AI search engines now grade the image before the alt text

What the patent describes

Google holds a patent, filed in 2023 and published in April 2026, that describes an answer flow where the cited source is chosen by image match first, and the surrounding text is pulled in afterward to build the answer. Andy Chadwick, who surfaced the filing, is clear it is a patent application, not confirmed production behavior. Google files thousands of these and most never ship as described. Even so, it is the clearest public signal that the picture, not the paragraph, decides whether a page gets pulled into an AI answer.

The volumes behind that shift are already large. Google Lens handles roughly 20 billion visual searches a month. On Pinterest, visual search is 30% of all searches and converts 62% better than text search, and Pinterest Lens alone processes 1.5 billion visual searches a month. Alibaba's camera-first shopping has passed 10 million visual searches a day for years.

Alt text was never the whole job

Alt text and captions were the fix when the only reader was a crawler parsing a filename or a tag. The image itself now carries more of the signal directly: whether the hero photo is original or a duplicate a dozen competitors also use, whether packaging text is legible enough for OCR to extract, and whether the page text actually confirms what the photo shows.

Search Engine Land's scorecard checks that with tools most SEO teams already have access to through Google Cloud Vision API:

• Ownership: run Web Detection on a hero image to see whether it is a duplicate, near-duplicate, or shared stock image.

• Legibility: run TEXT_DETECTION (OCR) on product shots. Character heights under roughly 30 pixels, or contrast under about 40 grayscale values, tend to fail.

• Context: run OBJECT_LOCALIZATION to see what else the model reads in frame alongside the product, since that builds the brand story a person or a model takes away.

None of this replaces alt text. Alt text tells the machine what the page claims; these checks tell you whether the image backs that claim up once the model strips it down to pixels.

What to do with it

The single highest-leverage move Search Engine Land names is replacing stock and shared hero images with photography nobody else has. It fixes the ownership problem and the duplicate problem at once.