Visual search optimization

TL;DR

Visual search rewards brands that treat images as first-class content. Rename the file, write real alt text, add ImageObject schema, keep EXIF sane, and feed the Merchant Center and Pinterest pipes cleanly. Google Lens now runs over 20 billion queries a month, so this is not a fringe surface anymore.

The surface, in one paragraph

Visual search is the class of queries where the input is an image, not text. A shopper points Google Lens at a chair in a friend's living room. A designer pulls up Pinterest Lens on a magazine photo. An Amazon customer opens StyleSnap on a jacket in a wedding photo. In each case the engine tries to identify the object, match it to inventory, and route the person to the closest product. The engines match on pixels, but they rank the results using the same signals that drive text search: page authority, structured data, freshness, and merchant feed quality. The brand that wins is the one whose product image is the best-labeled, best-structured version of that object on the open web.

Where this fits in Search Everywhere Optimization

VxSO is a specialized cousin of SEO, tightly coupled to PFO and KGO.

SEO is the substrate. Google Lens returns Google Images results, then layers matching. If the product page does not rank on the text query, its image loses too. Winning VxSO always starts with winning the underlying page.

PFO is the volume amplifier. Merchant Center product feeds are the fastest way to get product images indexed and matchable by Lens. A clean feed puts thousands of images into the visual index simultaneously.

KGO gates entity match. When Lens identifies a plant, a building, or a landmark, it draws from Google's Knowledge Graph. Brands with a Wikidata QID and Knowledge Panel get entity attribution. Brands without one become "similar looking product."

SEO and CWV compound. Fast-loading, well-marked-up pages get their images crawled and re-crawled. Slow, JavaScript-heavy pages get their images missed entirely.

FEO overlaps in Pinterest. Rich pins, Pinterest article pins, and feed-driven distribution move the same images through a second discovery pipe.

The five levers

1. Filename discipline

Every image on the site gets a descriptive, hyphenated filename. IMG_4271.jpg becomes walnut-mid-century-dining-chair-charcoal-linen.jpg. This one change moves the needle more than any other single VxSO task, and 90% of ecommerce sites still fail at it because their DAM exports with camera default names. I bake filename standards into the DAM export template and into the PDP template so this stops being manual work.

2. Alt text that reads like a caption

Alt text is the primary text signal Lens matches against. I write it as if describing the image to someone on a phone call: "Walnut mid-century dining chair with charcoal linen upholstery, three-quarter angle." Not "chair.jpg." Not the SKU. Not the entire product description. A caption. Every image, every page.

3. ImageObject schema everywhere

ImageObject schema on every product image and every editorial hero image. Fields I always fill: contentUrl, name, description, creator, license, copyrightNotice. For product images: representativeOfPage true on the hero. For editorial: creator points to a Person schema with sameAs to their profiles. This is what tells Lens the image is authoritative for that concept.

4. EXIF that helps, not hurts

Keep author, copyright, and geotag (when the location matters) in EXIF. Strip camera serial numbers, personal metadata, and edit software chains. Set IPTC caption and headline to match alt text. Most brands strip everything or preserve everything blindly. The middle path is disciplined: keep the fields that verify authorship and location, strip the fields that leak.

5. Feed and pin discipline

For product brands, a clean Google Merchant Center feed puts every image into Google Shopping and Lens simultaneously. Pinterest business account with claimed website, rich pins enabled, catalog uploaded. These are the two pipes that carry the most VxSO traffic for commerce. For editorial brands: Open Graph image tags with correct aspect ratio and dimensions, plus Pinterest article pins configured.

First 30 / 60 / 90 days

Days 1 to 30: audit and standards

Crawl. ScreamingFrog with image extraction, alt-text audit, and structured data extraction. Which images have alt text, which do not, which are named IMG_1234.jpg.

Merchant Center audit. Feed health, image quality flags, disapprovals. Pinterest business account state. Rich pin validation. Catalog upload state.

Filename convention drafted. Alt text style guide drafted. ImageObject schema template drafted for PDPs and editorial pages.

EXIF policy documented. What to keep, what to strip, at DAM export time and at CMS upload time.

Deliverable at day 30: a written standards document, an audit report of the current image state, and a prioritized rewrite queue for the top 200 images by traffic value.

Days 31 to 60: rewrite and ship

Filename remediation on the top 200 product images. 301 redirects from old paths where images were linked externally.

Alt text rewrites on the top 200 images. Editorial and PDP templates updated to enforce alt text at publish time.

ImageObject schema shipped on the PDP template. Editorial CMS gets an image block template with schema baked in.

Merchant Center feed fixes. Image quality flags resolved. Additional image slots populated with lifestyle and detail shots.

