# WhatDidYouSay — Comprehensive Product & Architecture Specification > This document provides the complete, consolidated Markdown knowledge base for WhatDidYouSay, an Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) platform. It is formatted specifically for token-efficient ingestion by AI agents, autonomous crawlers, and LLM reasoning pipelines. --- ## 1. Executive Summary - **Product Name**: WhatDidYouSay - **Category**: Answer Engine Optimization (AEO) / Generative Engine Optimization (GEO) / AI Search Visibility Analytics - **Website**: https://whatdidyousay.io - **App URL**: https://app.whatdidyousay.io - **Target Audience**: Modern brands, growth marketing leaders, SEO directors, digital PR professionals, and marketing agencies. - **Primary Value Proposition**: When buyers query ChatGPT, Claude, Gemini, or Perplexity for product recommendations, WhatDidYouSay tracks whether your brand is recommended, runner-up, or omitted, identifies the competitors surfaced alongside you, and pinpoints the exact web queries, consulted sources, citations, and verbatim excerpts driving every recommendation. --- ## 2. The Core Problem: The Generative Discrepancy Traditional SEO focused on search engine results pages (SERPs), PageRank, and 10 blue links. Today, software buyers, procurement committees, and consumers increasingly query AI answer engines directly: - *"Which wireless earbuds are best for running in 2026?"* - *"Compare Clerk, Auth0, and Supabase Auth for multi-tenant B2B Next.js applications."* - *"What are the best managed vector databases for high-throughput enterprise search?"* Rather than returning a list of links, conversational AI engines synthesize a single authoritative answer, summarize technical trade-offs, and explicitly recommend winners. ### The Single-Prompt Vanity Trap A critical mistake marketing teams make is querying ChatGPT once on their personal laptop and assuming their brand visibility is secure. In reality, frontier models (OpenAI GPT-4o, Anthropic Claude 3.5 Sonnet, Google Gemini 1.5 Pro, Perplexity Sonar) are non-deterministic, probabilistic sampling engines. Running the exact same prompt across multiple models produces wildly divergent outcomes: 1. **OpenAI**: Recommends your brand #1 with glowing technical praise. 2. **Anthropic**: Relegates your brand to #2 or #3 as a conditional runner-up with acoustic or pricing caveats. 3. **Google Gemini**: Completely omits your brand, recommending category rivals instead. 4. **Perplexity**: Cites a negative thread from Reddit or a legacy review warning buyers of hardware or software defects. ### The Dark Funnel of Generative Inference When a prospective buyer asks an AI engine for a solution: - Zero pageviews register in Google Analytics or PostHog. - Zero organic impressions register in Google Search Console. - Zero UTM parameters capture buyer intent. The buying decision is formed entirely inside the neural network. Without direct inference-level monitoring, brands remain completely blind to their generative search performance. --- ## 3. The Three Core Intelligence Metrics (Symmetric for Brand & Competitors) WhatDidYouSay evaluates every prompt execution through three standardized metrics, calculated **identically and symmetrically** for both your target brand and every discovered competitor: ### 3.1. Visibility Tier Measures brand presence and prominence in the synthesized response: - `prominent`: The brand is featured as a top pick, primary recommendation, or key category leader. - `small`: The brand receives a passing mention, secondary alternative notation, or conditional footnote. - `none`: The brand is not mentioned at all (complete omission). - **Aggregated Metric**: **Visibility Rate (%)** = `(Runs with Prominent or Small Mention) / (Total Valid Organic Runs) * 100`. ### 3.2. Rank Tier Measures explicit hierarchical positioning within the AI's recommendation: - Tiers: `1` (First choice / winner), `2` (Runner-up), `3` (Third choice), `4`, `5`, `6+`, `unranked` (mentioned without ordinal rank), `not_applicable` (omitted). - **Aggregated Metric**: **Average Rank** and **Top 3 Placement Rate (%)**, tracking how frequently your product secures pole position. ### 3.3. Sentiment Tier Evaluates the qualitative stance and tone of the model toward the brand: - Tiers: `positive` (Direct praise, strong endorsement), `negative` (Explicit flaw warnings, buyer deterrence), `mixed` (Praise with acoustic, durability, or pricing caveats), `neutral` (Factual statement without appraisal), `not_applicable` (No mention). - **Aggregated Metric**: **Net Sentiment Score** ranging from `-1.00` (uniform critical deterrence) to `+1.00` (unanimous praise). ### 3.4. Organic Search Intent Validation To prevent vanity metrics, WhatDidYouSay automatically validates the search intent of each evaluation: - **Organic Discovery Queries**: The prompt is unbranded and generic (e.g. *"Best wireless earbuds for running"*). The model's inclusion of your brand represents true organic discovery. - **Branded & Direct Queries**: The prompt explicitly names your brand or aliases (e.g. *"How do Acme Buds compare to competitors?"