Case study · Artificial intelligence
Making measurablea visibility nobody measures
A company knows exactly where it stands on Google, on social platforms and on ad networks. It has no idea whether an AI cites it. With LLM Monitor, we built the method that reconstructs that missing data.
- Client
- LLM Monitor
- Sector
- Marketing and artificial intelligence
- Our role
- Product design, data and AI
- Collaboration
- Market Connect
Which solution should I pick to track my brand's visibility?
Severalsolutionsaddressthisneed.Solution Aisoftencitedforcompetitivetracking,Solution Bforthedepthofitsreporting.Your branddoesnotappearinthisanswer.
What the answer actually contains
- Brand cited
- no
- Rank
- —
- Competitors cited
- 2
- Sources consulted
- 3
That absence — or that presence — is what LLM Monitor measures, one answer at a time.
01 — The blind spot
Four instrumented channels, a fifth one invisible
Every acquisition channel has its measuring instrument, and the habit is so old that nobody notices it any more. Answers generated by artificial intelligence are the exception: models publish no citation statistics at all. A brand can be recommended a thousand times a day, or never, without a single dashboard saying so.
Available measurement sources
- SearchSearch Console
- AdvertisingAd platforms
- SocialNative analytics
- AI answersNo source
The questions that turn strategic
- Is the company cited in the answers at all?
- How often does it appear?
- Which competitors are recommended ahead of it?
- On which topics and in which contexts is it mentioned?
- Which models surface it, and which ignore it?
- Does the result change with the model version, the persona or the wording?
- Which sources do the models consult to build their answer?
02 — The reversal
Since the statistic does not exist, it has to be manufactured
Rather than waiting for platforms to publish figures they have no reason to publish, the system works the other way round: it simulates a large number of real search situations, queries several models and several versions of the same model, then analyses what comes back. The data is not collected, it is reconstructed.
The combined axes
Personas
- SME owner
- Marketing lead
- Technical buyer
Intents
- Compare
- Choose
- Research
Topics
- Pricing
- Integrations
- Compliance
Wordings
- Direct question
- Scenario
- List request
Models
- Model A
- Model B
- Model C
Versions
- v1
- v2
- v3
personas × intents × topics × wordings × models × versions
that many queries to orchestrate, collect, then compare with one another
03 — The technical chain
From a conversational answer to a row of data
Most of the product's value happens before the screen. An AI answer is free text: for it to become comparable with thousands of others, it has to be orchestrated, collected, stripped of its entities, reconciled across name variants, then dated.
- 01OrchestrationPrompt scenarios are generated and campaigns are scheduled.
- 02QueryingCalls to the various models and their versions, at a controlled rate.
- 03CollectionFull answers are retrieved, along with the sources they cite.
- 04ExtractionBrands, competitors and referenced domains are identified.
- 05NormalisationName variants reconciled, duplicates removed, entities attached.
- 06HistoryDated storage, so one campaign can be compared with the previous one.
04 — The reading
What you see once the data has been reconstructed
Answers then become measurements: a share of voice, a rank against competitors, a map of the sources models consult, and a curve telling you whether things are improving.
Visibility score
share of answers citing the brand
Share of voice
Citations by persona and competitor
| You | C1 | C2 | C3 | |
|---|---|---|---|---|
| Owner | 2 | 7 | 5 | 3 |
| Marketing | 1 | 6 | 8 | 4 |
| Buyer | 4 | 5 | 3 | 6 |
| Technical | 3 | 8 | 4 | 2 |
Domains most consulted by the models
- domaine-a.fr
- domaine-b.com
- domaine-c.org
- domaine-d.fr
- domaine-e.com
Trend across six campaigns
Interface representation. The values shown illustrate how it reads; they are not a real brand's results.
The product
The visibility report: score, share of voice and average position
How it moves over time, one AI engine at a time
The sources models actually consult
Real product screenshots. The tracked brand, its competitors and their domains have been replaced with neutral labels: a case study has no business publishing the visibility of companies that did not ask for it.
05 — From measurement to decision
A dashboard is only worth what it sets in motion
The product does not stop at displaying metrics. The collected data feeds analyses, then concrete recommendations: what to write, on which topics, and why.
- Why the brand is missing
Answers that leave it out share traits: a persona, a topic, a wording. The product isolates them.
- Who takes the space instead
The ranking of cited competitors, by topic and by model, shows where the ground is taken and by whom.
- Which content feeds the models
The map of consulted domains shows what models actually read to build their answer.
- Which topics to work on
Subjects where the brand is absent while demand exists become a list of editorial priorities.
06 — What we built
A data product, not an interface
LLM Monitor started from an observation with no existing tool behind it. The method had to be invented before the screen could be drawn: this is data, orchestration and artificial intelligence work, of which the interface is only the visible part.
- Prompt orchestration
- Multi-model querying
- Version handling
- Large-scale collection
- Entity extraction
- Brand normalisation
- Citation and source analysis
- History and comparison
- Statistical aggregation
- Recommendation generation
- Data visualisation
Research and innovation
The project our CIR-CII accreditation rests on
LLM Monitor is the work for which MZ Informatique obtained its French research and innovation tax credit accreditation, granted by the ministry responsible for Research. This is not a decorative label: reconstructing a statistic nobody publishes takes genuine research work, and that is precisely what the accreditation certifies.
What it changes for you: sums you entrust to us for a research or innovation project count towards your own tax credit base. Without the accreditation, the very same invoice would not.
In short
SEO for conversational engines
What search engine optimisation did for search engines, LLM Monitor does for language models: measure first, understand next, act last. Visibility inside AI answers stops being a hunch and becomes an indicator tracked over time.
A similar project?
Tell us about your situation. The first conversation is free and without obligation, on site in Nice and the surrounding area.
See our artificial intelligence offerSee all our case studies