Google Added Multimodal Search Reporting. What Can You Learn From It?

What did Google actually announce here?

Google's Search Central blog published a post in September 2026 titled Announcing web multimodal Search performance reporting in Search Console. It brings reporting for a search type where people search with images rather than typed words. The detail that matters most for anyone who reads reports is that this traffic arrives without query strings.

I want to be careful about what I am telling you. I checked Google's own blog and Google's own Search Console documentation rather than the coverage that followed. The announcement itself is real and it sits on Google's Search Central blog under September 2026. The operational detail that changes how you work is not in the headline at all. It is in the Search Console Help documentation, and it is one sentence long.

That sentence is the whole story. A new search type shows up in your Performance report, and the column you have relied on for two decades is switched off for it. Everything else in this article follows from that one fact.

What does web multimodal search mean in plain terms?

Google's Search Console documentation states that multimodal searches mostly use images rather than text. In practice the search starts with a picture instead of a phrase. Someone points a camera at an object, uploads a screenshot, or selects something already on their screen, and Google returns web results from that image.

Text search asks a person to describe what they want. Image search asks them to show it. Those are very different acts. Describing something requires you to already know the words for it, which is why keyword research has always doubled as a map of how buyers think. Showing something requires nothing. A person who cannot name the part they need can still photograph it.

This is not new behaviour. People have been searching with their cameras for years, and Google Lens, Circle to Search and Google Images all live in that space. What is new is that a slice of it now shows up as its own search type in Search Console. Google has been reshaping the front door of search for a while, and I wrote about one of those changes in the multimodal search box redesign. The reporting is catching up with the interface.

Why does the missing queries dimension matter so much?

Google's documentation says it plainly: "Because multimodal searches mostly use images rather than text, specific text query data isn't available for this traffic. As a result, the queries dimension isn't available when this search type is selected." Every habit you have built around keywords stops working inside this search type.

Think about how much of your reporting runs through the query string. You find striking distance keywords by sorting queries by position. You spot cannibalisation by seeing two pages against one query. You prove content worked by watching a query climb. You brief writers from query lists. You report to a founder using query examples because a query is the closest thing you have to hearing the customer speak.

None of that exists here. You get a page, a count, and nothing about the person. It is closer to reading a footfall counter than reading a search report. That is not a flaw Google will patch later either, at least not in any obvious way, because the query genuinely does not exist. There was no phrase. There was a photograph.

I think this is the most interesting thing to happen to search reporting this year, and almost nobody is treating it that way. The industry has spent twenty years assuming that demand can be described in words, because the only demand anyone could measure was the kind that arrived as words. This search type quietly says otherwise.

How do you analyse a channel when you cannot see the query?

You change the unit of analysis. Stop studying the query and start studying the page. Filter the Performance report to this search type, sort your pages by impressions, and read that list as the answer to a question you never asked: which of my pages does Google treat as a good response to a picture?

That list is genuinely informative. If your product pages appear and your blog does not, Google is reading image-led intent as commercial. If a single comparison page carries most of it, something about that page is doing work the rest of your site is not. If nothing appears at all, that is also a finding, and a cheap one.

The second move is to compare. Put the page list from this search type next to the page list from ordinary web search. Where the two lists agree, you have a page that works for people who can name what they want and for people who can only show it. Where they disagree, you have found a gap worth understanding. I use the same side by side method when reading the AI citations report in Search Console, because the discipline is the same. Two reports of the same site, read against each other, tell you more than either one alone.

The third move is patience. A search type with no query data needs a longer window before it means anything. I would not read a week of it. I would read a quarter, and I would read it inside Search Console rather than piping it into Looker Studio or trying to reconcile it against Google Analytics sessions, because the two systems are counting different things.

Does this mean image SEO suddenly matters again?

Probably, but I am not going to pretend I know the mechanism. Nobody outside Google knows what the ranking inputs are for image-led results, and the announcement did not hand us a checklist. Anyone telling you exactly which image tactic drives this search type is guessing and charging you for the guess.

What I will say is that the safe work is the same work that was always sensible. Your images should be crawlable and should not be locked behind scripts that hide them. Googlebot should not be blocked from your image directories in robots.txt, and they should be described accurately in alt text, because accurate alt text helps real people first. If you are marking images up with Schema.org types such as ImageObject, keep that description honest too. The page around the image should explain the product in words, because the words are what a machine reads when it decides what the picture is of.

