Methodology
The RecoScope Framework
How we collect, normalize, and score AI recommendation data across models and categories.
The Three-Tier Model Classification
Independent AI
Claude (Anthropic)
No advertising revenue from recommendations. No integrated shopping or checkout features. Recommendations are based on training data and general knowledge without commercial incentives.
Search-Grounded AI
Perplexity
Retrieval-augmented model that searches the live web before answering. Recommendations reflect current web consensus rather than static training data. May surface brands with strong web presence that other models miss.
Commerce-Influenced AI
ChatGPT (OpenAI), Gemini (Google)
Models with active or announced commercial integrations. OpenAI has publicly introduced advertising into ChatGPT. Gemini is integrated with Google Shopping and Shopify agentic commerce. Recommendations from these models may be influenced by commercial relationships.
This classification is not a judgment of quality. All four models produce useful recommendations. The classification helps readers interpret differences in their outputs. When a commerce-influenced model consistently recommends different brands than an independent model, that divergence is worth examining.
Why This Matters
When consumers ask AI for product recommendations, they get answers shaped by each model’s training data, retrieval methods, and commercial integrations. Different models recommend different brands for the same question.
RecoScope exists to make those differences visible and measurable. We run standardized benchmarks across AI models so brands, agencies, and analysts can see exactly who AI recommends, where models agree, and where commercial influence may be shifting results.
How We Collect Data
Prompt Design
Each category gets three standardized prompts: an open-ended recommendation question, a constrained question with a specific use case or budget, and a brand comparison and ranking question. Prompts are identical across all models to ensure comparable outputs.
Model Querying
We run each prompt through ChatGPT, Claude, Gemini, and Perplexity during the same time window to minimize temporal variation. Evergreen categories are benchmarked monthly. Seasonal categories are benchmarked weekly during active periods.
Response Parsing
Every response is parsed for brand mentions. Each mention is recorded with its rank position (order of appearance), the agent that produced it, and which prompt triggered it. Brand names are normalized to handle variations in capitalization and formatting.
Scoring and Aggregation
Brands are scored by total mention frequency across all agents and prompts. We track first-mention rate (how often a brand appears first), top-3 rate (how often it appears in the top 3), and cross-model consensus (how many models independently recommend the same brand).
What We Track
| Metric | What It Measures |
|---|---|
| Total Mentions | How many times a brand appears across all agents and prompts |
| First Mention Rate | How often a brand is the first recommendation given |
| Top-3 Rate | How often a brand appears in the top 3 recommendations |
| Cross-Model Consensus | How many different AI models independently recommend the brand |
| Agent Classification Split | Whether independent and commerce-influenced models agree or diverge |
Limitations and Transparency
AI model outputs are non-deterministic. The same prompt can produce different results on different days. Our benchmarks capture a snapshot, not a guaranteed prediction.
We do not have access to the internal ranking algorithms of any AI model. Our three-tier classification is based on publicly available information about each company’s commercial integrations. We update classifications as new information becomes available.
RecoScope does not accept payment from brands to influence their ranking or visibility in our reports.
Report Cadence
Evergreen Categories
Benchmarked monthly. Categories with year-round consumer demand like office chairs, running shoes, and wireless earbuds. Reports track long-term trends in AI recommendation patterns.
Seasonal Categories
Benchmarked weekly during active periods. Categories with time-sensitive demand like lawn fertilizer, sunscreen, and space heaters. Reports track how AI recommendations shift through a season.
Want to see how your brand performs?
Get a free AI Visibility Audit. See exactly how AI models rank your brand, what they recommend instead, and where you can improve.
Request Your Free Audit