Product Specs in Images Don’t Count: What AI Can’t Read on Your Store

April 3, 2026

TL;DR

If your best product facts live only in hero graphics, comparison images, PDF spec sheets, or fragile widgets, they are much less useful to AI shopping systems than you think.

That does not mean AI tools are blind to every image. It means the systems deciding whether to cite, compare, and recommend your product rely far more heavily on text, structured fields, and merchant data than on visual inference.

Google’s Search docs explicitly say important content should be available in textual form and that structured data should match visible text.¹ Google’s image guidance also says it extracts information about the subject matter of an image from the content of the page, including captions and image titles.² Shopify’s current AI shopping guidance puts it even more bluntly: AI shopping platforms can process images, but they rely primarily on structured product data rather than visual inference to match products to shopper queries.³

That is why “all the important specs are in the image” is one of the most expensive product-page mistakes in AI commerce.

Why this problem is getting worse

Traditional ecommerce design got merchants used to hiding information in visual assets.

It made sense:

put the size chart in a nice graphic,
show a “what’s included” collage,
create a feature comparison image,
tuck the detailed spec sheet into a PDF,
or let the review app and accordion UI handle the rest.

Humans can work with that. AI shopping systems often cannot work with it well enough.

That matters more now because AI product discovery is increasingly comparison-led.

OpenAI’s shopping experiences now compare products side by side with price, reviews, and features.⁴ Shopping research in ChatGPT can build a buyer’s guide, compare items side by side, and translate technical specs into practical implications.⁵ Google’s AI features increasingly appear on shopping queries, and Shopify says complete, well-structured product data is essential to show up well in Google AI Shopping and similar AI-powered results.⁶

If the model cannot reliably extract the key facts from your PDP, the product does not just become “less optimized.” It becomes harder to recommend at all.

What AI actually reads on a product page

If you want the unglamorous answer, AI systems are usually much happier with these inputs:

visible text,
structured product data,
feed data,
headings and tables,
policy pages,
FAQs,
and consistent merchant metadata.

That is the raw material they can normalize and compare.

What Google says

Google’s “AI features and your website” guide says:

important content should be available in textual form,
structured data should match visible text,
and Merchant Center information should stay up to date.¹

Google’s image SEO guidance also says it derives information about the image from surrounding page content, captions, and image titles.²

That should change how merchants think about graphics. A beautiful feature image can support understanding, but if the actual product facts are not also present in nearby text, headings, or structured data, the image is doing less machine work than you think.

What Shopify says

Shopify’s Perplexity shopping guidance is unusually clear here. It says a person may infer from an image that a tote bag can hold a laptop, but AI shopping platforms rely primarily on structured product data like dimensions, materials, and weight capacity to match a product to a query.³

That single sentence should reset a lot of product-page strategy.

Because once you internalize it, you stop asking:

“Does the page look good?”

and start asking:

“Could a model answer the buyer’s question from this page without guessing?”

What OpenAI’s commerce docs imply

OpenAI’s product feed docs focus on structured fields: titles, descriptions, images, price, availability, and seller context.⁷ The “Get Started” guide emphasizes making your catalog understandable to ChatGPT with up-to-date product data through a documented schema and delivery model.⁸

That is the clue. Images matter. But images are not the primary substrate for reliable extraction.

The four places merchants hide critical product facts by accident

1) Hero graphics and comparison banners

This is the most common one.

A merchant creates a polished hero image that says:

64mm flat burrs
40 grind settings
low retention
for espresso and pour-over

It looks great.

But if those same facts are not also written in the page text or output in structured fields, the graphic is mostly serving humans, not the systems deciding whether your product matches:

“best grinder for espresso under $500”
“low-retention grinder for single dosing”
“flat burr grinder for pour-over”

2) PDF spec sheets

PDFs are convenient for internal teams and B2B buyers. They are a weak default for AI commerce if the facts only live there.

If your PDP says “Download spec sheet” and that is where the actual wattage, dimensions, voltage, compatibility, maintenance, or included accessories are stored, you are forcing both humans and models to do extra work.

The fix is simple: the PDF can stay, but the core specs need to be mirrored into the page itself.

3) Accordion and tab overload

Not every accordion is bad. The problem is when the page’s most important answers are hidden inside a deep UI pattern that is inconsistently rendered, sparse in text, or hard to crawl through infrastructure and app layers.

Google’s guidance is not “never use accordions.” It is “make sure important content is available in textual form.”¹

If the answer to the buyer’s real question only shows up after three clicks, two scripts, and a storefront app render, you are adding failure points.

4) App widgets that humans see but machines do not trust

This shows up with:

review widgets,
comparison tools,
financing widgets,
shipping estimators,
and “compatibility finders.”

Humans may see them. AI systems may not get a clean, stable version of the data they contain.

The solution is not to kill all widgets. It is to promote the most important outputs into stable page text, markup, and merchant data.

