AI visibility diagnostic for ecommerce brands.

AnswerRank AI

AnswerRank AI helps ecommerce brands understand how their products appear inside AI buying answers.

When a buyer asks: "What is the best magnesium glycinate supplement for sleep and muscle recovery in India?"

AnswerRank checks whether your product appears, which competitors outrank you, how visible your brand is, and what you should change in your listing.

Live Demo

https://answerrank-ai.vercel.app

Primary demo sample:

• Product: HK Vitals 100% Magnesium Glycinate
• Query: best magnesium glycinate supplement for sleep and muscle recovery in India
• Competitors:
  • Tata 1mg Magnesium Glycinate
  • HealthyHey Magnesium Glycinate
  • Himalayan Organics Magnesium
  • Carbamide Forte Magnesium
  • Wellbeing Nutrition Magnesium

Built For

Pixii.ai Founding Engineer project round.

Why This Exists

Ecommerce brands already track Google rankings, Amazon listings, paid acquisition, and social distribution. The buying surface is changing again: customers increasingly ask AI tools what to buy.

That creates a new visibility problem. A brand can have strong SEO and strong marketplace presence while still being absent from AI-generated buying answers. Or worse, AI may recommend a competitor with clearer copy, stronger trust signals, or better product framing.

AnswerRank AI is built to make that visible. It turns fuzzy LLM output into a structured diagnostic: whether a product appears, who outranks it, how confident the result is, and what listing changes would improve its chances of being surfaced next time.

What It Does

• analyzes AI buying answers for product visibility
• checks product and competitor mentions
• scores AI search visibility
• uses Gemini as a live answer engine
• uses Firecrawl for product-page context
• shows source metadata and provider coverage
• generates title, bullets, FAQ, and positioning fixes

Revenue-Ready MVP

AnswerRank now separates a free diagnostic from a deeper full audit path:

• Free diagnostic:
  • Gemini sample diagnostic
  • max 3 expanded buyer-intent queries
  • provider coverage transparency
  • premium CTA for full audit
• Full audit:
  • up to 20 buyer-intent queries
  • Gemini + OpenAI + Claude when configured
  • competitor share of voice
  • query-level visibility breakdown
  • lead capture for manual/premium follow-up

Core Workflow

diagnose visibility -> understand competitors -> fix the listing

Screenshots

Brand Assets

Custom SVG brand assets live here:

• public/brand/answerrank-logo.svg
• public/brand/answerrank-mark.svg
• public/brand/answerrank-wordmark.svg
• public/brand/favicon.svg

These are custom brand assets created for this project. They do not rely on third-party paid artwork or imported provider logos.

Architecture

flowchart TD
    A[User enters product, URL, query, competitors] --> B[/api/diagnose]
    B --> C[Validate input]
    C --> D[Firecrawl extracts product context]
    C --> E[Gemini answer engine]
    D --> E
    E --> F[Parse product and competitor mentions]
    F --> G[AEO scoring engine]
    G --> H[Visibility report]
    H --> I[/api/fix-it]
    I --> J[Rewrite title, bullets, FAQ, positioning]

Loading

Scoring Model

The AEO score is a structured visibility score, not a generic LLM confidence number.

At a high level it combines:

• mention frequency
• rank position
• sentiment/confidence
• competitor gap
• query relevance
• provider coverage adjustment

This lets the app distinguish between "the product was technically mentioned once" and "the product is consistently surfaced in a strong ranking position across the available answer engines."

Source Metadata

AnswerRank does not pretend full confidence when only one provider runs.

Every report surfaces:

• providers used
• providers skipped
• coverage count
• Firecrawl status
• mock/live mode
• cache status
• audit mode

That matters because a Gemini-only run should not be presented as if it covered the full AI answer surface. The app keeps the sampled score from the live result, then applies provider coverage adjustment when fewer than three planned answer engines are available.

Fix It Engine

After the visibility diagnostic, the Fix It Engine generates:

• rewritten product title
• listing bullets
• FAQ
• positioning statement

The goal is not just to explain why a product underperformed. It is to turn that diagnosis into listing copy a brand can actually use.

Streaming Diagnostic

AnswerRank keeps the existing JSON route for compatibility and adds a streaming route at /api/diagnose/stream.

Streaming emits staged progress events such as:

• validating input
• expanding buyer-intent queries
• extracting product-page context
• provider start / provider done
• scoring
• final result

The frontend prefers streaming and falls back to the JSON route if streaming fails.

Query Expansion

Query expansion is a core product differentiator. Brands do not care about one query; they care about the buyer-intent surface around a product category.