Pinterest rich pins validated. Article pins configured on editorial. Catalog product pins mapped to inventory.

Deliverable at day 60: 200 remediated images, template-level enforcement of standards going forward, clean Merchant Center feed, live Pinterest rich pins.

Days 61 to 90: measure and expand

GSC images-tab data. Image impressions, CTR, and query terms. Compare to day 30 baseline.

Pinterest analytics review. Lens-driven impressions, saves, outbound clicks.

Google Lens spot-check. Take twenty product photos, run them through Lens, verify the brand's product surfaces in the top results. Track over time.

Expansion. Editorial images beyond the top 200. Category page hero images. Blog post inline images. In-store or lifestyle shot rollout.

Deliverable at day 90: a working image discovery pipeline, measured lift on image-driven sessions, and a plan for the next 500 images.

Tools I use

• Google Search Console. Images performance tab. Image impressions and clicks separated from web.
• Google Merchant Center. Product feed hygiene, image quality flags, additional image support.
• Google Lens on device. Manual verification of visual match quality on top products.
• Pinterest Business account + Pinterest Analytics. Rich pin validation, catalog performance, Lens surfacing.
• ScreamingFrog SEO Spider. Site-wide alt text audit, image extraction, structured data validation.
• ExifTool. EXIF audit and remediation at scale.
• Cloudinary or Imgix. Image transformation and delivery. AVIF and WebP for CWV. Named transformations for DAM discipline.
• Rich Results Test. ImageObject and Product schema validation.
• Bing Visual Search. Secondary surface, worth spot-checking for the same products.

What kills the program

DAM exports with default filenames. The photographer ships IMG_4271.jpg and the CMS accepts it. Nothing gets fixed downstream.

Alt text stuffed with keywords. "Buy cheap chair discount furniture online store chair for sale" is not alt text. Lens penalizes it. Write captions, not spam.

JavaScript-only image loading. Lazy loading is fine when done with native loading="lazy" or a well-configured library. Client-side rendered images that never appear in the raw HTML get missed by crawlers.

Watermarks blocking the object. Heavy watermarks reduce match confidence. Keep watermarks subtle and off the primary product silhouette.

Stripping all EXIF. Losing author and copyright metadata makes it harder for engines to attribute the image to the brand. Blind stripping is worse than disciplined preservation.

Merchant Center feed neglect. Disapproved products stop appearing in Lens results. Weekly feed review is table stakes.

Confusing StyleSnap with a growth channel. If the brand does not sell on Amazon, StyleSnap does nothing for the brand. Do not waste cycles.

KPIs that matter

• Image impressions in GSC. Monthly trend on the Images tab.
• Image click-through rate. Rising CTR means alt and filename work is landing.
• Merchant Center approved product count. Approved products with valid images ready for Lens matching.
• Pinterest Lens surfacing rate. Sampled: given a hero product photo, does the brand's pin appear in top ten Lens results.
• Lens-referred sessions in GA4. Small but tracked.
• Rich pin coverage. Percent of eligible pages that resolve to valid rich pins.

FAQ

Is visual search actually a real channel?

Yes. Google Lens processes over 20 billion queries a month and half of Pinterest search is visual. For product, apparel, home goods, and plants, VxSO delivers measurable sessions.

What is the single biggest VxSO fix?

Renaming the hero image file from IMG_4271.jpg to walnut-mid-century-dining-chair-charcoal-linen.jpg, adding descriptive alt text, and putting ImageObject schema on the page. That trio moves the needle.

Does EXIF data matter?

Some. Lens reads geotags for a subset of queries. I preserve author, copyright, and geotag when appropriate. I strip camera serial numbers and personal metadata for privacy.

How do rich pins fit in?

Rich pins pull product, article, or recipe metadata from Open Graph tags on the source page. Pinterest Lens surfaces rich-pinned images more prominently because the metadata is machine-verified.

Does Amazon StyleSnap help non-Amazon brands?

Not directly. StyleSnap only matches to Amazon inventory. For DTC brands off Amazon, focus on Google Lens and Pinterest Lens. StyleSnap only matters if you sell on Amazon and want your ASINs surfaced.

How do I measure visual search traffic?

Google Search Console filters images searches separately. Pinterest Analytics shows Lens-driven impressions. GA4 referrer analysis catches lens.google.com and pin.it referrers. It is small but trackable.

Related reading

• Product feed optimization (PFO): the Merchant Center pipe that carries most VxSO product traffic.
• Voice search optimization (VSO): the audio-side sibling.
• Knowledge graph optimization (KGO): entity work that gates Lens attribution.
• Core Web Vitals: image loading discipline for LCP.
• Feed and Discover optimization (FEO): Google Discover surfaces images too.