GSC Multimodal Filter Explained: Camera Search, AI Overviews, Still No Queries (Sept 2026)

Google just gave you a line for searches that start with a photo. It still will not tell you what the photo showed, or whether anyone clicked the AI answer that cited you.

That is the story. Not “Search Console added a filter.”

If you run B2B SaaS organic or AI search, you already live with impressions without queries on Overviews and AI Mode, and with citation ≠ click when Overview links open AI Mode. September 24 added a second axis: what the searcher supplied (typed text vs an image). The two axes now cross. A screenshot can trigger an Overview. The Gen AI multimodal slice then shows that your page appeared. It still cannot show the prompt, the image, or a click.

This piece is DerivateX’s operator read of the rollout, Mueller’s confirmation that the counts are new, and the 25-minute play we want other agencies to have to copy.

---

What Google shipped

On September 24, 2026, Google’s Search Central blog announced web multimodal Search performance reporting. Trade coverage from Search Engine Roundtable (Barry Schwartz) and PPC Land documented the UI before and after the post (Google posted ~8:40 a.m. ET).

What is now in the Performance Search type filter, nested under Web:

1. Text-based: typed queries.
2. Multimodal: searches that use an image, photo, or screenshot as part of the query.

Four named entry points in Google’s post:

• Google Lens
• Circle to Search on Android
• Image uploads to Google Search
• Chrome right-click “Search this image”

The same split applies to the generative AI features Performance report, not only classic web results. Rollout is global from that day, and Google says you will see metrics if the property actually receives this traffic.

Google’s own help text, as quoted by Search Engine Roundtable:

• “Web-multimodal tracks search results triggered by a query that uses an image, photo, or screenshot. Text-only queries are tracked as Web text-based.”
• “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.”

Mueller on Bluesky, as reported by SE Roundtable: Search Console now reports on multimodal queries “like when you take a photo w/Lens,” you do not see the photo, and you can see which pages showed up and how often. After Dave Smart asked whether these impressions had previously sat under Web, Mueller checked with the team and said the data was not previously in the counts.

What the public package supports:

• A new input-type cut, independent of Image / Video / News (those are result tabs; multimodal is how the search started).
• New data, not a re-bucket of old Web totals, per Google.
• Page, country, device, date, and (on classic Search) clicks and position are described as available in the main report; queries are not.
• On the Gen AI report, you still get impressions without dedicated AI clicks or queries. Multimodal there is impressions by page / country / device / date after an image input.

What it does not prove on its own:

• What the camera or screenshot contained.
• Whether a typed refinement sat next to the image (docs say an image used “as part of the search,” and that searches “mostly” use images).
• Whether Search Live camera+talk sessions, or image prompts inside AI Mode, sit in this bucket. PPC Land notes neither is named.
• Whether the Search Console API or bulk export carries the new type (not mentioned in the announcement).
• How the top-level Web selector now aggregates. If it sums text-based + multimodal, Web totals will step up at the date the new series begins. Google has not published that rule in the blog post.

---

Why this matters more for B2B SaaS than Google’s Lens pitch

Mueller framed the change for image-heavy websites. Retail, travel, food, and product catalogs will feel it first. That is not the DerivateX buyer.

B2B SaaS buyers already search with screenshots:

• A competitor pricing table in a Slack thread
• A G2 or Gartner grid from a deck
• A product UI they want an alternative to
• An error state or integration screen from an incumbent tool
• A chart from an Overview they want to verify

Those inputs are closer to Circle to Search and “Search this image” than to pointing Lens at a lamp. If Google returns your comparison page, docs URL, or glossary inside an Overview after that screenshot, you now have a GSC line for it. You still cannot see the screenshot. You still cannot see the buyer question. You still cannot tie it to a demo without prompt-panel and CRM layers.

Two reporting asymmetries make this worse, not better:

1. Ads vs organic. Google Ads documentation already shows advertisers an approximated intent term for Lens / AI Mode / Overview searches. Organic Search Console shows the site owner nothing equivalent. Paid teams can exclude and bid. Organic teams get a page URL and a count.
2. Surface vs input. June 2026 gave you a generative surface report. September 24 gave you an input cut on both classic and generative views. A camera search that opens an Overview sits at the intersection: impression, no query, no click in the Gen AI slice.

For zero-click context, keep citation ≠ click next to this. For board reporting, keep GEO KPIs from mixing the new Web totals into last year’s baseline until you know whether top-level Web now includes multimodal.

---

What the report actually measures (and misses)

Question	Classic Web: multimodal	Gen AI: multimodal
Did our URL appear after an image input?	Yes (impressions, pages)	Yes (impressions, pages)
Which country / device?	Yes	Yes
Clicks / CTR / position?	Available in the classic report per public coverage	Still not a dedicated Gen AI click or position product
What did they search?	No queries dimension	No queries dimension
What was in the photo?	No	No
Did the Overview open a site, or AI Mode?	Not in this filter	Not in this filter (see the AI Mode destination change)

Do not add multimodal impressions on top of existing Web totals in a client deck until you confirm the aggregation rule. Google says the series is new. A year-over-year Web chart that quietly absorbs it will invent growth. The inverse error is also live: some SEOs reported text-based drops around September 10, and Mueller’s “not previously in the counts” answer means those drops need another cause (spam update, ranking volatility, logging). Annotate both.

