Product photo SEO 2026: the 13 fields Google and ChatGPT actually read

The shift in 2026 is quiet but total. Google’s AI Overviews now cite product listings the same way they cite Wikipedia paragraphs. ChatGPT’s web search reads your storefront and recommends your product by name. Perplexity returns three competitors and a price band before a shopper has typed half a query.

If your product photo arrives as IMG_0421.jpg with no alt text and a 12-word description, you are invisible to all three.

This guide breaks down the 13 fields that search engines and large language models actually read in 2026, ranked by impact, with the exact format each one expects.

Why image SEO changed in 2026

Three things flipped this year.

1. Google AI Overview rollout finished in March. Overviews now appear above 64% of commercial queries, and they cite source pages. The citations favor structured product data — pages with full JSON-LD product schema get cited 3–4× more often than pages with only HTML.
2. ChatGPT Search and Perplexity Shopping launched. Both LLMs read the title, description, image alt text and price from a product page and surface them inside the chat answer. There is no “click to view” step — the buyer sees the product description verbatim inside the LLM response.
3. Apple Foundation Models on iOS 26 + iPadOS 26 + macOS Sequoia. On-device LLMs now generate product copy that matches what server-side LLMs read. Indie sellers can finally produce LLM-readable listings without paying a copywriter.

The net result: the same 13 fields now serve both Google and the LLMs. Filling them well is no longer two jobs.

The 13 fields ranked by impact

#	Field	What it does	Google impact	LLM impact
1	Title	The primary identifier of the product across every surface	Critical — H1 + page title + structured data	Critical — first field LLMs cite
2	Slug	URL path segment, ranks for the keyword in the URL	High — top-3 on-page signal	Medium — LLMs prefer clean URLs
3	Filename	Image file name, indexed by Google Image Search	High — feeds reverse image search	Medium — used as alt-text fallback
4	Category	Taxonomy node the product belongs to	High — feeds Shopping Graph	High — narrows LLM answer scope
5	Meta description	The 150–160 character pitch in SERPs	High — drives CTR	High — verbatim citation source
6	Full product description	Long-form copy on the product page	Critical — main ranking text	Critical — primary LLM context
7	Alt text	The text shown when the image fails to load	Critical for Image Search	High — LLMs use as image caption
8	Tag 1 (style)	Style or attribute descriptor	Medium — feeds faceted search	Medium — refines LLM filters
9	Tag 2 (material)	Composition or fabric	Medium	Medium
10	Tag 3 (use case)	What the product is for	Medium	High — matches buyer intent
11	Tag 4 (audience)	Who the product is for	Medium	High — refines LLM persona match
12	Keywords	Comma-separated index hints (meta keywords + internal)	Low for Google	Medium — boosts LLM recall
13	Barcode (GTIN/UPC/EAN)	Unique identifier	High — feeds Shopping Graph dedup	High — disambiguates LLM answer

The order matters. Sellers who optimize only fields 1 and 6 leave the rest of the discovery funnel empty.

How LLMs read product listings

Here is what happens when a shopper types “best walnut cutting board for a 1-bedroom apartment” into Perplexity in May 2026.

1. Perplexity issues a web search via Brave Search.
2. The top 8 results are fetched in parallel.
3. Each page is parsed for: <title>, <meta description>, JSON-LD Product schema, all <img alt=""> text, and the first 2,000 visible characters of the body.
4. The LLM re-ranks the 8 results by semantic match — does the page describe a walnut board sized for small kitchens?
5. The top 3 are summarized into a 60-word answer with inline citations.

A page that has <title>Walnut Cutting Board – 12" Compact</title>, an alt text like walnut cutting board for small apartments, 12 inch, and a description that says “designed for studio and 1-bedroom kitchens” wins. A page with <title>Cutting Board</title> and alt text image1.jpg loses, even if the photo is better.

ChatGPT’s behavior is nearly identical, with one difference: ChatGPT also reads the filename and uses it as a fallback caption when alt text is missing. That makes the filename a second alt-text slot — most sellers waste it.

The on-device pipeline

Generating all 13 fields manually takes 8–12 minutes per product. Across 200 SKUs that’s 30+ hours of typing.

The 2026 alternative is an on-device LLM that reads the photo and writes all 13 fields in one pass. Apple’s Foundation Models framework (iOS 26+, iPadOS 26+, macOS Sequoia) runs a 3-billion-parameter LLM locally with no server round-trip and no analytics.

photoZseo uses the Apple Foundation Models framework together with the Vision framework to:

1. Detect the object in the photo (Vision’s VNClassifyImageRequest).
2. Extract any visible text — labels, packaging, model numbers (Vision’s VNRecognizeTextRequest).
3. Pull EXIF and computed color data.
4. Hand the structured signals to Foundation Models, which writes the 13 fields in your chosen language across 12 supported locales.

Output time per photo: under 30 seconds, 100% offline, zero data leaves the device.

The fields it produces map directly to the table above:

• Title (60–80 chars, keyword-front-loaded)
• Slug (lowercase, hyphenated, ASCII-safe)
• Filename (URL-safe, descriptive, JPEG/PNG/WebP suffix)
• Category (mapped to the closest marketplace taxonomy node)
• Meta description (150–160 chars)
• Full product description (180–400 words)
• Alt text (under 100 chars, keyword-natural)
• 4 tags (style, material, use case, audience)
• Keywords (8–12 comma-separated)
• Barcode placeholder if the photo shows one

Every field is editable — the AI is a draft, not a verdict.

A tactical checklist before you publish

Run this on the next product you upload to any marketplace.

• Title contains the primary keyword in the first 40 characters.
• Slug is hyphenated, lowercase, no underscores, under 60 characters.
• Filename describes the product — never IMG_*, DSC_* or a UUID.
• Category matches the marketplace’s existing taxonomy (Amazon Browse Node, Etsy category, Shopify Smart Collection).
• Meta description is 150–160 characters, ends without a period if you need every character.
• Body description includes material, dimensions, intended use, audience in the first paragraph.
• Alt text describes the photo, not the product abstract — "walnut cutting board on white background" not "walnut cutting board for sale".
• At least one tag per attribute axis: style, material, use case, audience.
• Keywords are unique to this SKU, not copy-pasted across the catalog.
• Barcode (GTIN/UPC/EAN) is filled if you have one — empty barcodes hurt Shopping Graph eligibility.

A listing that passes all ten boxes gets crawled, indexed and cited. A listing that fails three or more drops out of the Overview entirely.

Why “thin product pages” lose in 2026

Google’s 2024 Helpful Content Update merged into the core algorithm in late 2025. Pages with 50-word descriptions and no structured data are now classified as thin commerce and demoted in the SERP regardless of backlinks.

LLMs penalize them even harder. A product page with no description cannot be cited because there is nothing to quote. Perplexity will skip it and quote the page below it.

This is the single biggest behavior change of 2026: the LLM citation is the new featured snippet, and a featured snippet needs text to quote.

The takeaway

The 13 fields are not optional. They are the index. Whether a shopper arrives via Google Image Search, an AI Overview, a ChatGPT recommendation or a Perplexity comparison, the same 13 fields decide whether your product is in the answer.

The good news: filling them well is no longer a 10-minute-per-SKU task. On-device AI now drafts the full set in under a minute. The seller’s job in 2026 is to review and refine, not to type.

---

photoZseo is on the App Store. iPhone, iPad and Mac.