AI Keyword Research Playbook

AI keywords | Keyword Research

KR0 | Keyword command map

Keyword work on Amazon should not end with a giant export file. It should end with a usable term set that can support indexing, organic ranking, and Sponsored Products at the same time.

SATLIS turns keyword work into a closed operating loop:

Keyword Miner -> Keyword Refiner -> Relevance Test -> Ranking Test -> AI Listing -> Campaign Architecture Engine

The goal is to move from raw vocabulary to executable terms.

What a keyword must do on Amazon

A good keyword is not just “high search volume.” It needs to survive four filters:

indexing eligibility: can the listing be indexed for the term?
retrieval opportunity: can the term actually surface the ASIN in search or recommendation flows?
ranking probability: can the ASIN compete for meaningful placement on that term?
monetization: can the term drive orders at a cost structure the business can support?

If a term fails any of those layers, it may still look good in a spreadsheet, but it will not behave like a real growth asset.

Why the old keyword workflow breaks

Traditional keyword tools usually over-focus on reverse ASIN scraping and obvious head terms. That leaves three gaps:

weak coverage of long-tail buyer language
weak coverage of scenario and constraint phrases
weak connection between keyword research, listing buildout, and ad execution

Under Amazon’s newer search behaviors, that gap matters even more. Buyers increasingly search with more natural phrases, and Amazon surfaces results through both classic keyword matching and broader intent understanding.

SATLIS keyword outcome

SATLIS is not trying to collect the biggest possible keyword dump. It is trying to produce a verified keyword package that can be used in three places immediately:

Product Title / Bullet Points / Description / A+ Content
backend Search Terms
Sponsored Products campaign structure

KR1 | Keyword Miner: expand the entry pool first

Keyword Miner starts from buyer intent, not just from obvious head terms.

The system analyzes competitor ASINs and breaks demand into structured clusters such as:

shopper use case
audience
pain point
comparison language
size or feature constraints
scenario terms

From there, SATLIS generates a much broader expression pool, including:

main commercial keywords
long-tail converting phrases
use-case language
natural questions shaped by Alexa for Shopping (formerly Rufus)
COSMO-style semantic expressions

Why this matters

If the entry pool is too small, the listing can only compete in a narrow slice of traffic. That usually leads to:

higher ad dependency
weaker organic growth
poor coverage of real buyer language

Keyword Miner solves the upstream problem first: build a bigger, cleaner starting pool before verification begins.

Recommended workflow

Select a focused group of competitor ASINs.
Run Keyword Miner tasks against those ASINs.
Review the output by intent cluster, not just by word count.
Send the mined terms into Keyword Refiner in batches.
Merge external sources such as ABA or third-party exports only after SATLIS intent output is in place.

SATLIS vs traditional keyword tools

Workflow dimension	Traditional tools	SATLIS
Starting point	reverse ASIN or volume table	competitor ASIN + buyer-intent reasoning
Language coverage	head terms and close variants	head terms + long-tail phrases + scenario language + semantic expressions
Output focus	list building	executable growth assets
Downstream use	mostly ad targeting	listing + Search Terms + Sponsored Products

KR2 | Keyword Refiner: turn the pool into an operating asset

Keyword Refiner is the control center for the term pool. It does three jobs:

import and consolidate terms from multiple sources
remove duplicates and obvious noise
prepare the pool for verification

This is where the keyword list stops being a file and starts becoming a working asset.

What belongs inside Keyword Refiner

mined terms from Keyword Miner
validated legacy terms from older campaigns
competitor-derived terms
imported external research, when useful

The important point is not how many terms enter the library. The important point is whether the library becomes easier to verify and easier to operate.

What sellers should manage here

term frequency across benchmark ASINs
duplicate cleanup
category-specific naming consistency
split between core terms, scenario terms, and expansion terms

Keyword Refiner should be the single source of truth before you test indexing and ranking probability.

KR3 | Relevance Test: confirm indexing eligibility

The first gate is Relevance Test.

It answers one question:

Can this ASIN reliably index for the term?

If the answer is no, that term should not sit in the main operating pool. A non-indexing term can still consume attention and budget, but it will not build durable organic growth.

What Relevance Test protects you from

writing listing copy around terms the ASIN cannot hold
funding Sponsored Products traffic on terms with weak structural relevance
overestimating the size of the usable keyword set

Practical output

After Relevance Test, the library should split into:

indexing-ready terms
conditional terms that may need listing or catalog changes
discarded terms that do not belong in the main pool

KR4 | Ranking Test: confirm order potential

Passing indexing is not enough. A term may index well and still be a poor operating choice.

That is why SATLIS runs Ranking Test after Relevance Test.

It answers the second question:

If the ASIN can index for the term, does the term have enough ranking probability and order potential to deserve budget or listing real estate?

Signals behind Ranking Test

search intent strength
fit between the product offer and buyer expectation
competition density
likely click-through behavior
conversion economics

The result is a smaller but much more valuable set of terms:

terms worth using in launch or scale campaigns
terms worth embedding in the listing
terms that should stay in observation or test queues

KR5 | Turn verified terms into listing and ads

Once a term passes both gates, it becomes an executable asset.

Use verified terms in AI Listing

Place the terms into the right listing fields according to their job:

Product Title for primary commercial and intent-defining terms
Bullet Points for use cases, differentiators, and objection handling
Description for supporting relevance and buyer education
A+ Content for visual proof, comparison logic, and conversion support
backend Search Terms for additional indexing support

The goal is not to stuff every term everywhere. The goal is to give Amazon and buyers a consistent, high-signal description of the product.

