Our product page is one long image
With materials, dimensions, instructions and precautions all inside an image, the text a crawler reads is effectively zero. There is no sentence for an answer engine to cite in the first place.
ECOMMERCE
E-commerce AEO —
a product page that is one image
is a page AI cannot read
Customers ask AI to recommend a product. Yet product pages on Korean online stores are mostly one long image, leaving crawlers effectively no text to read.
In one paragraph
E-commerce AEO is the work of designing things so that when a customer asks AI to recommend a product, or narrows candidates by attaching conditions, your product information is cited as evidence. This industry has one structural bottleneck the others do not — product pages on Korean online stores are mostly a single long designed image, and the materials, dimensions and instructions inside it are information that does not exist as far as a crawler is concerned.
So AEO here does not start with making more content; it starts with getting the information already inside the image out into a form machines can read. It is work that can be done without changing the screen people see, and it usually takes effect faster than writing something new.
Product pages converted to text firstProduct structured dataNo 'lowest price' or 'number one' claims
Last verified
WHEN YOU NEED THIS
When you have these problems,
this is the work you need
These are the situations we hear repeatedly in consultations. If any of them apply, start by measuring.
Our product does not come up in recommendation questions
The answer to a recommendation question usually comes from review blogs and curation pieces. The reason the product page cannot take that slot is generally that its conditions are not written as text.
Marketplaces and our own store get treated as different brands
When the company name, brand name and product name differ slightly per channel, an answer engine cannot bundle them. The brand signal splits across as many channels as there are, and weakens.
Prices and stock get answered wrongly
Either the on-screen value and the structured data value diverge, or an old cache is being referenced. A wrong price loses trust at the moment immediately before purchase.
WHAT WE DO
What Navirang
actually does
Written as units of work rather than abstract proposals. The scope of an engagement is set from this list.
Converting product pages to text
We bring the information inside the image out as on-screen text. The default method is leaving the same content as text below the image or in a collapsible area, without changing the design.
- Materials, dimensions, capacity and compatibility separated out as tabular text
- Instructions, washing, storage and precautions stated as itemised lists
- Image alt corrected to the actual content rather than decoration
- Collapsible areas built with details — the full text stays readable to a crawler even when collapsed
Breaking down recommendation and conditional questions
Questions get organised the way customers look for a product. The axes of those conditions decide which items have to appear on the product page.
- Use case — asking for a recommendation for a given situation
- Condition-matching — attaching price band, dimensions, ingredients or compatibility as conditions
- Comparative — asking which of two products or two approaches is better
- Use and care — asking how to use and maintain it
Applying Product structured data
The facts written on screen get re-declared in a machine-readable form. Where the screen and the value disagree, the result is worse than not applying it at all.
- Product name, brand, identifier, price, currency, availability
- Connecting the on-screen value and the schema value so they read the same source
- No rating field where no ratings have actually been collected
Consolidating the brand entity
Notation scattered across your own store, marketplaces and social channels gets aligned to one form. The more channels an industry has, the greater the effect of this work.
- Fixing the canonical rules for brand name, company name and product name
- Building a per-channel notation comparison table, then correcting in sequence
- Declaring them the same entity through structured data and channel links
Checking mandatory disclosure consistency
We confirm the information required for e-commerce disclosure does not diverge from what is on screen. It is both regulatory compliance and the building of citation evidence.
- Whether the required product information disclosure items are stated
- Price, delivery, exchange and return terms consistent across channels
- Removing expressions that would fall foul of labelling and advertising rules
Re-measurement and correcting misinformation
We measure repeatedly under the same conditions, and where AI answers a price or specification incorrectly, we build the corrective evidence and track it.
- Fixed question wording, repeated logged-out sessions
- Counting whether the product appears as a recommendation candidate
- Recording the incorrect sentence and the source it appears to have drawn on
PROCESS
In what order
does it run
What you receive at each stage is stated alongside it. Durations are the working time Navirang controls; they are not a promise about when results appear.
- 01 1–2 days
Product range and scope
We confirm the main product lines, the sales channels, and how the product pages are produced.
Scope definition
- 02 2–3 days
Question type breakdown
Questions get organised along use case, condition, comparison and care, and the set to be measured gets fixed.
Question set document
- 03 3–5 days
Measurement and candidates
We put the questions to the answer engines and record the products currently recommended and the domains cited.
Observation record · recommendation candidate tally
- 04 Depends on product count
Product page text conversion
Information inside the images comes out as text and gets rebuilt in tabular form, applied to the lead products first.
Product page text structure plan
- 05 2–3 days
Structured data and notation
Product schema gets connected to the on-screen values, and brand notation gets compared and aligned across channels.
Schema application plan · notation comparison table
DELIVERABLES
What you
receive
We do not do work that ends in conversation. The documents below remain, and become the baseline for the next measurement.
Product page text structure plan
Which items to extract from the images and in what order to place them, delivered as a template per product line.
Question set document
The full set of customer questions used in measurement, with their condition-axis classification. It is the basis for re-measurement, so the client keeps it.
Recommendation candidate tally
A table of the products currently recommended per question and the domains that were the evidence, with whether your products appear written alongside.