---

Operator playbook this week

Layer 1: Split Web before you celebrate it

1. Open Search Console → Performance → Search results. Set Search type to Web: text-based and export 28 days.
2. Repeat for Web: multimodal.
3. Repeat both inside Performance → Generative AI features.
4. Note the date the multimodal series starts on your property. Put that date on the dashboard. Do not compare pre-series Web to post-series Web as if the definition held still.

Layer 2: The 25-minute Gen AI multimodal export

This is the DerivateX same-day move (the operator version of AEO Hack #59):

1. Generative AI features → Search type Multimodal → last 28 days.
2. Export top 10 pages by impressions.
3. Those URLs were returned in Overviews or AI Mode after an image or screenshot input. They are not “Lens vanity.” They are the pages Google is willing to cite when the query is a picture.
4. For each URL, ship the same day:
   • Filename and alt text that name the brand, ICP, or object a screenshot would contain (product UI, pricing grid, competitor logo-in-context only where accurate).
   • An opening sentence that answers the likely visual question (what is this screen, what plan is this, what tool is this) before the brand story.
   • A visible comparison or pricing module if the URL is commercial. Screenshot searchers are often mid-evaluation, not top-of-funnel readers.

Layer 3: Do not confuse camera citations with prompt-panel share

Run the usual buyer-prompt panel on ChatGPT, Perplexity, Claude, Gemini, and Google AI Overviews. GSC multimodal will not tell you when a screenshot of someone else’s UI caused Google to name a competitor and skip you. Recommendation share vs citation share still sits outside this filter.

Layer 4: Pipeline or it is theater

If multimodal Gen AI impressions rise on a glossary URL and CRM is quiet, you optimized for being a citation chip on a picture. If they rise on pricing / comparison URLs and demo notes start mentioning “found you after a screenshot / Lens / Overview,” you have a channel. Gumlet is still the public proof that revenue attribution beats a new GSC screenshot.

---

Worked example: mid-market workflow SaaS

Assume the new filter shows three URLs in Gen AI multimodal: a blog “what is RevOps,” a “HubSpot alternatives” page, and pricing.

Weak read: “We have Lens traffic now. Commission more stock photography.”

Better read:

1. The alternatives and pricing URLs are the money pages. The glossary impression is hygiene.
2. Someone (or an Overview expanding from a screenshot) is matching a visual of a competing workflow UI or a comparison grid to your pages.
3. This week: tighten alt, filenames, and the first 80 words on alternatives and pricing so they name the competitor and the job-to-be-done a screenshot implies. Do not stuff fake competitor screenshots.
4. Manually test: upload a competitor pricing screenshot and a product UI screenshot (logged out, relevant country). Note whether you are cited, named, or absent. That qualitative check is the query GSC will not give you.
5. Hold the prompt-panel cadence. A camera-triggered Overview is not a ChatGPT shortlist.

---

How DerivateX reads the multimodal filter

1. New input axis, old measurement debt. Camera and screenshot searches are now countable. Queries and Gen AI clicks are still missing. Do not let a new filter retire the stack from the GSC generative AI report piece.
2. B2B SaaS should assume screenshot intent, not catalog intent. Optimize the pages that already win evaluations (comparison, pricing, integrations), not a new “visual SEO” workstream for its own sake.
3. Web totals are about to lie. Annotate the series start. Keep text-based and multimodal as separate lines in any CMO chart.
4. Paid already has a better view of the same input. Until organic GSC gets inferred terms, prompt panels and sales notes remain the only way to recover “what they were looking at.”

If you want that baseline on your own brand, including whether camera-triggered Overviews actually touch commercial URLs, start with a free AI visibility audit. Engagement options stay on pricing.

---

FAQ

What did Google add on September 24, 2026?

A Web: multimodal search type in Search Console Performance for classic Search results and for generative AI features. It counts web results triggered by searches that use an image, photo, or screenshot.

Which products feed the multimodal bucket?

Google names Lens, Circle to Search on Android, image uploads to Search, and Chrome’s “Search this image.” Search Live and AI Mode image prompts are not named in the announcement.

Can I see the queries or the photos?

No. Google disabled the queries dimension for this type because the searches mostly use images. Mueller confirmed you do not see the photo.

Is this old Web traffic split out, or new data?

Google, via Mueller after a team check, says the data was not previously in the counts. Treat it as a new series. Watch whether top-level Web now sums both sub-types.

Does this replace the generative AI performance report?

No. That report is still impressions for AI Overviews and AI Mode, without dedicated clicks or queries. Multimodal is a filter on top of both classic and Gen AI views.

Why should a B2B SaaS team care if they are not “image-heavy”?

Buyers screenshot UIs, pricing, and comparison grids. If those images trigger Overviews that cite you, the new filter is the first official count of that path.

Where should we start this week?

Export 28-day multimodal vs text-based (classic and Gen AI), tag commercial URLs in the Gen AI multimodal top 10, fix alt/filenames/first sentences on those URLs, and keep prompt-panel plus CRM attribution running.