
Primary topic Mobile SEO
On November 16, 2026, Google added Lens reporting, brand classification and multi-value filters to the Search Analytics API. A retailer can now pull the product pages people find through image-based searches on phones in several countries, in one request. If your dashboard already pulls Search Console data, add a page-level multimodal series. Use the API for recurring reports; for a one-off check, select the multimodal search type in Search Console and export it.
| Goal | Use |
|---|---|
| Lens and Circle to Search pages | Query webMultimodal by page |
| Google Images tab | Keep type: image |
| Brand share in web search | Group by brandedClassification |
| Phones and tablets across markets | Add country and device lists |
Which searches count toward the new report?
webMultimodal covers searches made with inputs such as a camera image or a screenshot. The September 24 Search Console rollout began globally and listed Google Lens, Circle to Search on Android, image uploads in Google Search and Chrome’s Search this image command. Someone can type a shoe name and select the Images tab. That activity belongs to the existing image search type. Someone who uses Lens to identify a shoe from a picture is the kind of search the new report captures.
The reporting rollout includes the US, the UK and Europe; country filters select the audience you measure. Start by grouping results by page. You can see which destinations turn up often and which get clicks. The report combines several ways of searching by image, and specific query text is unavailable. A desktop result tells you which device class was recorded. It doesn’t identify the original picture or the tool the person used.
Pull one region and two device types
Send a POST to the Search Analytics query endpoint using your existing OAuth connection. For a reporting job, use the webmasters.readonly OAuth scope. The endpoint begins https://www.googleapis.com. In the path /webmasters/v3/sites/ followed by {siteUrl} and /searchAnalytics/query, encode the full property identifier as one segment. The Domain property sc-domain:example.com becomes sc-domain%3Aexample.com. A URL-prefix property uses its full URL exactly as listed in Search Console.
This body asks for pages found in Germany, Austria or Switzerland on a phone or tablet. The start and end are inclusive and use Pacific Time, even when the report covers European traffic. The country list accepts ISO three-letter codes. Use a device list for MOBILE and TABLET, or use a single device value if your report is phone-only.
{
"startDate": "2026-10-17",
"endDate": "2026-11-13",
"type": "webMultimodal",
"dimensions": ["page"],
"dataState": "final",
"dimensionFilterGroups": [{
"groupType": "and",
"filterLists": [
{
"dimension": "country",
"expressions": [
"deu", "aut", "che"
]
},
{
"dimension": "device",
"expressions": [
"MOBILE", "TABLET"
]
}
]
}]
}
A row needs a match in each list. Germany on a tablet qualifies. Germany on desktop and the UK on mobile fall outside this request. To report UK and Irish traffic instead, use gbr and irl; for the US, use usa. The country is where the search was made, so changing the list changes the audience in the report. If a job already has a country rule in filters, replace it when you add the country list. The API rejects filtering the same dimension in both fields.
Each row’s keys array follows the requested dimension order. Here it contains the canonical page URL. Keep clicks and impressions beside CTR and average position. API ctr is a fraction from 0 to 1; multiply by 100 to display a percentage. For a single regional total, omit dimensions and leave the filter lists in place. Add country or device to dimensions when you need separate rows. Page and property aggregation can count impressions differently, so compare totals using the same responseAggregationType. Leave aggregationType at auto when grouping or filtering by page; byProperty is invalid in that case. Responses are sorted by clicks, highest first; date-grouped responses run from oldest to newest. The default rowLimit is 1,000; set it to 25,000 and advance startRow for larger reports. Pagination still cannot guarantee every detail row, so obtain a site total with an aggregate request rather than summing page rows.
What a real regional total looked like
In a November 16 test on 22 sites, analyst Ewelina Podrez-Siama reported that a regional call for madeira-planer.de returned 65 clicks, 1,824 impressions and an average position of 16.07 for Germany, Austria and Switzerland from October 17 to November 13. Three individual country calls produced the same rollup when position was weighted by impressions. This was a check of the country filter. The phone-and-tablet Lens request above is a different slice.
The CTR for that slice is 65 divided by 1,824, or about 3.56%. Averaging the three country CTRs would give a small market the same influence as a large one. When the region request matches clicks, impressions and weighted position in your own parallel run, switch the recurring regional total to one call. Keep country rows if an editor or merchant acts on each market separately.
