Guide 1 of 6

Your first week with Promptrack

What happens from the moment you add your brand to your first action, when each thing runs, and what you should not conclude yet.

Read time 6 min · Product behavior verified on 2 Sep 2026

Day one: your setup

You give one thing: your domain. Everything below is derived from it, and everything is editable before anything runs.

We read your site

Menu, footer, structured data and a few priority pages, to work out what you sell and to whom. This is also how your market and language are pre-filled: a .de domain sets Germany, otherwise we read the language declared on your homepage.

We suggest who you compete with

Not the brands you would name from memory: the domains that rank alongside you on Google for your own topics, filtered and named automatically. It is a starting list, and you decide what stays.

We write your buyer questions

The purchase questions your buyers put to an assistant, in their language. Edit the list, drop what does not fit. This list is the single input that decides everything else, which is why it has its own guide.

Day one: your first run

Your questions go out and the answers come back one by one. Three engines are on by default because they cover most of the buying journey; the other three are one click away in your settings.

How each engine is queried

This matters when you compare our numbers with what you see yourself, so here is exactly what happens.

EngineDefaultWhat we query
Google AI OverviewsOnThe real Google result for that query, in your market
Google AI ModeOffThe real Google result for that query, in your market
ChatGPTOnGPT-4o, OpenAI's model, through its API
PerplexityOnSonar, Perplexity's own model, with web search bound to your market
ClaudeOffClaude Sonnet, Anthropic's model, through its API
CopilotOffGPT-4o, OpenAI's model, standing in for Copilot
!
Read the four model-based engines as a close proxy, not as a recording of the consumer app. We name the model we query so you can judge for yourself. A consumer chat window can route to a different model version, and it adds its own memory and personalisation, so a single answer you get yourself can differ. What stays comparable is the same question, put to the same model, week after week.

Every question is asked from your buyers' market

A question about running shoes does not get the same answer in Germany and in Australia, so we never ask it from nowhere. The market you picked at setup is carried into every single query, and it is set per prompt, so one workspace can track several markets at once.

Engine familyHow the market is applied
AI Overviews, AI ModeThe Google query is run from that country, so you get the result a buyer there would actually see
PerplexityThe model is told to answer for that country, and its web search is bound to the same country
ChatGPT, Claude, CopilotThe model is told to answer as it would for a user in that country

The language follows too: your questions are written in your buyers' language, not translated from English, which is why a German prompt set reads like a German buyer wrote it.

The Prompts table, with one row per question and the engines that cite you highlighted
Figure 1One row per question. A highlighted engine icon means you were cited in that answer, and Your rank is your best position across the engines that cite you.
i
The run fills in progressively, engine by engine. You do not have to wait on the screen: open the dashboard and watch the rows land.

What to read first

Your scorecard is the top row of the Overview. Four numbers, and each answers a different question.

The four KPIs at the top of the Overview screen
Figure 2The scorecard on our Nike demo workspace. Every screenshot in this documentation comes from that same workspace, so you can follow one example end to end.
MetricAnswers
Visibility scoreOn what share of the analysed answers does your brand appear at all?
Average positionWhen you do appear, how high are you in the answer?
Share of voiceOf all the brand mentions on your prompts, how many are yours?
Positive sentimentWhen engines talk about you, how do they talk about you?
!
Do not read a trend yet. Your first run is a baseline, not a direction. The arrows compare each prompt and engine with its own previous measurement, so they only start meaning something once you have a second week behind you.

Every Monday

Your questions are re-asked automatically and a new run lands. That is the whole point of the product: a single audit tells you where you stand, the weekly repeat tells you what changed and whether what you did worked.

The site audit runs separately, on Tuesday. It looks at your site as it is that day, so a change you ship on Monday evening is already in it.

Your first action

Every run ends with a ranked list. Take the card at the top of To do, the one with the highest estimated impact, do it, and read the following Monday. One action per week, measured, beats ten done blind.

When a number looks wrong

You seeWhat it meansWhat to do
No average positionEither you were never named, or you were named in prose without any ranking. Both leave nothing to averageOpen Prompts. The Absent filter lists the questions that never name you; the rest are mentions the engine wrote without ranking anything
No competitor suggestedYour topics are too narrow, or your site is too new to rank alongside anyoneAdd two or three by hand in Competitors
An engine still spinningIts answers have not come back yet, or some failedNothing. A failed pair is retried on the next run
Prompts that miss the pointThe generated set did not catch what you sellDelete them and write your own, then read guide 2
Documentation ← API reference