Presence
6 %
The share of answers in which the brand appears, across questions that do not name it. This is the acquisition measure: it says whether the brand surfaces when nothing points to it.
THE GOAL
Turning every AI search engine into a channel for influence, acquisition and conversion
The scope, measures and action plan of a GEO and APO analysis, walked through on a worked example. Vernaz is a fictional company: a French manufacturer of technical lighting for architecture and commercial buildings, mid-way through a website rebuild, whose agency is driving its growth. The figures and the structure of the deliverable are those of real analyses.
The scoping, analysis and action plan are produced on the ResonanceMetrics platform, which we co-founded.
The scope measured
An analysis is judged not by its thickness but by what it covers. Here is the reach of a standard case.
- 315
- questions put to the engines
- 966
- answers collected
- 4
- engines queried
- 3
- audiences modelled
- 5
- competitors tracked
- 200
- website pages analysed
- 15
- authority domains identified
- 30
- days of observation
The four engines queried are ChatGPT, Claude, Gemini and Perplexity. The questions span the whole journey, from exploration to decision, and are distributed across the audiences defined at scoping.
A buyer in trade distribution
Looking for a dependable supplier. Decides on lead times and consistency of quality.
A specifying architect
Looking for a technical answer. Decides on compliance and performance.
An electrical contractor
Looking for practical material. Decides on ease of installation and availability of advice.
The audiences are defined at scoping with the agency, from what it knows about its client. They are what determines the questions put to the engines.
Six measures, six levers
Each measure corresponds to a distinct lever. The values below are those of the worked example.
Sentiment
68 / 100
What the engines say about the brand once they know who is being discussed. Here: favourable, in first position, with no unfavourable citation.
Authority
40 %
The share of the sector's reference sources in which the brand is cited. Six domains secured out of fifteen identified.
Semantics
29 %
The gap between the vocabulary the brand claims and the words the engines use to describe it. Seven times out of ten, neither the words nor the messages are picked up.
Source ranking
3e
Where the brand's website ranks among the sources the engines consult on its own market. Two competitors are ahead of it.
Technical
59,8 / 100
How readable the website is to an AI agent: structure, markup, structured data, performance. Here, no structured data at all across the entire catalogue.
These six measures are not there to produce a score. They are there to identify, for each one, the workstream that will move it.
Vernaz is known to the engines and well regarded once named to them. It is absent from answers when nothing points to it — which is precisely where acquisition happens. The brand does not have a reputation problem, it has a presence problem.
From finding to workstream plan
This is the part that separates an analysis from a monitoring report. The deliverable ends with an ordered list of actions, each with an owner, a level of effort and a priority rank.
The order is not negotiable: until the first two workstreams are handled, no editorial work produces a measurable effect.
| Rank | Nature of the action | Owner | Effort |
|---|---|---|---|
| 1 | Roll out product markup across the whole catalogue | Agency | Medium |
| 2 | Write the missing metadata | Agency and brand | Medium |
| 3 | Fix the category pages returning errors | Agency | Low |
| 4 | Produce the priority comparison content | Brand | High |
| 5 | Take a position on the missing sector references | Brand | Ongoing |
Every line can be priced by the agency against its own costs. That is what turns the analysis into a commercial proposal.
Down to page level
The analysis does not stop at site level. Every page in scope is assessed on three axes and gets its own actions.
Brand alignment
What the page says about the brand, against what the brand claims.
Audience intent
What the page answers, against what the audiences are looking for.
Technical fitness
What an AI agent can actually extract from the page.
| Page family | Pages | Brand alignment | Intent | Technical fitness |
|---|---|---|---|---|
| Product pages | 142 | 54 | 65 | 15 |
| Collection pages | 39 | 49 | 57 | 12 |
| Service pages | 19 | 44 | 52 | 0 |
An identical technical score across every page in a family is the signature of a template problem — and therefore of a single fix that applies across the whole site.
What gets measured, and when
The deliverable sets targets at three and six months, measured with the same instrument as the starting point. That is what lets the agency evidence its work to its client.
| Indicator | Start | 3 months | 6 months |
|---|---|---|---|
| Presence in AI answers | 6 % | 12 % | 20 % |
| Structured data | 0 | 80 | 90 |
| Authority domain coverage | 40 % | 53 % | 67 % |
| Semantic alignment | 29 % | 38 % | 47 % |
The targets are projections based on the scale of the planned fixes and on the assimilation lag observed across the engines. They are not a guarantee of results.
What the agency does with it
The deliverable is handed over in its brand. It serves three purposes, depending on the moment.
In pre-sales
A priceable workstream plan, presented to the client as the agency's own recommendation.
In a pitch
An AI component documented on real findings, inside a pitch response.
In ongoing management
A quarterly measurement baseline that evidences the work done to the brand.
The next step happens on one of your clients
This page shows the structure of an analysis, not its content. The content only exists in relation to a specific brand, its market and its competitors. We can show you that on an account in your portfolio.
We never approach a brand introduced by an agency directly. An NDA is signed before any conversation that names one.