing Google has not published is how far back the multimodal series is backfilled. Do not assume; test it in your own property. Filter to Web: multimodal, set the date range to the maximum 16 months, and look for the earliest date with a non-zero impression. That date is the real start of your usable multimodal history, and everything before it is an absence of data rather than an absence of activity.

No queries. So what do you actually do with the report?

A performance report without queries feels useless, and most of the coverage stopped at pointing out the gap. It is workable, because the other four dimensions still tell you something when you read them together rather than separately. The method below substitutes structure for keywords.

1. Start with Pages, sorted by multimodal impressions. This is the only list that matters. A page here surfaced because something visual on it, or something visually similar to it, matched a camera query.
2. Ask what is photographable on each page. A product, a label, a plant, a part number, a dish, a symbol on a dashboard, a building. If you cannot name the object, the match is probably incidental and you can stop.
3. Check Devices. Multimodal should skew heavily to mobile; Lens and Circle to Search are phone behaviours. A desktop-weighted profile usually means Chrome's "Search this image", a different user, often a researcher or a competitor, not a buyer.
4. Check Countries. Circle to Search is Android, and Android share varies enormously by market. A country that over-indexes in multimodal but not in text is telling you where the visual behaviour is concentrated.
5. Compare position within the type. A page at position 4 for text and position 22 for multimodal is not one ranking; it is two, and the second one is addressable with images rather than copy.
6. Cross-check the Generative AI report. Multimodal impressions there mean your page was pulled into an AI Overview or AI Mode answer off the back of a photograph. That is the clearest signal available that the page's images, not its text, are doing the work.

None of this gives you a keyword. All of it gives you a page, an object and a device, which is enough to decide whether to act.

Which pages get multimodal impressions, and what to check on them

The pattern is fairly consistent: multimodal traffic follows physical objects and visual ambiguity. It rewards sites that document things precisely and ignores sites that write about concepts.

The last row matters as much as the others. Multimodal impressions on a text-heavy article are typically noise, and optimising for them is how a small reporting change turns into a wasted quarter.

A visual-search readiness checklist, mapped to the four entry points

Because the four surfaces work differently, "optimise for visual search" is not one job. This is the version we would actually run, cheapest item first.

Notice how much of this is ordinary discipline rather than a new specialism. Clean product photography, specifications as text, honest structured data and fast mobile pages were already the right answer for a well-built site. What the new report does is let you see, for the first time, one of the returns on them.

When is multimodal visibility worth acting on?

Here is the threshold we would use, because the honest answer for most sites is "look once, then ignore it".

1. Under 1% of total impressions and no clicks. Record the baseline, change your reporting filter, do nothing else. You have a reporting artefact, not a channel.
2. 1% to 5% of impressions, clicks present but CTR well below your text CTR. You are being shown and not chosen. Work the top five pages only: isolated image, specs as text, answer-first opening line.
3. Over 5% of impressions, or any meaningful click volume on commercial pages. Treat it as a channel. Audit product imagery across the range, fix variant images, and check whether the multimodal position gap against text points at specific categories.
4. Multimodal impressions in the Generative AI report on pages that convert. Highest priority regardless of share. Something you photographed is being used to answer a question in AI Mode, and the page is one edit away from being the cited source rather than a listed one.

What this does not mean

Three claims to resist, because they are already circulating.

It is not a ranking change. Nothing about how Google ranks pages changed on 24 September 2026. A reporting filter was added. Multimodal search itself has been running for years; only the measurement is new.

It is not a reason to rebuild your site around images. For most B2B and service businesses, multimodal will be a rounding error, because nobody photographs a consultancy. The businesses with real exposure sell, stock, service or grow physical things.

It does not replace the AI visibility question. Multimodal reporting tells you when an image started a search. It says nothing about whether text answer engines cite you, which is a separate problem with separate mechanics. Both matter; they are not the same report.

What it does do is close a measurement gap that had been open for years, and hand anyone selling a physical product a list of pages where a photograph already reaches them. That list is worth an hour of your time this week, and for a good number of sites, nothing after that.

Frequently asked questions

What is the Web: multimodal search type in Search Console?

It is a Search type filter Google added to the Search Console Performance report on 24 September 2026. It isolates searches that were triggered by an image rather than typed text: Google Lens, Circle to Search on Android, an image uploaded to Google Search, and Chrome's "Search this image" action. The old Web option is now split into Web: text-based and Web: multimodal.

Why did my Search Console impressions jump in late September 2026?

Most likely because multimodal impressions were added to your totals rather than moved into a new bucket. John Mueller confirmed that "the data wasn't previously in the counts, the report has new data." Camera and screenshot searches were never included in web totals before 24 September 2026, so any site with visual search visibility shows a step change on that date with no underlying change in performance.

Why are there no queries for multimodal searches?

Because there is usually no text to report. Google's dimensions documentation states that "Because multimodal searches mostly use images rather than text, specific text query data isn't available for this traffic." You still get pages, countries, devices, dates, clicks, impressions, CTR and average position, but the Queries dimension is unavailable when the multimodal search type is selected.

Does the multimodal filter mean Google changed its rankings?

No. Nothing about ranking changed on 24 September 2026. Multimodal search has existed for years through Lens and Circle to Search; what changed is that Search Console now reports on it. Treat it as an accounting change to your measurement, not a search change to your site.

How should I compare year-on-year Search Console data now?

Apply the Web: text-based filter to both periods before comparing anything. The unfiltered Web total now contains a data category that earlier periods did not, so year-on-year impressions, CTR and average position on the full Web total are not comparable across 24 September 2026. Because Search Console retains 16 months of data, clean unfiltered comparisons do not return until late September 2027.

Which kinds of website actually get multimodal search traffic?

Sites documenting physical, photographable things: products and variants, spare parts and components, menus and dishes, plants, materials and finishes, signage and error symbols, and buildings or venues. Sites selling intangible services usually see negligible multimodal volume, because there is nothing for a customer to point a camera at. If it is under about 1% of your impressions with no clicks, record the baseline and move on.

The takeaway

A reporting filter is an unexciting thing to write about and an easy thing to get wrong, because the first symptom is a number going up. Do two things this week. Change your standard search report to the Web: text-based filter so your trend lines stay honest through 2027, and open the Pages tab under Web: multimodal once to see which of your pages a camera already finds. For most service businesses that second list will be short and you can leave it alone. If you sell, stock or service physical things, it is the cheapest visibility audit available to you right now: clean isolated product images, specifications as text rather than baked into pictures, and the identification question answered in the first line under the photo.