FPAI Business Logic Documentation

Smart Listing Optimizer

A unified platform that takes an Amazon product and produces an optimized listing grounded in real customer intent, differentiated from competitors, keyword-rich, compliance-safe, and AI-visible.

Then tests it via A/B experiments, creates off-Amazon content for AI platform visibility, and proves the results with before/after measurement.

View Pipeline Data Flow

The Pipeline

Each module produces structured data that feeds the next. One continuous assembly line from product intake to proven results.

UNDERSTAND

ANALYZE

M0/M1Product Intake

LQSBaseline Score

M1.5AI Landscape Scanoptional

M2Competitor Discovery

M2.1Customer Intent

DIFFERENTIATE

BUILD

VALIDATE

M2.2USP Extraction

M2.3USP Evaluation

M3Keyword Intelligence

M4Listing Creation

LQSQuality GateGATE ≥ 70

score ≥ 70

Track A — Amazon

M5MYE A/B Testing

Track B — AI Visibility

M6Off-Amazon Contentoptional

30 days later

Measure

M7AI Rescore Loopoptional

Hover nodes for details

optionalCan be skipped

GATEMust pass to proceed

How data flows

Every module consumes artifacts from upstream modules and produces new artifacts for downstream modules. The product profile feeds into competitor analysis, which feeds into customer intent, which feeds into USP extraction, and so on. No module operates in isolation — each decision is grounded in the accumulated intelligence of all previous steps.

Pipeline Modules

20+

Data Artifacts

Quality Dimensions

MODULE 0

No-ASIN Product Intake

Entry point for products that don't have an Amazon listing yet.

When this is used

Pre-launch

Product isn't on Amazon yet — seller wants an optimized listing ready for launch day

Demo runs

Testing the platform without committing a real product

Cross-platform

Product exists on Walmart/Shopify/eBay but not Amazon

Documentation only

Seller has spec sheets, manuals, or marketing PDFs but no live listing

Three additional input channels

Product URL

Link to the product on another marketplace. System scrapes it to pre-fill form fields.

Image Uploads

Product photos, packaging shots, ingredient labels, certifications. Vision AI extracts text, specs, and certifications.

Document Uploads

PDFs or Word docs (spec sheets, safety data sheets, marketing materials). System extracts structured product data.

The merge rule

User-typed fields always win. Everything else fills in blanks.

If the user typed “500ml” and the URL scrape says “16oz”, the system keeps “500ml”. If the field is empty, the system fills it with whatever was found.

Output

product_profile — the exact same data structure as Module 1. Everything downstream works identically regardless of entry point.

MODULE 1

Product Context

Standard entry point when the product already has a live Amazon listing (has an ASIN).

Step 1: Product Data Intake

Required: Market, ASIN, MSKU

Optional: Brand name, product features, up to 10 images

If ASIN is provided, system fetches public Amazon data: current title, bullets, description, category, images, rating, review count.

Step 2: Product Profile Creation

AI analyzes all available information to determine:

Product type and purpose

Key attributes and specs

Target audience

Category and subcategory

Brand tone and positioning

Initial keyword seeds

Output

product_profile — Product identity, category, attributes, features, benefits, target audience, use cases, brand tone, and keyword seeds. The foundation for everything that follows.

MODULE 1.5

AI Landscape Scan

Scan what AI platforms (ChatGPT, Perplexity, Google AI) say about the product before any listing work begins.

Why this exists

Traditional Amazon SEO (A9/A10) has only 22% overlap with AI shopping recommendations. A product optimized only for Amazon search is invisible to shoppers who ask ChatGPT or Perplexity for advice.

How it works

Generate 20-30 Shopper Questions

AI creates questions real shoppers would ask AI assistants — not search terms, but conversational queries like "What's the best manuka honey for immune support?" across 6 intent types: best-of, comparison, problem-solution, attribute, use-case, safety.

Query AI Platforms

Send each question to ChatGPT (API), Perplexity (API), Google AI Overviews (SerpAPI), and Amazon Rufus (manual). Record every response.

Analyze Responses

Extract: product mentions and position, citation sources (which URLs/sites the AI references), sentiment, competitor set, language patterns, and hallucination checks.