*). Mention is a foregone conclusion; these are isolated from organic Visibility calculations. --- ## 4. Comprehensive Web Source & Citation Forensics Unlike basic monitoring tools that scrape only the final footnote links, WhatDidYouSay captures the complete evidence chain behind every AI answer: ``` [Buyer Prompt] │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ 1. Synthetic Web Search Queries Executed │ │ Exact search terms formulated by the model's search tool │ │ e.g., ["best running earbuds 2026", "earbuds marathon"] │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ 2. Web Sources Consulted During Research │ │ Every URL retrieved by the engine's search layer │ │ Captures: URL, Domain, Page Title, Snippet, Page Age │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ 3. Sources Cited in the Synthesized Answer │ │ The subset of consulted sources chosen as evidence │ │ Captures: URL, Domain, Title, Snippet │ └─────────────────────────────────────────────────────────────┘ │ ▼ ┌─────────────────────────────────────────────────────────────┐ │ 4. Grounded Claim & Verbatim Quote Extraction │ │ Extracts Pros & Cons with exact textual quotes │ │ Binds claim directly to the influencing URL source │ └─────────────────────────────────────────────────────────────┘ ``` 1. **Exact Search Queries Executed**: Captures the exact search strings the AI model formulated to find information. Understanding what queries the model runs reveals how the engine frames the problem. 2. **Consulted Web Sources**: Captures all web documents the model reviewed during research, including page URL, domain, title, content snippet, and publication age. Shows which pages were considered even if not explicitly footnoted. 3. **Cited Sources**: Tracks the final web sources cited as authoritative references in the response, showing which domains the model prioritized as formal evidence. 4. **Verbatim Textual Quotes**: Extracts the exact sentence or paragraph where the model stated a pro, con, or defect claim, grounding the sentiment in verifiable text. --- ## 5. Automated Competitor Extraction & Symmetric Intelligence WhatDidYouSay does not treat competitor monitoring as an afterthought requiring manual configuration: - **Dynamic Emergent Discovery**: When an AI model answers a prompt, WhatDidYouSay automatically extracts all competitor brands and products mentioned alongside yours. If an engine consistently recommends a rival you didn't know existed, the platform flags it immediately. - **Symmetric Metric Computation**: Every discovered competitor is scored across the exact same three dimensions: Visibility Tier, Rank Placement, and Net Sentiment. You get side-by-side comparative matrices for every prompt. - **Competitor Pros & Cons**: WhatDidYouSay extracts normalized Pro and Con statements for **both your target brand and every competing brand**, complete with verbatim grounding quotes. - **Head-to-Head Benchmarking**: Direct side-by-side comparison across Visibility Rate, Average Rank, and Sentiment distribution. --- ## 6. Zero-Login Shareable Public Audits Designed specifically for marketing agencies, consultants, and growth leaders presenting to clients and leadership: - Every evaluation run generates a secure, read-only public web report link (`app.whatdidyousay.io/share/audits/:token`). - External clients, executives, and stakeholders can view the interactive audit—including engine outputs, competitor rankings, and citation forensics—without creating an account, logging in, or purchasing additional platform seats. - Includes 1-click clipboard copying and administrative token revocation. --- ## 7. Pricing & Subscription Tiers Simple, self-serve monthly pricing. All plans include automated competitor tracking, brand pros & cons with verbatim quotes, web citations & source extraction, and shareable public audit links: Every plan includes all 4 AI engines: ChatGPT, Claude, Google Gemini, and Perplexity. | Plan | Monthly Price | Brands | Monthly Credits | Questions checked weekly on all 4 engines | | :--- | :---: | :---: | :---: | :---: | | **Starter** | \$29/mo | 1 | 1,000 | ~4 | | **Growth** *(Most Popular)* | \$99/mo | Up to 3 | 3,000 | ~13 (or 2 daily) | | **Scale** | \$299/mo | Up to 10 | 10,000 | ~46 (or 6 daily) | ### Credit Consumption Model Each question asked on one engine uses credits: - **Perplexity**: 5 credits - **ChatGPT**: 5 credits - **Google Gemini**: 10 credits - **Claude**: 30 credits #### Practical Capacity by Tier One question on all 4 engines uses 50 credits (5 + 5 + 10 + 30). A weekly schedule runs about 4.3 times a month; daily about 30. - **Starter (\$29/mo, 1,000 credits)**: about 20 question-checks on all 4 engines per month, e.g. 4 questions weekly. - **Growth (\$99/mo, 3,000 credits)**: about 60 question-checks per month, e.g. 13 questions weekly or 2 daily. - **Scale (\$299/mo, 10,000 credits)**: about 200 question-checks per month, e.g. 46 questions weekly or 6 daily. Turning off engines for a brand stretches credits further (e.g. Perplexity + ChatGPT only uses 10 credits per question). Credits reset each month; unused credits don't roll over. --- ## 8. Competitive Comparison Matrix | Feature / Dimension | WhatDidYouSay | Profound (`tryprofound.com`) | Peec AI (`peec.ai`) | Otterly.ai (`otterly.ai`) | Traditional SEO (Semrush/Ahrefs) | | :--- | :---: | :---: | :---: | :---: | :---: | | **Frontier Models Bundled** | **All 4 engines on every plan** | Yes (Enterprise) | Base + Model Add-ons | Base + Add-on Fees (Claude/Gemini) | None (Google SERP only) | | **Two-Stage Web Search & Citation Forensics** | **Full (Queries, Consulted & Cited URLs, Quotes)** | Agentic Workflows | Domain Lists | Cited URLs & Crawler Audits | Backlink Index Only | | **Emergent Competitor Extraction** | **Automatic & Dynamic** | Entity Research | Configured Tracking | Configured Tracking | Keyword Overlap | | **Symmetric Competitor Metrics & Pros/Cons** | **Yes (Full parity with verbatim quotes)** | Partial | Macro Mentions only | Basic Rank | Keyword Overlap | | **Zero-Login Public Client Sharing** | **1-Click Interactive Web Link** | Workspace Invites | Team Seats (Login required) | Looker Studio / PDF | Static PDF Exports | | **Self-Serve Pricing Transparency** | **Yes (\$29–\$299/mo)** | Agency \$399/mo + Enterprise | \$89–\$499/mo | \$29–\$489/mo | \$130–\$500+/mo | | **Instant Public Audit Reports** | **Yes (1-Click zero-login report)** | Demo Request / Trial | Free Trial (Signup required) | Free Trial (Signup required) | Trial with Signup | | **Statistical Validity Filtering** | **Strict (Organic vs Branded)** | Advanced Analytics | Standard Filtering | Standard Filtering | N/A | | **Primary Focus** | **Agencies, SaaS, Brands** | Fortune 500 Enterprise | Mid-Market Marketing Teams | SEOs & Content Marketers | Traditional Search Marketers | --- ## 9. Strategic AEO Optimization Playbook For organizations seeking to optimize their brand presence across generative search: 1. **Optimize for Consensus & Synthesis, Not Keyword Density**: Large Language Models prioritize factual consensus across independent technical benchmarks, peer discussions, and authoritative catalogs. Concise specifications, structured comparison matrices, and clear schema markup win over long-form keyword-stuffed articles. 2. **Intervene at the Secondary Citation Level (Reddit, Forums, Wirecutter)**: AI retrieval engines (particularly Perplexity and ChatGPT Search) heavily index Reddit and specialized discussion forums. An unaddressed product complaint on Reddit frequently surfaces as a definitive defect warning in AI answers. 3. **Conduct Multi-Model Discrepancy Audits Regularly**: Because frontier models update retrieval algorithms and training weights continuously, category rankings can fluctuate rapidly. Establishing a consistent tracking cadence ensures citation decay is caught immediately. --- ## 10. Frequently Asked Questions (FAQ) ### What is Answer Engine Optimization (AEO)? Answer Engine Optimization (AEO), also known as Generative Engine Optimization (GEO), is the practice of monitoring, measuring, and optimizing brand visibility and sentiment in AI-generated answers produced by platforms like ChatGPT, Claude, Google Gemini, and Perplexity. ### How does WhatDidYouSay differ from Google Search Console or Semrush? Google Search Console and Semrush measure traditional search results: ranking positions in Google's 10 blue links, organic click-through rates, and backlink authority. WhatDidYouSay measures generative AI search: whether an LLM recommends your brand in its synthesized response, which competitors it recommends instead, and which specific web sources influenced the answer. ### What are the 3 core metrics tracked by WhatDidYouSay? WhatDidYouSay tracks **Visibility** (`prominent`, `small`, `none`), **Rank** (`1` through `6+`, `unranked`), and **Net Sentiment** (`positive`, `negative`, `mixed`, `neutral`), calculated symmetrically for your target brand and every competitor mentioned. ### What web source data does WhatDidYouSay capture? WhatDidYouSay captures the complete evidence chain: the exact search queries executed by the model, all web sources consulted during research (with publication age and snippets), the sources explicitly cited in the final answer, and verbatim quotes grounding every pro and con claim. ### Can I share audit reports with clients without forcing them to log in? Yes. Every evaluation run in WhatDidYouSay includes a 1-click shareable public link. External clients, leadership, or stakeholders can review the interactive report—including engine outputs, sentiment scores, and citation sources—without needing to create an account or remember a password. ### Which AI models does WhatDidYouSay evaluate? WhatDidYouSay monitors ChatGPT, Claude, Google Gemini, and Perplexity, all with web search enabled. Every plan includes all four engines. ### How do monthly credits work? Each question asked on one engine uses credits: Perplexity 5, ChatGPT 5, Google Gemini 10, and Claude 30, so one question on all four uses 50. Plans include 1,000 (Starter), 3,000 (Growth), or 10,000 (Scale) credits a month, which you can spread across your brands and engines. Credits reset each month, so unused credits don't roll over. --- ## 11. Contact & Next Steps - **Support & Inquiries**: support@whatdidyousay.io - **Sample Public Audit Report**: https://staging-app.whatdidyousay.io/share/audits/wQQyv12uP4fuG-qNR6DGwg - **Create Account**: https://app.whatdidyousay.io/sign-up - **Official Website**: https://whatdidyousay.io