onpage.app

Methodology

How the scoring works

1. Query execution

You provide up to 5 queries that your customers actually ask. We send each query to Claude (Anthropic), Gemini (Google), and GPT-4o (OpenAI). These are real API calls, not cached or simulated. Each model generates a fresh response.

We recommend a mix of query types: category queries ("best [product category]"), comparison queries ("[your brand] vs [competitor]"), and problem queries ("how to [solve problem your product addresses]"). These are the queries where AI answers shape purchasing decisions.

2. Response parsing

Each LLM response is analyzed programmatically for four signals:

  • Mention presence: Is your brand explicitly named in the response?
  • Position: If mentioned, where? First recommendation carries more weight than fifth.
  • Sentiment: Is the mention positive, neutral, or negative? Context matters.
  • Competitor landscape: Which other brands appear? How often? In what context relative to yours?

3. Scoring

Each query gets a score from 0 to 100 per model, based on three components:

Mention present: +40 points
Position #1: +30 points | #2: +20 | #3: +10
Positive sentiment: +30 points | Neutral: +15 | Negative: +5

Your overall score is the average across all queries and models. We show per-model and per-query breakdowns so you can see exactly where you're strong and where you're invisible.

4. Recommendations

This is what makes us different. After parsing, we feed the complete audit data back into Claude with a specialized prompt that generates 5 specific, actionable recommendations. Each one includes:

  • Priority level (high, medium, low) based on visibility gap severity
  • Page type to create (comparison page, FAQ, technical guide, case study)
  • Target queries from your audit that this recommendation addresses
  • Competitor context explaining who shows up instead and what they're doing right

The goal: you finish reading the report and know exactly what content to create this week. Not "improve your content strategy" — specific pages, specific angles, specific gaps.

Transparency

Every audit shows the raw LLM responses. You can expand any query result and read exactly what Claude, Gemini, or GPT-4o said. No black boxes. If you disagree with the scoring, you can see why the system scored it the way it did and make your own judgment.