Can you pull Search Console's new web multimodal search type from the API? Tested the day Google announced it: 7 candidate values, 7 rejections

What Google announced, in its own words

The primary source is the Search Central post Announcing web multimodal Search performance reporting in Search Console, dated Thursday, September 24, 2026, signed by Harsh Kharbanda (Product Manager Lead, Google Lens) and Moshe Samet (Product Manager Lead, Search Console). The operative sentences: "Today, we are excited to introduce Search Console reporting for web multimodal search, in the report for performance on Search results as well as in the report for Generative AI features." And: "The data reported here includes searches with Lens, Circle to Search on Android, image uploads to Google Search, and the Chrome right-click 'Search this image' feature." The rollout is global "starting today"; the metrics appear "if your site is receiving traffic from these queries."

Both help pages were updated to match. The Performance report help page now reads: "Web: text-based - Includes traditional search traffic originating from text queries entered into the standard Google search bar. Web: multimodal - Includes web search results where an image was used as part of the search." The Generative AI performance report help page carries the same two definitions. Neither page mentions the queries dimension; the Search Console interface does, and Search Engine Roundtable and Search Engine Journal both quote it: "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."

The test: seven candidate API values, seven rejections

The Search Analytics API takes a `type` parameter. Google's API reference, read today, lists six allowed values: `discover`, `googleNews`, `news`, `image`, `video` and `web`. Nothing about multimodal. We sent the same 28-day query (2026-08-25 to 2026-09-22, our own property, the `date` dimension) once per candidate value from a script committed beside this post and recorded every response.

Read at 2026-09-24T16:41Z. Every documented value returned 200; every guess at the new type returned the same 400. This is one property on one day and a sample of names, not proof that no name exists: Google can add an enum value without changing the reference page. What the table does establish is that on the day of the announcement the six documented types are the six that work, and a pipeline that reads the API gets the old `web` total with no way to say how much of it came from an image. It is the shape we found in August for the Generative AI report: six candidate values then, six rejections, interface and export only.

Why this matters for a business that cares about AI visibility

• Image-triggered search is now a line you can read, and it has no query text. A customer who points a phone camera at a storefront, a menu or a product and lands on your page shows up under Web: multimodal with a page, a country, a device and a date, never a query. You can see whether that route sends anyone; you cannot see what they photographed.
• The Generative AI features report gets the same filter and keeps the same blind spots. That report shows impressions inside AI Overviews and AI Mode with no clicks and no queries. A multimodal filter on it says an image search produced an AI answer that included your link. It does not say the answer recommended you.
• Dashboards that read the API cannot see the split. Reporting built on `searchAnalytics.query` cannot separate text-based from multimodal web traffic today, and the `web` total may already contain both. The split lives in the interface and its export, so a client's multimodal number is a manual pull for now, with no query column.

What we will re-read

The probe re-runs in one command. We will run it again when Google's API reference changes or a second outlet reports a working value, and update this post with the date. Search Console still does not answer the question underneath for any search type: did the AI answer an image search produced name the business? A free 60-second check samples that directly on ChatGPT and Perplexity. The $79 audit adds Gemini and Claude, 25 questions sampled three times each.