Give brand reporting its own request
For your web report, set type to web and replace the page dimension with brandedClassification. Keep the country and device lists if you need the same audience. The row keys distinguish BRANDED and NON_BRANDED. To request just the second group, use a brand filter with equals and NON_BRANDED. A query-text filter is incompatible with brand grouping or filtering, and Podrez-Siama’s live request combining brand classification with webMultimodal was rejected. Keep brand and multimodal reporting in separate request bodies. The same live test rejected branding with discover, but accepted it with image.
For the API, brand history starts February 21, 2026. Use a top-level property covering the site: path-only and subdomain properties are ineligible, as are sites without sufficient impressions and query volume. The UI filter was announced November 20, 2025, and reached all eligible sites March 11, 2026; those rollout dates are separate from the API history cutoff.
The test covered eight of the analyst’s sites and 14 client or friend properties. Twelve returned the split; visible-query coverage varied from 8% to 57% of clicks by site. Their two classes matched total clicks, while visible queries covered 45% of clicks across the sample. A regex over query rows therefore measures a different population. Google’s classification also handles misspellings and names of products associated with the site. Prefer the new split for eligible web reports after a label audit, because it can include clicks whose query text is hidden. Keep historical regex results labeled as query-text estimates.
For that audit, request the two dimensions together. Grouping by query is allowed here; filtering on query text is what conflicts with branding. This body returns each visible query and its classification in that order:
{
"startDate": "2026-10-17",
"endDate": "2026-11-13",
"type": "web",
"dimensions": [
"query",
"brandedClassification"
],
"dataState": "final",
"rowLimit": 25000
}
To reproduce a regional brand-and-device report, instead use brandedClassification and device as the dimensions and the country list from the first body. For a brand-share total, group only by classification and divide branded clicks by the sum of both classes. Audit the labels before adopting the metric, then check them monthly. The classifier handles multiple languages, but a generic phrase can still receive a branded label. These labels help reporting; they do not change rankings.
The label audit matters even for an exact name. On podrez.pl, the query ewelina podrez-siama received 195 impressions: 50 branded and 145 non-branded, with the label changing by day. On ms-fox.pl, 180 of 186 impressions for the same query were branded. Conversely, the classifier found 231 branded clicks on ms-fox.pl where a regex for the blog name found 25. Include exact names, typos, associated products and generic phrases in the label audit.
Across the 12 classified sites, branded CTR ranged from 3.6% to 75%, versus 0.6% to 3.3% for non-branded traffic. Nine sites had branded click shares below 5%; the other three ranged from 18% to 30%. Those are different businesses, so use your own matched-period comparison to judge change.
When a request comes back empty
Run a broad request for the same property and finalized dates before investigating individual pages. First remove the country and device filters. Then remove grouping and compare the aggregate with the matching Search Console report. A row with impressions and zero clicks is a useful signal. An absent row calls for a scope check. For the branded report, check the API history cutoff and property eligibility first. There is no published numeric eligibility threshold; test the intended property rather than adopting another site’s traffic level as a rule.
For the Lens report, check whether the Web multimodal view shows data in that period. Keep dataState: final for reconciliation. Use all only when you want recent, potentially changing data. For multimodal detail, group by page, country, device, date or search appearance; adding query produces an error. A startRow beyond the available rows also produces an empty successful response.
If your client library rejects the new field before sending it, try the authenticated REST request. When the API itself rejects it, inspect the returned field error and compare the body against the current reference. Save the property, search type, date range and filters alongside the results. Store separate statuses for an empty response, a rejected request and a row with zero clicks.
Choose a page to inspect
Concentration is more useful than the sample average. On podrez.pl, multimodal impressions reached 5% of web impressions and 91% of that activity went to one AI Mode guide; a quarter of the guide’s multimodal impressions were desktop. After screenshots were added on October 27, the guide rose from at most 12 impressions a day to 138–231 a day from November 1. That sequence identifies a page to investigate, but does not establish that the screenshots caused the rise. The sample’s earliest multimodal data was September 10, before the UI announcement; check your property’s available dates rather than assuming the release day is its first data day.
In the 22-site sample, 15,806 multimodal impressions produced 105 clicks, a rounded 0.7% CTR; impressions were equivalent to 0.4% of web impressions. Among sites with at least 40 multimodal impressions, average position ranged from 1.5 to 2.3 and phones accounted for 74–96% of impressions. For your own store or publication, look for a page with repeated impressions over finalized dates, then check that its main image, title and destination make sense together on a phone. If clicks look steady but sessions fall, trace the landing and referrer before changing the photo or the report.
For an existing warehouse pipeline, the update also adds multimodal data to active BigQuery bulk exports. Keep the API for a small dashboard; use the daily bulk export when you need detail beyond API top-row limits.