The method

Every Refinea number, from prompt to calculation

Prompts come from your data, cleaned and grouped by meaning. We read the answers every day from the AI engines' interfaces. Here you can see every step with its parameters, and how the data becomes AI Visibility Index, Share of Voice, Average Mention Position and Sentiment.

Want to see how AI engines see you today first? Get the free analysis.

  • ChatGPT
  • Gemini
  • Claude
  • Google AI Mode
  • AI Overview
  • Perplexity
  • Copilot
  • DeepSeek
  • Grok
  • Mistral
  1. Search ConsoleRunning shoes55,200 impressions
  2. PromptWhat are the best shoes for running a marathon?
  3. Engines
  4. AnswersNike named in 3 of 4 engines
  5. AI Visibility Index58.3%+5.4 pts
Refined Marketers
  • Caffeina
  • AIRC
  • Finbooks
  • WeHunt
  • GECOS
  • Grenke
  • Garmont Tactical
  • StoryWalking
  • Omni
  • Seoleader
  • Fabio Antichi
  • Axyon AI
  • Unflat
Named in thesurvey of the tools Italian SEO experts use (2026)
Pre-accelerated byB4i·Participant in TEF Ignition byTech Europe Foundation·Finalist atStartCup Lombardia
How it works

From a Google search to the number in your dashboard, in six steps.

Every number you see in Refinea goes through these steps, every day.

  1. We collect and clean your site's data.

    You connect Search Console, Bing Webmaster Tools and Google Analytics. We read the last 90 days of searches and merge Google and Bing queries, keeping the source of each one. We clean the data and group the queries by meaning, keeping only the clusters that stay stable.

    Nike's topic breakdown: active prompts, visibility, Share of Voice, impressions and clicks from Search Console
  2. Searches become prompts.

    Every cluster of searches, every product and every segment of your market becomes a prompt, written the way a person would ask an AI assistant. Then we rewrite the same prompts from the point of view of each buyer persona and run them through the validator. Each one is compared against a public archive of one million real conversations with AI assistants and checked against writing rules, so it reads the way a real person writes. We choose them together. You see every one and can change them anytime.

    Nike's buyer personas: amateur soccer player, parent, marathon runner, beginner runner, sneakerhead, trail runner
  3. The same prompts, every day, to every engine.

    We run the prompts on ChatGPT, Gemini, Claude, Google AI Mode and the other AI engines you choose, in the countries and languages of your market. The prompts stay the same every day, so you can compare one day with the next.

    Daily readings · Sep 22–28
    Mon 22Tue 23Wed 24Thu 25Fri 26Sat 27Sun 28
    ChatGPT
    Gemini
    Claude
    Google AI Mode
    AI Overview
    Perplexity
  4. We read the answers the way your customer reads them.

    Refinea reads each answer from the engine's interface, as the person asking sees it, without going through the APIs. In each answer it recognizes the brands named using the dictionary of your brand and your competitors, names and variants included. For each answer it records the order of the brands, the tone it takes about you and the sources it cites.

    Question · Running shoes

    What are the best shoes for running a marathon?

    • ChatGPTNike named4 sources
    • GeminiNike named7 sources
    • PerplexityNike named10 sources
    • AI OverviewNike missing3 sources
  5. From answers to numbers, with the formulas in plain sight.

    We count the answers to all active prompts, engine by engine, day by day. The formulas are below, one per metric, each with an example.

    Share of Voice for Nike and its competitors over time
  6. A cross-check that doesn't depend on the prompts.

    Your site's logs tell you which AI bots read your pages and when. We verify every bot visit with the providers' own tools: reverse DNS for Google, Bing, Apple and Amazon, and the published IP address lists for OpenAI, Perplexity and DuckDuckGo. With Google Analytics you see how many visits come from AI assistants.

    AI bot visits to a site over time, split into citations, training and indexing

Explore AI Monitoring Explore AI Crawlers

The formulas

Six metrics, six calculations.

The same definitions you find in the app, next to every number.

AI Visibility Index

How many answers you appear in.

answers that name you ÷ answers analyzed × 100

Nike appears in 40 of 100 answers. AI Visibility Index: 40%.

Share of Voice

How much space you get compared with your competitors.

answers you appear in ÷ (answers you appear in + answers each competitor appears in) × 100

Nike appears in 50 answers, all competitors together in 450. Share of Voice: 50 out of 500, or 10%.

Average Mention Position

How early you get named.

average of your position among the brands named, only in the answers that name you. 1 means first.

1.8 means Nike is usually the first or second brand in the answer.

Sentiment

How they talk about you.

positive mentions ÷ mentions with an assigned tone × 100, with the themes of each mention (quality, price, design, reputation and others)

14 positive mentions out of 22 in the last 7 days. Positive sentiment: 63.6%.

Citation Share

What share of the cited sources points to you.

for each engine, citations of your page ÷ all citations from that engine; then the average across engines

5 citations out of 50 on one engine and 25 out of 100 on another. Citation Share: (10% + 25%) ÷ 2 = 17.5%.

Citation Coverage

How many answers use your site as a source.

answers that cite your site ÷ all answers × 100

216 of 1,286 answers cite nike.com. Citation Coverage: 16.8%.

AI Visibility Index and Share of Voice answer two different questions. A brand can appear in half the answers and still hold a small share when there are many competitors.

The rules

Every engine counts the same.

Engines write their answers in different ways. These rules apply to all of them and make the numbers comparable.

One page, one citation per answer

ChatGPT repeats the same source at every reference in the text; Claude doesn't. We count each page once per answer, with the same rule for every engine.

The same address is the same page

nike.com/running, www.nike.com/running/ and the version with ?utm_source=openai are a single page. A YouTube video counts once, whatever format the link comes in.

Each engine gets one vote

In source shares we average across engines, so an engine that cites long lists of sources doesn't weigh more than the others.

Tone prompts stay separate

Prompts like “What do you think of Nike?” feed Sentiment and are left out of AI Visibility Index and Share of Voice.

  • https://www.nike.com/running/
  • nike.com/running?utm_source=openai
  • https://nike.com/running#top
nike.com/running1
“Refinea is a strategic partner that will let us expand our firepower and the services we can offer our clients, thanks in part to AI Agents.”
Alessandro ZainaHead of SEO at CaffeinaCaffeina

Put agents to work on everything AI reads about you.

With Pro, we set up Refinea with you on day one. Agents prepare the work, and your team approves it before anything goes out.

Want to see how AI engines see you today first? Get the free analysis.

Frequently asked questions about the method

Do you use the AI engines' APIs?

No. We read the answers from the engines' interfaces, as the person asking sees them.

Where do the prompts come from?

From searches in Search Console and Bing Webmaster Tools, from Analytics data, from your products and from your market segments, plus the same prompts asked by each buyer persona. We choose them together and you can change them anytime.

How do you check that the prompts are realistic?

Every prompt goes through the validator: we compare it against a public archive of one million real conversations with AI assistants and check it against writing rules, so it reads the way a person would write it. Then we choose the prompts together with you.

How often do the numbers update?

Every day. The first analysis arrives within 30 minutes of setup.

How do you recognize my brand in an answer?

With a dictionary of your brand and your competitors, names and variants included.

Do paused prompts count?

No. Only the answers to active prompts are counted.

Which days is the period calculated on?

On full calendar days. The comparison is always with the previous period of the same length.