GEO tool for agencies: measure and report your clients' AI visibility

A client account in Refractia can hold several brands, with two roles (client admin and viewer) and per-client limits configured by Refractia. Each brand has its own identity profile, competitors, questions and scan history. If a client sells lines that compete in different arenas, each line is measured separately under the same brand (Product lines).

Scans are started by a person, from your team or from Refractia's. That leaves the calendar in the agency's hands: you measure at kickoff, after a content delivery, and before the results meeting.

• Command Center as a PDF: visibility, share of voice, sentiment, engine ranking and competitor arena for the scan you choose, in a single document.
• Partial report as a PDF, and a group report for clients with several product lines: an executive summary per line plus each line's full report.
• Technical Workshop in PDF and Excel: 56 checks with an evidence grade (A, documented or mechanically verifiable; B, independent study; C, convention with zero weight on the score). It gives the client's development team a list that separates defects from observations.

Every rate appears with its sample ("k of n"), and an engine that didn't answer is not counted as zero. That heads off the awkward question in the meeting: "where does this percentage come from?"

Each market is measured separately and in its own language, with the country named in full in every question. Google AI Overviews are fetched with the country's location for Argentina, Mexico, Colombia, Chile, Peru, Uruguay, the Dominican Republic, Brazil, Spain and the United States. For a regional client, that means one measurement per market instead of an average that describes none of them.

Refractia handles the measurement and the diagnostic; the work that moves the numbers can be done by your agency. One possible structure, using the platform's modules:

1. 01

   Baseline diagnostic

   A first scan: whether the client appears, in which questions, against which rivals (Share of Voice), how it is described (AI Perception) and which sources are cited.

2. 02

   Technical fixes

   The Technical Workshop backlog: AI crawlers blocked by the CDN even though robots.txt allows them, pages AI can't extract, and grade-A defects first.

3. 03

   Content

   The content plan from the GEO analysis and the Content Workshop tasks, line by line (GEO Optimization).

4. 04

   Third-party sources

   In Refractia's data, a brand's own site is between 0.5% and 2.8% of what AI cites. Work with media, directories and comparison sites is a service in its own right (Influence Sources).

5. 05

   New measurement

   A scan after the work. Evolution shows the trend for the client and for each competitor across scans.

We have a program for partner agencies; terms are agreed case by case. If you want to see the platform with one of your clients' brands, the starting point is a free diagnostic.

• Refractia measures five engines: ChatGPT, Claude, Gemini, Google AI Overviews and Muse AI (Meta AI in WhatsApp, Instagram and Facebook). It does not measure Perplexity, Copilot, Grok, DeepSeek, or Google AI Mode as a separate surface.
• It is not continuous monitoring: each scan is started by a person.
• Plans are quoted; there are no public prices and no online checkout.
• No tool can promise that AI will name a client. What you can promise is to measure with a method and show what changed.