Schema application plan
The value for each Product structured data field, connected to the on-screen element it comes from.
Notation comparison table
The differences in brand name, company name and product name across your own store, marketplaces and social channels, with a correction order attached.
What decides citation in this industry
- The biggest bottleneck
- A product page that is one long image — a state where the text a crawler reads is effectively zero
- What comes first
- Not adding content but converting the information inside the images to text. It can be done without changing the design
- Entry requirements for conditions
- Materials, dimensions, capacity, compatibility and price band have to be written as text to match a conditional question
- Structured data
- Connected so it reads the same source as the on-screen value. A schema whose values diverge is worse than none
- What we do not use
- 'Lowest price', 'number one', 'the only one in Korea' superlatives, and rating fields with no measurement behind them
Navirang does not manufacture review counts or ratings for you. Putting a rating into the schema when no ratings have been collected declares to machines something that is not on the screen — an item this site has prohibited for itself. Evidence of trust gets built from confirmable facts: specifications, ingredients, certifications, delivery and exchange terms, and the update date.
HOW IT CONNECTS
How it connects
to the other work
Our work moves as one piece. SEO builds the foundation for being found by search engines, AEO raises the odds of that information being cited in an answer, structured data helps machines understand the facts, and content supplies the evidence there is to cite.
| Area | Relationship to this work |
|---|---|
| AEO audit | The single measurement of current appearance using recommendation and conditional questions |
| Structured data | Declaring the on-screen price and specifications as machine-readable facts |
| Citable content design | Rewriting the information extracted from images into citation-sized sentences |
| Entity SEO | Binding several sales channels into one brand |
| AEO by industry | The parent page for comparing question types across other industries |
| B2B and SaaS | The same condition — questions that narrow candidates by attaching conditions |
FAQ
Frequently asked questions
Q Why is it a problem that our product page is an image?
A Because there is no text for an answer engine to read. Product pages on Korean online stores are mostly a single long designed image, and the materials, dimensions and instructions written inside it do not exist as far as a crawler is concerned. To a person the page looks rich with information, while the sentences a machine can cite are close to zero. So AEO in this industry does not start with writing new content but with getting the information already inside the image out as text. The method leaves the design as it is and places the same content as text alongside, so the selling screen is untouched.
Q Isn't filling in the image alt enough?
A It is not. Alt is a short sentence describing one image, unsuited to carrying a full specification table or a set of instructions, and pushing long text into alt is poor practice from an accessibility point of view as well. What is needed is body text. Items such as materials, dimensions and capacity belong on screen as a table, and instructions and precautions as an itemised list. Even collapsed, using a details element rather than a JavaScript accordion keeps the full text readable to a crawler. Correcting alt is the next task after that.
Q We have thousands of products. Do we have to fix them all?
A Not all of them — in order. The audit first confirms which questions actually arise, and the product lines connected to those questions get done first. And rather than fixing individual products by hand, we build a template per product line — once the item structure for materials, dimensions and instructions is settled, newly listed products only need their values filled in. Products already listed get worked backwards from the largest revenue contributors. Done in this order, the initial workload does not scale with the number of products.
Q Can we do AEO if we only sell through marketplaces?
A The range of what is possible changes. On a marketplace product page the area under your control is limited, and you do not set that domain's crawling policy. There is still work to be done — unifying product and brand notation across channels, filling the detail information in as text, and stating the required product information disclosure items accurately. But with no domain of your own for an answer engine to cite directly, there is also nowhere to accumulate a brand signal. We will say plainly that having your own store, or at minimum a brand site, is the stronger position over the medium term.
Q AI states our product's price incorrectly.
A Either the on-screen value and the structured data value have diverged, or older information is being referenced more often. First we check whether the price, currency and availability in the Product schema read the same source as the screen. A schema with values typed in by hand diverges every time the price changes, so connecting it to read the same data as the screen is the underlying fix. Then old price statements left on other channels get cleared, and the same question gets measured repeatedly to track whether the answer changes. A wrong price is the type that loses trust immediately before purchase, so we hold it at high priority.
Q Can we not write 'lowest price'?
A We do not recommend it. Superlatives such as lowest price, number one or the only one in the country become a labelling and advertising problem when used without objective basis, and they easily stop being true as time passes. From an answer engine's point of view they are unverifiable claims and do not serve as citation evidence. What can go in the same place is a fact with its conditions attached — pack quantity, price per unit volume, delivery terms, warranty period, values that can be confirmed. Those values actually match in conditional questions, so they perform better than a superlative.
Q We have few reviews. Does that hurt us?
A Few reviews does not exclude you from citation. What an answer engine references when answering a product question is not only ratings but the product's factual information, and that part we can fill in without reviews. If anything, a product with accurate specifications, ingredients, compatibility and instructions has the advantage in conditional questions over one with many reviews whose information is trapped in an image. What we will not do is manufacture ratings that do not exist — stating a rating when none has been collected declares to machines something that is not on the screen, an item Navirang has prohibited for itself.
Does your product come up in recommendation questions?
Give us the product lines and the store address and we will run the free audit on recommendation and conditional questions. We also measure how many characters your product page amounts to for a crawler.
We reply within one business day.