Show our work
How the checker works
No black box. These are the exact questions we ask, the OpenRouter models we fan out to, and how we turn their answers into your visibility score — the same definitions the checker runs on, published straight from the code.
prompt set: checker-en-v1
The models we ask
Every prompt is audited across each of these models via OpenRouter, so the score reflects how different AI assistants answer the same buyer question.
- GPT-4o miniopenai/gpt-4o-mini
- Claude Sonnet 4.5anthropic/claude-sonnet-4.5
- Gemini 2.5 Flashgoogle/gemini-2.5-flash
The 12 fixed prompts
The same 12 category questions run for every brand. The category term shown here (for example solutions, worldwide, or the market leaders) is replaced with your category, location, and a leading competitor when the check runs.
What are the best solutions available today?
recommendation
Which solutions offer the best value for the money?
recommendation
Who are the leading solutions brands worldwide?
makers
Which companies make the most popular solutions worldwide?
makers
How do the top solutions brands compare?
comparison
What sets the best solutions providers apart from the rest?
comparison
What are good alternatives to the market leaders for solutions?
alternatives
Besides the market leaders, which brands are worth considering for solutions?
alternatives
Which brands are known for the best solutions worldwide?
best-of
What is the most trusted name in solutions right now?
best-of
Which solutions would you recommend for everyday use, and why?
use-case
If someone is shopping for solutions, which brands should they look at first?
use-case
The score
The share of model answers that mentioned the brand: the count of answers with a footprint divided by the total answers collected (prompts × configured OpenRouter models).
score = footprints / total_responsesThe result is a fraction from 0.0 to 1.0, shown as a 0–100% headline. With 12 prompts × 3 models = 36 answers, 12 mentions is 33%.
What this score does — and does not — tell you
One sample per prompt, for now
Each of the 12 prompts is asked once per configured model today. A single answer can vary run to run, so treat the score as a directional signal, not a precise ranking. Repeat sampling is on the roadmap.
The score is binary today
Each answer is scored as a plain yes/no: did the brand get mentioned or not? A weighted 0–100 version — rewarding earlier, more prominent mentions — is coming. Today the headline percentage is simply the share of answers that named the brand.
We measure unprompted visibility
The 12 prompts ask about the category and never name your brand. We then search the answers for you. That is the whole point: we measure whether a model brings you up on its own, not whether it can talk about you when asked.
English only
The checker runs in English today. Turkish is not yet supported — "no Turkish beats bad Turkish" — so a non-English brand is still asked the English category questions.
Results are cached for 24 hours
A brand + category checked twice within 24 hours returns the same cached result, so the score is stable across a session and we keep LLM costs sane. A fresh run happens after the cache expires.