How We Test AI Visibility
Every audit runs live queries against five AI systems, captures every response verbatim, and scores the results algorithmically. Here's exactly what happens under the hood.
The audit pipeline, step by step
From query generation to final score — nothing is black-boxed.
Step 1
We generate buyer-intent queries
Based on the business category and location, we build the exact questions a real prospect would type — "best landscaper in Virginia Beach," "top-rated dentist near me," "who should I hire for roofing in Portland." These are discovery queries: broad, buying-intent questions that test whether AI recommends the business unprompted.
Step 2
We send those queries to five AI systems via API
Every query is sent to ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews — the five systems that cover roughly 95% of how customers discover businesses through AI today. We use the official API for each platform, with a consistent system prompt that asks for specific business recommendations.
Step 3
We capture every response verbatim
Nothing is paraphrased or summarized. The full text of every AI response is recorded and stored. You can read exactly what ChatGPT said, what Claude recommended, which businesses Gemini listed, and whether Google showed an AI Overview at all.
Step 4
We analyze mentions, position, and competitors
Our detection engine checks whether the business was mentioned by name, where it appeared in the response (top 3, top 5, or just mentioned), and which competitors showed up instead. We also run entity queries — asking each AI directly about the business — to measure how well the AI knows the brand.
Step 5
We crawl the website for AI-readiness signals
Beyond AI queries, we check the business website for structured data, FAQ schema, AI crawler permissions (robots.txt), content depth, author attribution, citable facts, and technical SEO foundations. These signals influence how AI systems extract and present information.
Step 6
We score and report
All data feeds into a weighted scoring model across four Tier 1 modules (AI Discovery, Entity Recognition, Competitive Position, Content Readiness) and three Tier 2 modules (Local Relevance, Trust & Authority, Technical Foundations). The result is a single AI Visibility Score from 0 to 100, delivered in a white-label PDF report.
Why results may differ from manual searches
This applies to every AI audit tool on the market, not just RediForAI. Understanding these factors is essential to interpreting any AI visibility data.
API vs. consumer chat are different experiences
When you open ChatGPT.com in your browser, you get web browsing, personalization, location awareness, and your conversation history influencing results. Our audit uses the official API — no browsing, no personalization, no location context. The API tests the AI's core knowledge, not the enhanced consumer experience. This is true for all AI audit tools, not just RediForAI.
AI responses are non-deterministic
Ask ChatGPT the same question twice, an hour apart, and you may get different businesses listed. AI systems use controlled randomness (temperature) to generate varied, natural responses. Every audit is a snapshot in time — a data point, not a permanent verdict. Manual searches are also snapshots; they just happen to be different snapshots.
Google AI Overviews don't appear for every query
Google selectively shows AI Overview blocks. Commercial and local queries like "best plumber in Vancouver WA" frequently return no AI Overview at all — Google shows regular search results instead. When no overview exists, we accurately report that. This is not a bug; it's how Google works.
Web-grounded vs. training-data providers
Perplexity and Google AI Overviews pull from live web results — they reflect current search rankings. ChatGPT, Claude, and Gemini respond from their training data, which has a knowledge cutoff. A business that ranks well in Google may not yet exist in ChatGPT's training data. Our audit tests both angles to give a complete picture.
The five AI systems we test
Each provider has different strengths, data sources, and behaviors.
| Provider | Data Source | Web-Grounded? | Best At |
|---|---|---|---|
| ChatGPT | Training data | No | Broad knowledge, entity recognition |
| Claude | Training data | No | Detailed analysis, nuanced recommendations |
| Gemini | Training data | No | Google ecosystem knowledge |
| Perplexity | Live web search | Yes | Real-time web presence, current rankings |
| Google AI Overviews | Google Search index | Yes | Search visibility, AI-enhanced results |
Common questions about our testing
If you're comparing your manual searches to our report, start here.