What Amazon Means by Relevant: The ESCI Model

Most sellers use the word relevant loosely. Amazon does not. When Amazon released a public research dataset of real shopping searches, it labeled each search and product pair with one of four grades. That labelling scheme, known as ESCI, is the clearest public statement we have about what Amazon means when it says a product is relevant to a search.

It is worth understanding because it changes how you audit a listing. Instead of asking whether your keywords are good, you start asking whether a shopper looking at your page would say yes, that is exactly the thing I searched for.

Reviewed by the SellerShorts editorial bench. SellerShorts runs an AI tool marketplace for Amazon teams.

Our Amazon Listing Optimizer takes an ASIN and returns a full optimized listing, covering title, bullets, description, and backend keywords, in one run. Push live to Seller Central in one click.

What ESCI actually is

ESCI is a relevance labelling scheme that came out of Amazon's own research. Amazon published a large public dataset of real shopping searches paired with products, and every pair carried one of four labels. The dataset was released so researchers could build and test better product search systems.

Two things are worth being clear about before you use it.

• It is a research dataset, not a live ranking dial. Amazon has not said that live search assigns these exact grades to your ASIN in real time.
• It is still the best public definition we have. It shows how Amazon's own people think about relevance, and that is more useful than a guess from a forum.

The four grades at a glance

Here is the whole model in one table, with a plain example for each grade. Imagine a shopper searching for a stainless steel 20 oz insulated water bottle.

Exact: the only grade you want

Exact means the product fully satisfies everything the shopper asked for. That includes the parts of the search that are specs, not just the parts that name the product category.

This is where sellers usually lose. A search often carries several requirements at once, and a listing that answers three of four is not treated as almost right.

• The category has to match. The obvious part, and the part most sellers get right.
• The named specs have to match. Size, count, color, material, and compatibility when they appear in the search.
• The implied specs have to match too. A search for a laptop stand for a 16 inch laptop implies the stand supports that size, even though the shopper never used the word supports.

Substitute: close, but not what they asked for

A Substitute is a product a shopper might buy instead of the thing they searched for. It is in the right neighbourhood and does a similar job, but at least one part of the request is not met.

Substitutes are not failures. Plenty of sales come from shoppers who searched narrow and then bought something adjacent. But you do not want to be a Substitute for the searches you built your product around.

• Wrong size or capacity. The right product in the wrong dimension.
• Wrong pack count. A single unit when a multipack was searched for, or the reverse.
• Wrong material. Plastic when the search said steel, or synthetic when the search said cotton.
• Wrong finish or variant. The right item in a color the shopper did not ask for.

Complement: it goes with the thing, it is not the thing

A Complement is a product that pairs with the searched item without being it. Cases, cables, refills, brushes, filters, and mounts are the usual examples.

Complements matter for two reasons. First, if you sell accessories, being clear that you are the accessory keeps your listing honest and reduces returns. Second, if you sell the main product, an accessory sitting in your search results is competing for the click without competing for the sale.

The practical move for accessory sellers is to state the main product you work with, by name and model, so a shopper searching for the accessory finds you as an Exact match for their search rather than as a Complement to someone else's.

Irrelevant: the grade you can fall into by accident

Irrelevant means the product does not answer the search. Most sellers assume this only happens to products in completely wrong categories, and that assumption is the expensive part.

You can land here through omission. If a shopper searched with a spec that is central to what they want, and your page never states that spec, there is nothing on the page that answers the search on that point. Your product may be perfect in reality and still fail to show that it is.

The one spec problem, and why it is not a small drop

The most useful takeaway from the ESCI model is that relevance is stepped, not smooth. Missing a central spec can move you a full grade or more, not a few positions.

Think about how a shopper would judge it. If someone searched for a 6 foot USB C cable and you sell a 3 foot one, they are not going to say it is 50 percent right. They will say it is the wrong cable. The grade drops in one jump.

• A spec is central if a buyer would return the item over it. That is the simplest test.
• Central specs need to be visible, not implied. If it only exists on the back of the box, it does not exist for search.
• Being close does not earn partial credit. The model has grades, not a percentage.

The specs that usually decide the grade

Across most categories, the same handful of specs are the ones that flip a listing from Exact to something lower.

How to audit your listing against ESCI

This is a short exercise you can do with a notepad, and it finds real gaps fast.

• Pick your top five searches. The ones you actually want to win, not every phrase you could rank for.
• Write down every spec each search implies. Include the ones the shopper did not type but clearly expects.
• Open your live listing and hunt for each spec. Title, bullets, description, and the attribute fields.
• Mark every spec you cannot find. Each miss is a place where you could drop below Exact.
• Fix the attribute fields first. They are structured data, so they are the cleanest place to state a spec.
• Then state the same spec in words. Attributes help systems, the visible copy helps shoppers.

Common mistakes this model exposes

Once you look at listings through the four grades, some familiar habits look worse than they did.

• Leaving pack count out of the title. Very common on multipacks, and it costs you every multipack search.
• Hiding compatibility in the description. Shoppers searching for a fit will not scroll that far, and neither should you rely on it.
• Writing a poetic title with no specs in it. Brand plus adjectives reads nicely and answers nothing.
• Chasing broad head terms while missing the spec words. The broad term gets you seen, the spec words get you bought.
• Blank attribute fields. Leaving them empty removes the cleanest signal you have.

What ESCI does not say

Being honest about the limits keeps you out of trouble.

• It does not give you ranking weights. Amazon has never published how much relevance is worth versus price or sales history.
• It does not confirm a live grade on your ASIN. The grades came from a research dataset, not from a Seller Central report you can look up.
• It does not replace conversion work. Being an Exact match gets you into the running. Images, price, and reviews still decide what happens next.

Conclusion

Amazon's own research defines relevance in four grades, and the gap between them is bigger than most sellers assume. Exact means every part of the search is answered, including the specs. Slip on one central spec and you can fall further than you expect.

The fix is not clever. Write down what each of your target searches implies, then make sure your page says it, in the attribute fields and in plain words. For the field by field build, our Amazon Listing Optimizer does it in one run. Next reads: why Amazon attributes matter more than keywords, how Amazon search works, matching and ranking, and how Amazon handles vague searches.

References

1. Shopping Queries Dataset: A Large-Scale ESCI Benchmark for Improving Product Search (arXiv 2206.06588)
2. Amazon Science: ESCI shopping queries dataset
3. Amazon: Amazon SEO seller guidance