Skip to content
Georadar
Sign inCreate an account

Asking once proves nothing

Querying an assistant once proves nothing, since the same prompt never returns the same answer twice. Georadar replays each question several times on each engine, which turns an impression into a usable measurement.

4
assistants queried
5
repetitions per prompt and per engine
20
answers analysed for a single prompt

Input

The prompt is run in real conditions

A prompt is never measured in isolation: the platform runs the question in the conditions your customers ask it, and keeps everything needed to explain a figure afterwards.

  • The prompt and its categoryThe question as a customer would phrase it, attached to a product world so results can be read by range.
  • The engine and its modeChatGPT with and without browsing, Perplexity, Gemini and Claude, each queried separately since their answers diverge.
  • The repetitionsThe same question replayed several times, so the variability inherent to the models is smoothed before any calculation.
  • The raw answerThe full text returned by the assistant, timestamped and stored, viewable at any time from the prompt record.
prompt #17 · relevé du 23 mars

Quelle marque de café en grain choisir ?

Rép. 1 rang 2 recommandée positif
Rép. 2 rang 3 citée neutre
Rép. 3 absente
Rép. 4 rang 1 recommandée positif
Rép. 5 rang 4 citée neutre
80 %taux de mention
40 %taux de recommandation
2,5rang moyen
0,80stabilité

Output

Every answer is broken down

Each answer is analysed for four separate pieces of information, which then feed the Radar Index and every tracking screen.

01

The brands cited

Every brand present in the answer, including those you were not yet tracking.

02

The rank

Where each brand appears in the answer, a citation opening the text being worth more than one closing a list.

03

The status

Brand merely listed or actively recommended, counted by two distinct rates.

04

The tone

Positive, neutral or negative mention, so you know in what terms you are described.

How it runs

How a weekly reading runs

  1. 01
    Queue

    All active prompts are queued with their engines and repetition count, and the run cost is estimated before launch.

  2. 02
    Query

    Calls go out in batches to each assistant, with temperature fixed so the protocol stays comparable week to week.

  3. 03
    Analyse

    Each answer goes through extraction, which returns the brands, their rank, their status, their tone and the sources cited.

  4. 04
    Aggregate

    Repetitions are consolidated into rates, share of voice is computed against the other brands, and the Radar Index lands.

Go further

Your real competitors are the ones the assistant cites Every brand cited is extracted, including those you were not watching, with its share of voice and its trajectory. Explore The map of the ground to cover A ranking of the domains assistants lean on to answer your prompts, with your own domains flagged. Explore Your measurement plan stops being guesswork Search Console builds your measurement plan from your real queries, GA4 closes the loop on your conversions. Explore

All features

Your own figures beat any demonstration

Create your free account, approve your prompt plan and launch the first reading. What this page describes then appears on your own brand.

Create my free account Book a demo