Outputs

ai_visibility_baselineHow visible the product is across AI platforms (e.g., mentioned in 3/28 queries)

ai_gap_reportWhich queries/platforms the product is missing from, what external sources are missing

citation_source_mapWhich external sources (Reddit threads, blogs, YouTube) the AI platforms are actually citing

Feeds forward to

M2 — AI-recommended competitorsM2.1 — Real AI platform questionsM3 — Natural language query patternsM4 — GEO-optimized writing signalsM6 — Gap & citation data for content

MODULE 2

Competitor Discovery

Find, analyze, and select the products that compete directly with the user's product on Amazon.

Three-step process

Competitor Collection

System suggests 10 search terms based on the product profile. User selects 5. System searches Amazon (20-25 results per term), captures title/brand/price/rating/bullets, deduplicates. Result: competitor_list_raw (40-50 listings).

Relevancy Scoring

Each competitor is scored on: category alignment, shared features, similar positioning, target audience overlap. A Sales Velocity Score is estimated where possible.

User Selection

Top 20 displayed (competitor_list_trimmed). User picks 5-10 true market comparables (competitor_list_final). User can drag and reorder by priority.

Why two lists matter

Trimmed (20 competitors)

Broad market context. Used for frequency analysis — “how common is this claim in the market?”

Final (5-10 competitors)

The user's actual competition. Used for strategic decisions — “is this USP differentiated against my direct rivals?”

MODULE 2.1

Customer Intent Extraction (Rufus)

Understand what customers actually care about — independent of what the product claims or competitors say.

Step 1: Collect Raw Signals

Sources:

Competitor reviews (titles + bodies)

Competitor Q&A (questions on PDPs)

User's own product reviews

Long-tail search queries from category

AI-inferred questions

Review statements are converted to questions: “This serum feels sticky” becomes “Will this serum feel sticky?”. Result: 50-200 raw signals.

Step 2: Cluster Into Themes

AI groups signals into themes (e.g., Texture, Ingredients, Performance, Safety, Value). Each theme gets:

Frequency Score (0-100)

How often customers mention this

Importance Score (0-100)

Complaint severity, desire strength, emotional tone

Step 3: Package Output

Top 5-10 intent themes with scores

Curated customer questions (up to 10)

Pain points — frustrations ("sticky texture", "doesn't absorb")

Desire points — outcomes sought ("long-lasting hydration")

Natural search-language phrases

Feature-Interest Map — which features customers care about most

Structured tags (TEXTURE, HYDRATION, etc.) for cross-module reference

MODULE 2.2

USP Extraction

Identify every possible unique selling point by analyzing the product, competitors, and customer intent.

Three data sources compared

Product Profile

What does the product claim? Features, benefits, ingredients, specs.

Competitors (20)

What do competitors emphasize? Common claims, repeated selling points, and gaps.

Customer Intent

What do customers care about? Themes, pain points, desire points.

What the AI finds

The AI compares across all three sources to identify:

Product features competitors don't mention

Gaps in competitor messaging that the product can fill.

Customer concerns competitors fail to address

E.g., customers ask about "stickiness" but no competitor addresses it.

Advantages aligned with high-importance themes

Product attributes that match what customers care about most.

Output

usp_raw — Each potential USP with: clean text, competitor frequency, product presence (0/1), impact score (0-100), linked intent themes, pain and desire points.

MODULE 2.3

USP Evaluation

Score and rank USPs to determine which ones are most valuable for the listing.

Three scoring dimensions

45%Customer Relevance

How strongly this USP aligns with what customers care about. Uses theme frequency, importance, and pain/desire alignment. Pain-heavy USPs score higher — solving a problem beats fulfilling a want.

25%Competitive Uniqueness

How differentiated vs. competitors. Formula: (1 - competitor_frequency / total_competitors) x 100. Being unique but irrelevant gets penalized.

30%Market Impact Potential

How much this USP can realistically move CTR, CVR, and revenue. Based on impact_score from extraction.

Formula

Total USP Score = 0.45 x CR + 0.25 x CU + 0.30 x MI

Validation & Selection

High-scoring USPs are validated for factual accuracy, Amazon compliance, clear phrasing, and content fit. Then selected:

Primary Set (3-5 USPs)

High scores, strong validation, diverse themes. Lead the listing.

Secondary Pool

Good scores or niche themes. For supporting bullets, A+ content, MYE retests.

MODULE 3

Keyword Intelligence

Build a comprehensive, scored keyword strategy from five data sources.