What a machine-readable PDP actually looks like

A strong AI-commerce PDP is rarely fancy. It is usually just disciplined.

The page exposes core facts in visible text

This means the page should directly state, in plain language:

dimensions,
materials,
capacity,
compatibility,
included accessories,
warranty,
maintenance,
return terms,
and the main use case.

Not everything has to be above the fold. But the facts need to be there.

The page has a real spec table

If the product is even moderately technical, you want a proper text-based spec table.

For a coffee grinder, that might include:

burr type,
burr size,
motor power,
grind range,
hopper capacity,
dosing style,
dimensions,
weight,
voltage,
warranty,
cleaning requirements.

For cookware, it might include:

material,
coating,
diameter,
weight,
oven-safe temperature,
induction compatibility,
dishwasher safety,
warranty.

These are not just CRO details. They are comparison inputs.

The page’s markup supports the visible text

Google’s merchant listing docs say product markup can make the page eligible for merchant listing experiences that show price, availability, and shipping and return information.⁹

That means the structured layer should reinforce—not contradict—the visible page.

The page answers real questions in FAQ form

Shopify explicitly recommends adding FAQs for Perplexity shopping and says merchants can add FAQ schema so AI systems can identify and interpret those Q&A pairs.¹⁰

Good FAQ content covers things a buyer would actually ask, such as:

Is this compatible with induction?
Can this grinder handle espresso?
What comes in the box?
Is the bowl dishwasher-safe?
Does this fit a standard 58mm portafilter?
What is the return window?

The page connects product truth to merchant trust

Shipping, returns, and warranty should not be left to guesswork.

Google now supports merchant shipping policy data and merchant return policy data through structured data, and merchants can also provide shipping and return settings directly in Search Console.¹¹¹²¹³

This matters because recommendation systems are not only asking “What is the product?” They are also asking “Is this merchant safe to send a buyer to?”

The kitchen-and-coffee examples that matter for StoreSteady

StoreSteady’s launch niche makes this especially obvious.

Example 1: espresso machine pages

A lot of espresso machine PDPs bury the real differentiators:

boiler type,
heat-up behavior,
water tank capacity,
recommended skill level,
included portafilter baskets,
cleaning cycle,
and water line compatibility.

If those details live only in a brochure image or manufacturer PDF, AI systems have to improvise. That leads to weak recommendations and bad comparisons.

Example 2: grinder pages

This category is brutal because the buyer questions are technical.

“Best grinder for espresso under $400” is not answered well by a page that says “professional performance.”

It is answered by:

burr type,
grind precision,
retention,
dosing workflow,
noise,
footprint,
voltage,
and maintenance.

If the page does not expose those facts clearly, the product becomes hard to match.

Example 3: cookware pages

For cookware, buyers care about:

induction compatibility,
oven-safe temperature,
coating type,
PFOA/PTFE claims,
dishwasher safety,
handle material,
and warranty.

Those are exactly the kinds of details merchants love to put in polished infographics. They also happen to be the details AI shopping systems need in text.

What to fix first if your specs are trapped in visuals

Here is the short list.

1) Mirror the image facts into text

Do not redesign the whole PDP first. Just extract the high-value facts from the images and place them into readable text near the product description or in a spec table.

2) Create a normalized spec table

Use consistent field names across the category. That makes both comparison and downstream structured data cleaner.

3) Add or repair product markup

Make sure Product / Offer data supports price, availability, identifiers, and relevant merchant listing fields.⁹

4) Publish FAQ blocks

Use the actual questions customers ask, not generic filler.

5) Promote policy facts onto or near the PDP

Return window, shipping expectations, and warranty support should be easy to verify.

6) Keep the PDF only as an extra

The PDF can remain for power users. It should not be the only place the truth lives.

What not to do

Do not assume alt text solves this

Alt text helps provide image context, but Google’s own guidance frames image understanding around surrounding content, captions, titles, and page relevance—not as a substitute for proper product data.²

Do not leave critical facts in a custom widget only

If the widget fails or is not cleanly extracted, your facts disappear.

Do not use AI-generated fluff instead of concrete specs

Google’s product data spec is still about accurate fields. OpenAI’s feed docs still care about structured facts. There is no shortcut where vague “conversion copy” replaces product truth.⁷¹⁴

Do not create a mismatch between page text and markup

Google explicitly warns structured data should match the visible text.¹

Where StoreSteady fits

This problem is exactly why StoreSteady has both a crawl layer and a fix layer.

Truth Graph extracts what is actually present on the store and maps it against the attributes AI shopping systems need.
Replay shows the moment the model hesitates because your specs are hidden, incomplete, or weak.
Fixes turn image-only facts into text, spec tables, FAQ blocks, and stable product markup inside Shopify.

The goal is not to make your PDP uglier. It is to make the underlying truth easier to extract and trust.

FAQ

Can AI read images at all?

Sometimes, yes. But for product discovery and recommendation, platforms still rely primarily on structured product data and text for reliable matching and comparison.³⁷