None of that is a growth hack. It is hygiene. The difference now is that hygiene has a report attached to it, and a report is how hygiene gets funded inside a company. If you have been unable to get image work prioritised, this search type is the argument you were missing.

What should you check on your own site this week?

Open Search Console, go to the Performance report for Search results, and look at the search type selector. Find out whether this type is showing anything for your property at all. That single check takes about two minutes and it decides whether this article is urgent for you or merely interesting.

If your site runs on Webflow, the check is the same, because this is a Search Console setting rather than a platform feature. If you see nothing, note the date and move on. You have learned that image-led discovery is not currently a channel for you, which is worth knowing and costs nothing to recheck next quarter. Plenty of businesses will sit here. A payroll software company is not going to be photographed.

If you see something, spend the next twenty minutes on the page list rather than the totals. Ask which page types appear, whether they are the pages you would have chosen, and whether the pages that appear are actually good at converting a visitor who arrived with no words in their head. That last one catches people out. A visitor who searched with a camera did not read your headline on the way in. They landed cold.

How does this fit with the other reporting Google shipped this year?

It fits a pattern. Google's Search Central blog published Introducing Search Generative AI performance reports in Search Console in June 2026, and this multimodal reporting arrived in September 2026. Google is splitting one Performance report into several views of several surfaces, and each view answers a narrower question than the last.

For a team, that is a real cost. One report was easy to look at every Monday. Four views of a report is a meeting. Most marketing teams I talk to have not adjusted their reporting rhythm at all, and they are quietly getting a less complete picture every quarter while feeling like nothing changed.

My take is that the number of surfaces will keep going up and the richness of each surface will keep going down. Text search gives you queries, pages, clicks and positions. Generative surfaces give you less. Image-led search gives you less again. The trend line points at reporting that tells you a page is working without telling you why, and teams that only know how to explain wins through keywords are going to struggle to explain anything.

What will this report not tell you?

It will not tell you what the person photographed, what they wanted, or what they did next. It has no query, no intent signal and no conversion. It tells you that a page of yours was shown, and sometimes that it was clicked. Treat it as a presence signal, not as a performance story.

This is where I see people about to make a mistake. A brand new report with a low number in it creates pressure to do something, and doing something usually means inventing a narrative. Somebody will decide the number is low because of alt text and rewrite five hundred alt attributes. Somebody else will decide the number is high because of a product shoot and commission another one. Neither will know.

Restraint is a skill here. The right response to a report you do not yet understand is to record the baseline and wait. That is unglamorous and it is also the only version of this that survives contact with a sceptical finance team six months later.

Is this worth building a dashboard around yet?

Not yet, in my view. A search type with no query dimension and no history behind it does not deserve a dashboard tile on day one. Check it monthly by hand, write the number down, and revisit in a quarter. Build the dashboard when the number is big enough to argue about.

If your site is large enough that manual checks are painful, the sensible middle path is to get the raw data flowing somewhere you can query it later without committing to a visual yet. That is exactly what a bulk export is for, and I walked through the setup in the BigQuery bulk export guide. Collecting data early is cheap. Interpreting it early is expensive.

There is a version of marketing work that confuses visibility of a metric with control of a metric. A dashboard tile makes you feel like you are managing something. For a channel this young, you are not managing it. You are watching it, and watching it in a spreadsheet is honest about that.

What should you do next?

Do three small things. Check whether this search type shows anything for your property. If it does, read the page list rather than the totals and write today's baseline down. Then make sure your images are crawlable and honestly described, because that work pays off regardless of what this report eventually turns into.

The bigger shift is worth sitting with. Search is slowly becoming something people do with a camera as well as a keyboard, and the measurement you get for it is thinner than the measurement you are used to. The teams that handle that well will be the ones who already know how to judge a page on its own merits rather than on the keywords attached to it.

If you want a second pair of eyes on what this looks like for your site, or you are staring at a Search Console property and not sure which of these reports deserves your attention first, reach out. I am always happy to look at a real account and tell you honestly whether there is anything there yet.