Five keyword sources

Product Profile

What the product is

Competitors

How the market describes it

Customer Intent

How customers search

External Tools

Search volume data

AI Expansion

Synonyms & variants

Scoring formula

base = 0.60 x product_intent_relevance + 0.20 x competitor_alignment + 0.20 x search_demand

keyword_strength_score = min(100, base + usp_bonus)

USP bonus: +10 (strong), +5 (weak), +0 (none)

Tier	Score	Description	Placement
Primary	>= 75	Most important, defines the product	Title, Bullet 1
Secondary	60-74	Strong support, reinforces positioning	Later bullets, description
Long-tail	40-59	Specific, high-intent phrases	Backend, description
Excluded	< 40	Weak, irrelevant, risky, or competitor-only	Not used

USP Keyword Bundles

For each approved USP, all supporting keywords are grouped into a bundle (primary, secondary, long-tail). Neutral category keywords stay separate as baseline terms. When MYE tests different variants, each variant emphasizes a different USP bundle while keeping category keywords constant — making tests clean and attributable.

MODULE 4

Listing Creation

Turn approved USPs and keyword bundles into an actual Amazon listing. Hybrid approach: rules decide structure, AI writes copy.

Inputs

Product Truth Set

Facts from the product profile. The listing can only claim what's in here.

USP Package

Approved USPs with safe phrasing, proof points, placements.

Keyword Package

Keywords with scores, tiers, and USP links.

Compliance Config

Banned terms, risk words, tone guidance, character limits.

Five-step creation process

Step A — Build Listing Blueprint (Rules Only)

Decide: which USP leads, USP order for Bullets 1-5, which keywords go where, minimum inclusion rules, dedup rules, compliance constraints. No AI — pure structural planning.

Step B — Draft Copy (AI Within Blueprint)

AI generates title, bullets, description, backend terms — constrained by the blueprint. Must follow: must-include keywords, allowed USP phrasing, banned terms, character limits.

Step C — QA Pass 1 (Automated Rules Check)

Validate every section: length limits, required keywords included, no keyword stuffing, no banned terms, USP alignment per bullet. Output: pass/fail per section.

Step D — Targeted Rewrite (AI Fixes Failures)

Only failed sections are regenerated with explicit instructions: "Remove banned term X", "Include missing keyword Y", "Reduce by N characters." Passing sections untouched.

Step E — QA Pass 2 (Final Check + Fallbacks)

Re-check everything. If still failing, apply deterministic fallbacks: trim lowest-priority phrase, swap keyword for same-USP alternative, replace with pre-approved safe phrasing.

Key principles

Truth-first

Never invent facts

Compliance-first

Risk-high language never enters

Structure-first

Rules decide, AI writes

No stuffing

Reuse caps + section quotas

Explainable

Every phrase maps to a source

LQS

Listing Quality Score

Scores listings across 6 dimensions. Used twice: before optimization (baseline) and after (quality gate).

LQS runs at two points in the pipeline

LQS BASELINE

Before optimization (after M1)

Scores the existing live listing using the same 6 dimensions. Establishes a measurable starting point so the client can see exactly where their listing is weak and how much improvement the platform delivers.

Input: current Amazon listing (fetched via ASIN)

Output: lqs_baseline_report

LQS GATE

After optimization (after M4)

Scores the optimized listingproduced by Module 4. Must score ≥ 70 to proceed to MYE testing. Below 70 blocks the listing and requires fixes. This ensures only strategically strong listings reach customers.

Input: M4 final listing + all pipeline data

Output: lqs_report (gate decision)

Six quality dimensions

Keyword Optimization (25%)

Keywords in the right places?

USP Effectiveness (20%)

Approved USPs present and unique?

Readability (15%)

Flesch score, scannability

Competitive Position (15%)

Differentiated vs. competitors?

Customer Alignment (15%)

Addresses pain points?

Compliance (10%)

Zero banned terms, format rules

Before/after comparison example

BASELINE (before)

57.3

Grade F

OPTIMIZED (after)

78.5

Grade C — MYE eligible

The delta between baseline and optimized LQS is the measurable proof of the optimization's value.

Grade	Score	MYE Eligible
A	90-100	Yes (auto-approve)
B	80-89	Yes (auto-approve)
C	70-79	Yes (review recommended)
D	60-69	No (require fixes)
F	< 60	No (major rewrite)

MODULE 5

MYE Integration (A/B Testing)

Test the listing on Amazon via Manage Your Experiments. One attribute at a time.

