Image Metadata That Helps Fashion Products Get Found

“Add metadata” sounds like one task. In practice, a fashion image can carry information in at least six places: its filename, the HTML that embeds it, the page around it, structured data, a commerce feed, and metadata stored inside the file. Each layer serves different readers and systems.

The mistake is to search for one perfect field. Alt text is essential for accessibility and useful for image understanding, but it is not a product feed. A caption can explain why an image is synthetic, but it does not replace a rights record. Embedded provenance can record origin and edits, but it does not prove that a visual claim is true.

A durable workflow assigns a clear job to every layer.

Layer 1: the file

Use a short, descriptive filename that remains stable when the visual itself has not changed. ink-cotton-field-jacket-front.webp gives a person maintaining the library more information than IMG_8439.webp. It also gives search systems a light contextual clue, according to Google’s image guidance.

The file should have the correct extension and MIME type, enough resolution for its use, reasonable compression, and known dimensions. Store a master separately from web renditions. Avoid repeatedly changing the URL for trivial edits because other systems may need to recrawl it.

File-level metadata can include creator, copyright, credit, description, license information, and digital source type. The IPTC Digital Source Type vocabulary distinguishes different ways digital media can originate or be transformed. Those values can support provenance, but only if tools preserve them during export, optimization, CDN processing, and download.

Layer 2: accessible HTML

The alt attribute describes the image’s function and relevant visible content in its page context. It is not a bag of keywords. If the same image is decorative in one location and informative in another, the appropriate alt treatment may differ.

For a product page, useful alt text might be: “Ink-blue cotton twill field jacket, front view, with four patch pockets and drawcord waist.” It should mention only features that are visible and verified. “Sustainable premium jacket” is not a visual description and may be an unsupported claim.

W3C’s image decision tree provides a practical starting point for deciding whether an image needs a text alternative and what kind. Human accessibility testing still matters because a syntactically present alt attribute can be unhelpful.

Use real HTML image elements, responsive sources where appropriate, declared width and height, and captions when readers need context. Google documents that it finds images in standard image-element source attributes and uses page content, captions, titles, filenames, alt text, and computer vision as context.

Layer 3: the landing page

The page supplies facts that do not belong in alt text: material composition, care, dimensions, fit notes, price, availability, return conditions, reporting method, creator credit, and correction history. Place the image near the text it supports.

For editorial work, a caption can distinguish documentary photography, a supplied brand image, a diagram, and an AI-generated editorial illustration. The disclosure should tell readers what they are seeing, not merely say “AI-assisted” somewhere in a site-wide policy.

For commerce, the selected image must agree with the visible variant. A detailed file record cannot repair a page that shows an ink jacket while the purchasable selection says stone.

Layer 4: structured data

Structured data describes the visible page in a machine-readable form. It should not introduce hidden product facts. Google’s product documentation explains markup for product snippets and merchant listings. Required and recommended properties depend on the result type.

An editorial article may use Article and ImageObject properties appropriate to what readers can see. A transaction page may use Product and Offer information that matches price, availability, and variants shown on the page. Validate the output and check rendered HTML, not only a CMS field.

Google’s image-license metadata guidance also describes creator, credit, copyright, and license information that can be supplied through structured data or embedded IPTC fields. A rights field is only as reliable as the records behind it.

Layer 5: commerce feeds

A commerce feed is an operational contract with a platform. The primary image URL, additional images, title, color, material, pattern, size, item group, availability, price, and identifiers need to refer to the same product state.

Feed requirements can change. The current Merchant Center image-link specification includes format, size, URL, staging, and overlay rules. The team responsible for export should monitor diagnostics instead of assuming the CMS view proves that the feed is valid.

Layer 6: provenance and synthetic-media disclosure

Google’s guidance on generative content encourages accuracy, quality, relevance, process context where useful, and image metadata. It also points ecommerce publishers to more specific Merchant Center policies for AI-generated content.

For FashionMember, the minimum reader-facing record for a synthetic editorial image is a clear caption, accurate alt text, an article-level AI disclosure, the saved prompt, the generated master, the web rendition, and a rights/source log. When the toolchain supports it, embedded source-type metadata should travel with the asset.

That stack still does not transform a fictional garment into product evidence. Provenance answers questions about origin and history; product truth requires separate verification.

What our ten-image audit found

FashionMember audited the ten completed article hero images in Batches 01 and 02. The machine-readable results are stored in content/data/FM-044-image-metadata-audit.csv.

All ten files had descriptive filenames, 1672 × 941 web renditions, article-level alt text, AI-labeled captions, and AI disclosure fields. File inspection found no EXIF or XMP chunks in the WebP renditions. That is a real gap in portability: if an image leaves the article context, its source information does not travel inside the optimized file.

The finding does not mean the images are undisclosed on FashionMember. The disclosure exists at the page and project-record layers. It means the production workflow should test a metadata-preserving export and CDN path before claiming portable provenance.

A minimum fashion image record

For each important image, preserve:

• asset ID and descriptive filename;
• master and derivative locations;
• width, height, format, and crop relationship;
• creator, owner, license, credit, and expiration where relevant;
• product ID, variant, color, material, and image role;
• alt text and contextual caption;
• documentary, supplied, edited, composite, or synthetic source type;
• prompt and generation tool for AI-created work;
• approval date, approver, and restrictions;
• canonical landing page and replacement history.

The record should live in a system the team can audit, not in one employee’s desktop folder.

What metadata alone cannot prove

This audit covers ten project-owned editorial WebP files, not a live ecommerce catalog or CDN. It checks presence of project fields and file chunks, not accessibility quality with users, indexation, ranking, rights validity, or platform acceptance. Search and merchant requirements must be checked again when the assets are deployed.

Sources and verification

• Google Search Central: Image SEO best practices — HTML discovery, page context, filenames, alt text, quality, and performance.
• W3C Web Accessibility Initiative: Image decision tree — functional decisions for text alternatives.
• Google Search Central: Product structured data — visible product information and result-type requirements.
• Google Search Central: Image license metadata — creator, credit, copyright, and license fields.
• IPTC: Digital Source Type vocabulary — controlled terms for digital-media source types.
• Google Search Central: Generative AI content guidance — accuracy, quality, disclosure context, metadata, and ecommerce policy links.
• Google Merchant Center: Image link specification — current product-image rules and recommendations.