Current behavior:

• deterministic expansion is always available
• free mode includes the seed query plus up to 3 total queries
• full mode supports up to 20 queries
• optional LLM-driven expansion is gated by ANSWER_RANK_LLM_QUERY_EXPANSION=true

Example expansions from the HK Vitals seed query include:

• best magnesium glycinate supplement for sleep in India
• HK Vitals 100% Magnesium Glycinate vs HealthyHey Magnesium Glycinate
• magnesium glycinate supplement with high absorption India

Rate Limiting and Cost Guards

AnswerRank includes typed validation and request limits to keep model costs predictable:

• strict field-length limits for product context and queries
• max 8 competitors
• /api/diagnose free limit: 5 requests per IP per hour
• /api/diagnose/stream free limit: 5 requests per IP per hour
• /api/fix-it: 5 requests per IP per hour
• /api/leads: 10 requests per IP per hour

If Upstash Redis is configured, rate limiting uses Redis. Otherwise the app falls back to an in-memory development-safe limiter and logs that the fallback is not production-safe.

Report Caching

Successful reports are cached using a SHA-256 key built from:

• product name
• product URL
• buyer-intent query
• competitors
• audience
• region
• audit mode
• expanded query count
• provider policy

Cache behavior:

• free reports: 6 hours
• full reports: 24 hours
• mock reports: 24 hours

If Upstash Redis is configured, cache storage uses Redis. Otherwise the app uses an in-memory fallback for development.

Lead Capture

The report includes a premium CTA:

• Unlock full tri-engine audit
• collect email + optional company name
• submit to /api/leads
• respond with:
  • Full audit request received. We’ll send the next steps.

This is intentionally a lightweight revenue path for the MVP. It does not add billing or auth yet.

Tech Stack

• Next.js App Router
• TypeScript
• Tailwind CSS
• Gemini API
• Firecrawl API
• optional OpenAI / Anthropic adapters
• Vercel

Environment Variables

Use .env.example as the starting point:

NEXT_PUBLIC_DEMO_MODE=true
ANSWER_RANK_FULL_AUDIT_ENABLED=false
ANSWER_RANK_LLM_QUERY_EXPANSION=false
GEMINI_API_KEY=
FIRECRAWL_API_KEY=
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
UPSTASH_REDIS_REST_URL=
UPSTASH_REDIS_REST_TOKEN=

Rules:

• NEXT_PUBLIC_DEMO_MODE=true uses deterministic mock mode
• NEXT_PUBLIC_DEMO_MODE=false enables live provider mode
• ANSWER_RANK_FULL_AUDIT_ENABLED=false keeps the product in free diagnostic mode by default
• ANSWER_RANK_LLM_QUERY_EXPANSION=false keeps query expansion deterministic unless explicitly enabled
• UPSTASH_REDIS_REST_URL and UPSTASH_REDIS_REST_TOKEN enable production-safe rate limiting and caching
• API keys must stay server-side
• do not expose secrets as NEXT_PUBLIC variables

Running Locally

npm install
npm run dev

Then open http://localhost:3000.

Production / Vercel Notes

Changing NEXT_PUBLIC_DEMO_MODE requires a redeploy because it is bundled at build time.

Expected live setup:

NEXT_PUBLIC_DEMO_MODE=false
ANSWER_RANK_FULL_AUDIT_ENABLED=false
ANSWER_RANK_LLM_QUERY_EXPANSION=false
GEMINI_API_KEY=your_gemini_key
FIRECRAWL_API_KEY=your_firecrawl_key
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
UPSTASH_REDIS_REST_URL=
UPSTASH_REDIS_REST_TOKEN=

Notes:

• Gemini is the main live answer-engine provider
• Firecrawl extracts product-page context when a product URL is present
• OpenAI and Anthropic are optional adapters and can stay empty
• mock mode remains available for reproducible demos and reviewer testing

Live Demo Deployment

For live Gemini + Firecrawl mode on Vercel, set:

NEXT_PUBLIC_DEMO_MODE=false
ANSWER_RANK_FULL_AUDIT_ENABLED=false
ANSWER_RANK_LLM_QUERY_EXPANSION=false
GEMINI_API_KEY=your_gemini_key
FIRECRAWL_API_KEY=your_firecrawl_key
OPENAI_API_KEY=
ANTHROPIC_API_KEY=
UPSTASH_REDIS_REST_URL=
UPSTASH_REDIS_REST_TOKEN=

Then redeploy.

Expected source card:

• Source: Live partial
• Providers used: Gemini
• Providers skipped: OpenAI, Anthropic
• Coverage: 1/3
• Firecrawl: used or attempted

Only NEXT_PUBLIC_DEMO_MODE is public. API keys must remain server-side.

Revenue MVP Notes

This repository is intentionally positioned as a revenue-ready audit MVP:

• free mode stays cheap and strict
• full mode is feature-gated
• provider coverage stays honest
• lead capture exists before payments
• manual full-audit fulfillment is a valid early revenue path

Social Preview

Repository and deployment preview asset:

• public/social-preview.svg

This is a custom SVG social preview designed for GitHub and deployment metadata.

What I Would Build Next

• weekly AI visibility tracking
• query expansion for buyer-intent clusters
• OpenAI, Claude, and Perplexity provider runs
• historical visibility charts
• shareable PDF reports
• Shopify and Amazon listing integrations
• automated listing improvement experiments

Keywords

AI search optimization, answer engine optimization, AEO, ecommerce AI, AI buying answers, LLM SEO, product listing optimization, Amazon listing intelligence, Gemini API, Firecrawl, AI visibility tracking.