Method
We replay the same reading every week
A measurement that changes protocol between runs measures nothing. Georadar runs the same reading, at the same moment, on the same prompt plan, so the movement you read comes from the market rather than from the method.
The loop
Every publication is measured before and after
The four stages run continuously, and the baseline frozen at the outset serves as the comparison point for everything that follows.
-
01
Measure
Your strategic prompts are run across the four assistants with several repetitions, then each answer is analysed for the brands cited, their rank, their status and the tone used.
Output: share of voice, Radar Index, raw answers stored
-
02
Spot
Prompts where a competitor is ahead rise to the top, ranked by gap, and the ranking of cited sources points at the domains actually feeding those answers.
Output: a prioritised list of prompts and domains to target
-
03
Publish
Content is produced from the platform, approved by you, then put live on the network site whose editorial line matches the question being worked.
Output: dated articles and inserts, tied to their target prompt
-
04
Measure again
The next reading is compared with the baseline, crawler traffic reads page by page from the moment content goes live, and the prompt plan adjusts to what moved.
Output: movement since baseline, crawl rhythm per article
The protocol
The protocol does not shift mid-course
Comparability rests on a handful of rules held without exception. Here they are, so you can check that two points on the same curve really describe the same thing.
-
The same prompt planA prompt added along the way starts its own series and does not enter retroactively into the aggregates of earlier weeks.
-
The same momentThe automatic reading runs in the same slot every week, since answers vary with whatever is in the news.
-
The same settingsTemperature and call parameters are fixed per engine and versioned, so any change of setting shows up in the history.
-
The same number of passesThe repetition count is constant across the period, a rate computed on three passes not being comparable with one computed on ten.
-
Engines kept apartEach assistant keeps its own series, and the consolidated view serves only for the overall read.
-
Answers storedThe full text of every answer is archived and timestamped, which makes any figure traceable back to its source.
Le même enchaînement se rejoue chaque semaine au même créneau, ce qui rend deux points d'une courbe comparables.
La référence ne bouge jamais. Elle répond à « nos actions servent-elles », là où la semaine précédente capte les mouvements courts.
The comparison
A score always reads against a marker
Reading a score in isolation teaches you nothing. Georadar always sets it against two markers: your baseline, and the previous week.
-
The baselineFrozen when the campaign opens, it never moves. It is what answers the question "are our actions working".
-
The previous weekIt catches short movements, a publication that has just been crawled or a competitor pushing on a category.
-
The engine detailA rise carried by a single assistant does not read like an overall rise, and the consolidated view can hide it.
-
The prompt detailThe overall score rarely climbs everywhere at once: the detail shows which questions moved and which are holding out.
A cadence that holds
The calendar of an engagement
Baseline measurement. The first reading is frozen and becomes the comparison point for every measurement that follows.
Automatic full reading. All your prompts, all the engines, at the same point in the week, so the curves stay comparable.
One-off reading. Just before and just after a publication, a product launch or a move from a competitor.
AI traffic. The pixel records every crawler visit to your published pages, without waiting for the next reading.
Timescales
Assistants do not respond at the same pace
Assistants do not all respond to published content at the same pace, and conflating the two situations leads to misreading a flat curve.
Engines that browse directly
Perplexity, AI Overviews and ChatGPT in search mode consult the web at the moment of answering. Indexed and crawled content can appear there within days, which the AI traffic pixel shows you before the next reading even runs.
Horizon: days to weeksAnswers without browsing
When the assistant answers from its training, presence depends on the models' update cycles, whose calendar belongs to their publishers. Georadar tracks this series separately so it does not dilute the read on the faster engines.
Horizon: model training cyclesOur commitments
You can check what we claim
Assistant answers are not advertising slots, so progress is built through content rather than bought. Our commitment covers the mechanism, and it can be checked week after week.
A reproducible reading
The same protocol every week, multiple repetitions, raw answers stored and readable prompt by prompt.
An open calculation
The six Radar Index axes and their weightings are published, so you can redo the calculation yourself.
Engines kept apart
Assistants that browse the live web and those answering from training are tracked separately, which shows where the progress is actually happening.
Dated production
Every published article carries its date, its target prompt and its host site, with AI traffic recorded page by page.
Go further
Your baseline can be set today
Create your account, approve the prompt plan proposed, and the first reading launches straight after. That reading becomes your comparison point.