AI commerce
Help buyers find a product when they only have a picture
Connect useful product images to clear page facts, then measure multimodal search without mistaking a visual match for confirmed product fit.
TL;DR
- A buyer may know a product's shape before knowing its name; the destination page must turn that visual clue into a safe comparison.
- Use clear images, descriptive filenames, useful alt text and visible product details together.
- Google's multimodal reporting includes image-led search experiences; it is different from treating all visual discovery as ordinary image-search traffic.
- Measure the relevant page and inquiry quality. A visual match alone does not establish compatibility.
A buyer may start a search with a photo of an old fitting, a broken part or a product on a site. They may not know its model number. The page they reach should explain what it is, what it fits and what to check before buying.
Adding a keyword to an image does not solve that task. Connect the photo to a named product. Show its size and give buyers a way to check an uncertain match.
What can now be measured?
Google announced multimodal Search Console reporting on September 24, 2026. The announcement names Lens, Circle to Search, image uploads and Chrome's image-search action. It describes a multimodal search-type filter in Search results and generative AI performance reports, with visibility dependent on receiving relevant traffic. This confirms the feature, not its presence in every individual account. See Google's announcement.
Save the filter and date range with each export. An empty report does not prove that the images are poor. Check the traffic, the report's availability and the view you chose. Do not rename the report “AI product recommendations.” It measures the documented search activity, not a buyer's final product decision.
Make the image and the page describe the same thing
Google's image guidance explains that image understanding uses page context, alt text and computer vision. It recommends descriptive filenames and relevant nearby text. Alt text is one input, not a substitute for a useful image or a complete page.
Use a view that distinguishes the product from nearby alternatives. Put a model name and meaningful specification beside it. If the image shows an accessory that is sold separately, state that in readable text. Put dimensions in HTML as well as in any diagram, so the buyer can compare them without zooming into a picture.
Worked example: an unknown cabinet hinge
This is a fictional hardware supplier. A buyer uploads a photo of a concealed cabinet hinge. Two products look similar, but they need different door preparations. The supplier's page uses IMG_8842.jpg, the alt text “quality hinge” and a lifestyle image as the main photo.
Check the real product photos, approved spec sheet, page copy and catalog record. The gap is clear: the page never links the image to the sizes that define fit.
| Element | Proposed revision | Acceptance test |
|---|---|---|
| Main image | Use the actual H35 concealed hinge with the mounting plate visible. | Photo matches the offered SKU and included parts. |
| Filename | h35-concealed-cabinet-hinge-mounting-plate.jpg | Name describes the asset without a list of search phrases. |
| Alt text | H35 concealed cabinet hinge shown open beside its mounting plate. | Text describes what this image contributes to the page. |
| Visible fit details | Publish cup size, opening angle and door requirements from the approved sheet. | The buyer can compare the old fitting with the specification. |
| Uncertain match | Invite the buyer to send dimensions and a second photo for a fit check. | The request explains which evidence the sales team needs. |
The content owner checks every detail against the product record. The page reviewer then hides the images and reads the remaining information. Next, the reviewer inspects the images without the marketing copy. Both views should identify the same product and leave the same fit questions open.
W3C's image decision tree adds an essential distinction: text alternatives depend on the image's function. An informative product photograph needs a useful description; decoration does not need a sales pitch in its alt attribute.
Compare discovery and business outcomes separately
Save a baseline for the relevant pages and multimodal view. After the change, compare like periods, places and devices where the report allows it. Note shifts in stock and demand. Impressions show that a result was seen. Clicks show visits. Neither is an accepted order.
For a quote-led business, review whether image-led visitors supply the details needed to identify the part. More vague “is this the same?” requests may reveal missing size details. Use those questions to improve the page. Do not count every request as a win.
Common questions
Should we rename every existing image file?
Prioritize images on important product pages. Before changing a live URL, account for existing links, caches and references. A descriptive name is useful, but does not justify careless asset migrations.
Can an image prove product compatibility?
A photo may narrow the options. Exact dimensions, materials, model versions and installation requirements can still differ. Make those checks available before the buyer acts.
Semanta can help map visual buyer questions to product facts and useful next steps. Bring a product family with frequent identification questions to discuss a content and discovery review.