Use verified terms in Sponsored Products

Verified terms should then flow into the Campaign Architecture Engine:

high-confidence terms form the core growth layers
exploratory terms stay isolated in test layers
low-value terms stay out of the main spend path

This prevents the ad account from getting larger while becoming less clear.

KR6 | Operating discipline

Keyword research only works if the cycle closes.

Review the system on a fixed rhythm:

add new term discoveries from search term reports and competitor movement
retest terms when the listing changes materially
remove terms that lose relevance or economics
promote newly proven terms into the main structure

The end state should be simple:

the listing language matches the winning search language
the Sponsored Products structure matches the verified term set
organic and paid growth reinforce each other instead of drifting apart

Official terminology references

Amazon Ads keyword targeting guide: https://advertising.amazon.com/library/guides/keyword-targeting
Amazon Ads Sponsored Products guide: https://advertising.amazon.com/library/guides/targeting-with-sponsored-products
Amazon Seller guidance on search terms: https://sellercentral.amazon.com/forums/t/optimize-your-search-terms/839864
Amazon listing AI overview: https://sell.amazon.com/blog/amazon-listing-ai

After the task is started, the interface is as shown below:

Figure 39 Inclusion ratio test in progress

After waiting for the task to be completed, a green check mark will be displayed, as shown in the figure below:

Figure 310 Collection ratio task successful

The second gate: opportunity verification (deciding whether it is worth fighting)

This gate answers only one question: does the term have the potential to make a sale and is it worth investing budget and resources into?

Operation process:

Select the required [Keyword Refinement Library] to enter the initial keyword library, and click [Step 2: Create a new keyword order probability task]:

Figure 311 Create a single probability task

In the pop-up dialog box, click [Automatically create in batches]:

Figure 312 Automatically create batch tasks

You can select the number of keywords for a single batch task according to the computer configuration, and a front-end crawler task will be created, which mainly consumes memory:

Figure 313 Select the number of keywords for batch tasks according to computer configuration

The subsequent steps are the same as the [Collection Ratio Task], just wait for the task to be completed.

Accurate keyword generation (20,000 → 706)

When the two types of tasks are completed:

Select all tasks → click [Generate precise keywords] → get the executable vocabulary library:

Figure 314 Click to generate precise keywords

The system automatically selects precise keywords based on the traffic distribution mechanism within Amazon's site. In the new dialog box that pops up, click the button [Continue Generation], as shown in the figure below:

Figure 315 Click to continue generating

Finally, 706 precise keywords were obtained, as shown in the figure below:

Figure 316 Obtained 706 precise keywords

One sentence to remember: Purify the entry pool into: word packages with natural position opportunities + transaction potential.

Precision word application scenario A: copywriting burying words (making the Listing "recognized by the system")
**Goal: Improve collection and recall, so that Listings can be "found" in more real-life scenarios. **

Figure 317 Precise keywords - words buried in copywriting

Key points of operation (only do 3 steps):

Keyword stratification (dividing labor first and then copywriting)

Core word: decide the main intention (1 group)

Scenario/Constraint Word: Determine the long-tail entrance (multiple groups)

Variant words: synonyms, spellings, combined expressions (complementary coverage)

Slot allocation (put words where they should appear)

Title: Put “Core Intent + Key Points of Difference” (few but accurate)

Bullets: Each item is bound to 1 "intention cluster" (giving the paragraph a task)

Backend Search Terms: Put words that “can be included but are not suitable for the text pile” (complementary coverage)

A+/QA: Provide more natural scene expression (undertake long-tail and conversational expression)

Verification after publication (use the results to push back the copy)

Check first: whether the included coverage is up to standard (not included = written in vain)

Let’s look again: whether the natural position can stably enter the visible range (visible = only growth can occur)

**One sentence summary: Precision words first solve "qualification and recall", and copywriting then solves "conversion and persuasion". **

Accurate word application scenario B: Advertising cold start (let the word "bring orders" as quickly as possible instead of burning money)

**Goal: Use the minimum trial and error cost to quickly get the "words that can be typed" out of the order and have a replicable structure. **
During new product promotion, precise keywords can directly cold-start user ads:

Figure 318 Precise keywords - "Advertising cold start"

Common misunderstanding: Why “the more you do, the more expensive you become, so you won’t be able to afford it alone”

Misunderstanding 1: Treating "initial vocabulary library" as "accurate vocabulary library"

Typical performance:

There are a lot of words, and it looks like I’m trying very hard, but naturally I can’t get up.

The ads are becoming sparse and sparse, and ACoS/TACoS cannot be promoted.

Essential reasons:

The initial vocabulary is only an "entry candidate", which does not mean that the system will approve it, nor does it mean that the transaction will be completed.

If you don't run "Qualification (Inclusion) + Opportunity (Probability of placing an order)", it is equivalent to using "unverified entrance" to burn your budget.

Correct approach:

First make a large entrance pool, and then use two gates to purify:

Proportion of inclusion (qualification) → Probability of placing an order (opportunity) → Generate precise keywords

Only “precise words” are used in copywriting and main advertising

Misunderstanding 2: Write the copy first and then verify the inclusion (only to find out "not included" after writing)

Typical performance:

The copywriting is very full, but the actual collection is very small.

I thought it was because the advertisement was not effective, but the result was "The system does not recognize the entry you wrote."

Essential reasons:

Copywriting is a "conceptor", not a "conveyor"

The entrance is not qualified, and no matter how good it is, it will not receive natural traffic.

Correct approach:

The order must be reversed:

First verify the inclusion → then allocate slots → then write the acceptance copy

One sentence to remember: let the system recognize it first, and then let the buyer believe it