Skip to content

How it works

How we measure

remake asks AI models the prompts your customers would type, stores every answer word for word, and counts who each answer names. Every number on your screens comes from those stored answers.

How a run becomes your numbers: your prompts go to up to 12 AI models; every answer is stored word for word; Visibility, Share of voice, Position, Tone and Sources are counted from the stored answers, and Site score from your website.YOU WRITEYour promptsAs customers type themUP TO 12 AI MODELSChatGPTGeminiAI OverviewsClaude and moreEVERY ANSWERStored answersWord for word, with linksTHE SIX NUMBERSVisibilityShare of voicePositionToneSourcesYour websiteSite score

What is measured

Number Read from
How often answers name you: Visibility The text of each answer
Your share of all brand mentions: Share of voice The text of each answer
Where you are named in the list: Position The order brands appear in each answer
How answers speak of you: Tone A fixed list of praise and criticism words
Which sites answers link to: Sources The links each answer returns
How readable your site is to AI: Site score Your site's raw HTML

Visibility, Share of voice, Position, Tone and Sources are counts over stored answers. No AI model judges them. An AI read picks out which names across all the answers are businesses: your rivals, up to 25 per run.

What each number means

  • Position is your average place in the answers that name you. AI answers have no fixed top ten, so read it as a trend. A brand needs 3 or more answers that place it before Insights ranks it by place.
  • Each number reads AI answers, not your traffic. The AI referrals card on My website counts the visits AI answers send you.
  • A location changes a prompt's words. Every AI model is asked from the same place, whatever the prompt's location or country. See Countries and languages.
  • Rivals come from the answers. A rival you add by hand joins the boards once a run picks it out of the answers.

Answers vary

The same prompt can get a different answer each time it is asked. One answer is one sample.

More prompts and more runs give a steadier picture. Read a change over several runs before you act on it.

The app shows how much sits behind each number:

  • Each tile shows its count, such as "10 of 40 answers".
  • Insights shows the number of answers in view. Under 10, the line under the headline adds Few answers — read with care.
  • An AI model, topic or other slice with fewer than 3 answers shows a dash, not a number.
  • 1 run beside a Visibility figure means it rests on one run of your prompts. Hover it for the reason.

Asked where customers ask

Every prompt is asked live, on the day of the run, the way a customer would ask it. Each AI model is reached one of 3 ways, all with live search.

How each AI model is reached, and what follows from it
  • Read from the real product: ChatGPT, Gemini, Google AI Overviews and Google AI Mode. You see the answer a customer sees on screen, from a fresh chat with no history.
  • Asked with web search on: Claude, Grok, Kimi, DeepSeek, Mistral, Qwen and Meta AI. Each reads today's web pages for the prompt, then answers.
  • Asked with its own search: Perplexity.

No AI model is asked without live search. A model that answers from memory alone misses businesses that a customer's own search would find.

The models asked with web search on read the same pages for a given prompt, so their sources often overlap. Their answers can differ from the chat apps, which pick their own searches.

Meta AI is Meta's own AI model, asked with web search on. It is not the meta.ai app. That app answers only people signed in to a Meta account, so no one can read it for you.

Copilot is not tracked for now. Its earlier answers stay in your history.

ChatGPT and Gemini have been read from the real product since 8 October 2026. Before that they were asked with web search on, so a trend line for either can step at that date.

An answer read from ChatGPT or Gemini sometimes comes back cut off. It is asked again once. If it is still cut off, it is left out of the numbers and you are not charged for it.

Each prompt is asked once per AI model per run, the way a customer types it. The answer is stored as it comes back, with its links. Open any prompt on the Prompts page to read the answers behind its numbers.

Countries and languages

Every prompt belongs to your brand's country, and each brand has one. The AI writes its prompts in English for now, whatever the country. The country still sets the place words, such as "in Switzerland".

Your own prompts, and where the models are asked from

Prompts you add yourself can be in any language.

ChatGPT, Gemini, Google AI Overviews and Google AI Mode are asked from the United States. The other AI models search the web with no location set.

Counting sources

A page counts once per answer, however many times that answer links it. Links that differ only by www, a trailing slash, a tracking code or a jump to a quoted passage count as one page.

Your stored answers keep every link exactly as it came back. See Sources.

How your website is read

Each run reads up to 8 pages of your site as raw HTML, with no scripts run. That is what AI crawlers read. A fact that appears only after scripts run is missing for them, and the checks report it.

The website checks and the fix re-check read the site the same way. No AI model is involved in either.

How fresh the numbers are

Every tile reads your latest run. The header on each screen says when that run finished.

How often runs happen is up to you: from every hour to every 2 weeks. See Schedules.

Changes in Brand settings, such as a new alias, count from the next run. Stored runs keep the answers and verdicts they had.

Failures on our side are free

You pay for each answer as it comes back. An answer that fails on our side is refunded and left out of every number. See Runs.

FAQ

Why don't my numbers match what ChatGPT told me?

Your chat was one answer, at one moment. A run asks many prompts across several AI models. Read the trend over several runs, not one answer.

Does an AI decide my Tone?

No. Tone counts praise and criticism words around your name, from a fixed list. See Tone.

Why did my Visibility change when I changed nothing?

AI answers change as the models and the web change. Small moves between two runs are normal. A trend over several runs is the signal.

Open Insights