How it works
From a question to a measurable signal.
Here is exactly what happens when Citewise AI runs a check, and how every number in your dashboard is calculated.
The four steps
You define the project
Brand, alternative names, domain, competitors, the prompts to track and which engines to ask.
We ask each engine
Each prompt is sent, unchanged, to each engine's official API with web search enabled.
We analyse the answer
Mentions, order of brands, cited sources and your domain are extracted with deterministic rules.
You see the trend
Results roll up into visibility, share of voice and citation metrics, compared run over run.
Methodology
What is a check?
A check is one prompt answered by one engine. It costs one credit. When a project runs, Citewise AI performs one check for every combination of prompt and selected engine: 10 prompts on 4 engines is 40 checks. If a check fails (for example, the provider times out), it is recorded as failed and its credit is returned to your balance automatically.
Engines and models
We call each provider's official API with its web search or grounding feature switched on, so answers can reflect current information and include source links:
- ChatGPT: OpenAI API with web search enabled.
- Gemini: Google API with web search enabled.
- Perplexity: Perplexity AI API with web search enabled.
- Claude: Anthropic API with web search enabled.
Your prompt is sent exactly as you wrote it, as a single user message, with no system prompt and no hidden instructions. The model version that answered is stored with every check.
API answers vs. consumer apps. The ChatGPT, Gemini, Perplexity and Claude apps can personalise answers using a signed-in user's history, location and settings. API answers are not personalised, which makes them a consistent baseline for measurement, but they will not always match what an individual user sees.
How mentions are matched
Your brand counts as mentioned when the answer contains your brand name, one of your alternative names, or your domain, matched case-insensitively on word boundaries. “Acme” matches “Acme's pricing” but not “Acmeville”. Competitors are matched the same way. Choose distinctive names: a brand called “Apple” will also match the fruit.
A check counts as cited when one of the answer's source links points to your domain or one of its subdomains.
How metrics are calculated
| Metric | Definition |
|---|---|
| Visibility | Successful checks mentioning your brand ÷ all successful checks |
| Position | Your order among the tracked brands mentioned in an answer (1 = named first) |
| Average position | Mean position across checks where you were mentioned |
| Citation rate | Successful checks citing your domain ÷ all successful checks |
| Share of voice | Your mentions ÷ mentions of all tracked brands (you + competitors) |
| Top sources | Number of answers citing each domain at least once |
Why answers vary
Language models are not deterministic, and web results change daily. The same prompt can produce a different answer an hour later. That is why Citewise AI focuses on trends across many prompts and repeated runs rather than any single answer. As a rule of thumb, treat a change as real when it holds across several runs and several prompts.
What we don't do
- We don't guarantee that any engine will mention or recommend your brand.
- We don't attempt to manipulate AI systems, generate fake reviews or post on your behalf.
- We don't scrape consumer chat apps or use personal accounts. We use official APIs only.
- We don't share your prompts or results with other customers.
Put a number on your AI visibility.
Free workspace with 25 credits every month. No card required.