Core principle

One attribute at a time, in sequence: Title first, then Bullet 1, then Bullets 2-5, then Description.

If you change title and bullets simultaneously and performance improves, you don't know which change caused it.

Six-step process

Preparation

Pull all inputs, validate content, create Test Subject Registry. AI generates 2-3 title variants: USP-led, keyword-led, balanced.

Experiment Setup

Create experiment via MYE API. Set control (current live) and treatments (AI variants). One experiment per ASIN per attribute.

Data Collection

Automated daily metric retrieval: impressions, clicks, CTR, CVR, sessions, sales. Typically runs 2-4 weeks.

Result Evaluation

Win: replace live listing, move to next attribute. Loss: AI analyzes patterns (CTR drop = unclear USP, stable CTR but low CVR = disconnect), creates Retest Recommendation.

Retest Queue

Maintains queue per ASIN: Running/Passed/Failed/Retest Pending. Failed tests get new variants that address identified issues.

Learning Record

Every test result is saved: which USP + keyword + tone combinations produced the strongest performance. Cross-product learnings improve future optimizations.

MODULE 6

Off-Amazon Content Package

Create external content for AI platform visibility. AI assistants pull from Reddit, blogs, YouTube — not Amazon listings.

Seven content types

Step	Content Type	Generation	Publishing
1	Content Strategy	Fully automated	Prioritized action plan
2	Reddit Content	Automated	Manual (needs karma)
3	Quora/Forum Answers	Automated	Manual
4	Blog/Articles (GEO-optimized)	Automated	Semi-auto (CMS)
5	YouTube/Video Scripts	Automated	Manual (record)
6	Outreach Emails	Automated	Semi-auto (send API)
7	Technical SEO/AEO	Automated	Manual (dev implements)

GEO visibility impact

+42.6%

Expert quotes

+32.8%

Statistics

+28.7%

Fluent writing

+27.7%

Source citations

Technical AEO package

robots.txt — Allow AI crawlers (OAI-SearchBot, PerplexityBot, ClaudeBot)

llms.txt — Table of contents for LLM crawlers

JSON-LD schema markup — Product, FAQPage, HowTo

SSR audit — Flag client-side pages invisible to AI crawlers

MODULE 7

AI Rescore Loop

Prove it worked. Re-run the same AI queries after 30 days and measure the change.

The cycle

M1.5

Baseline 3/28 visible

→

M2-M4

Amazon Listing Opt

→

Off-Amazon Content

→

Wait

14-30 days

→

Rescore 14/28 visible

→

Proof

Client Report

Delta analysis

Visibility change (new mentions, lost mentions)

Position change (3rd to 1st mention)

Sentiment change

Citation source change (new blog/Reddit cited)

Competitor displacement

Hallucination resolution

Attribution

For each visibility gain, attribute to: Amazon listing changes (M4), Reddit content (M6), blog content (M6), technical fixes (M6), or organic/external. Attribution is probabilistic with confidence levels.

Complete Data Flow

What each module produces and who consumes it.

Module	Produces	Consumed By
M0/M1	product_profile	LQS Baseline, M1.5, M2, M2.1, M2.2, M3, M4
LQS Baseline	lqs_baseline_report (pre-optimization score)	Client report (proof of improvement delta)
M1.5	ai_visibility_baseline, ai_gap_report, citation_source_map	M2, M2.1, M3, M4, M6, M7
M2	competitor_list_raw, trimmed, final	M2.1, M2.2, M3, M4, LQS
M2.1	intent_themes, Customer Intent Package	M2.2, M2.3, M3, M4, LQS
M2.2	usp_raw	M2.3
M2.3	usp_test_set (approved USPs)	M3, M4, M5, LQS
M3	Keyword Package, USP Keyword Bundles	M4, M5, LQS
M4	Final Listing + QA Reports	LQS Gate, M5, M6
LQS Gate	lqs_report (score, grade, MYE eligibility)	M5 (must pass >= 70)
M5	experiment_results, learning_record	M4 (retest), M7
M6	content_package (Reddit, Quora, Blog, Video, etc.)	M7
M7	rescore_report, attribution, client_report	M1.5 (next cycle)

SLO - Smart Listing Optimizer | Business Logic Documentation

Generated April 2026 | FPAI Deliverable