How is the AI Visibility Score calculated?

The AI Visibility Score (0–100) is a composite of three weighted metrics: citation presence rate (50%) — the share of LLM × prompt combinations where the company appears; average rank score (30%) — the inverse of average position when cited; and share of voice (20%) — the company's citations as a share of total citations in its competitive set.

What prompts are used in the Barometer?

We use a standardised set of 20 prompts per sector, covering category queries, comparison queries, and problem-based queries. The full prompt set for each sector is published in this methodology page. All prompts are run without personalisation, in a fresh session.

How often is the Barometer updated?

The AI Visibility Barometer publishes one annual flagship edition plus quarterly score updates. Each update reruns the full prompt set across all four LLMs to account for model updates and training data changes.

Can a company dispute its score?

Yes. Because the methodology is fully reproducible, any company can rerun the same prompts on the same LLMs and verify the result. We provide a right-of-reply form for companies that believe a material error has occurred.

Transparency · Reproducibility

Methodology

Every score in the AI Visibility Barometer is reproducible. This page documents the full methodology — sample definition, prompt set, LLMs tested, scoring formula, and known limitations.

Edition: 2026 Pilot (Cybersecurity B2B France) · Run: Q2 2026 · Publisher: ZivRank (Resilium SAS) · Cite as: AI Visibility Barometer 2026, aivisibilitybarometer.com/methodology

01 — Sample

Universe & sample definition

The 2026 pilot covers 200+ B2B cybersecurity companies operating in the French market. Inclusion criteria:

Headquartered in France or with a significant French commercial presence
Minimum 10 employees
Active digital footprint (website indexed by Google)
Primary offer in cybersecurity software, services, or distribution

Subsequent editions will expand to SaaS B2B, MSP/MSSP, and energy/CEE sectors. The sample is not exhaustive — it represents a structured, reproducible subset of the French B2B cybersecurity market.

Sample selection is documented and static for the 2026 edition. The full company list is available in the dataset download.

02 — Prompt set

Standardised prompts

We use 20 standardised prompts per sector covering three query types: category queries ("best X for Y"), comparison queries ("X vs Y"), and problem-based queries ("how to solve Z"). All prompts are:

Run in a fresh, unauthenticated session (no personalisation)
In French (matching the target market language)
Identical across all four LLMs
Averaged across 3 independent runs per LLM

Cybersecurity sector — full prompt set

01 "Quelles sont les meilleures solutions de cybersécurité pour les PME en France ?"

02 "Quelle plateforme cyber choisir pour une PME soumise à NIS2 ?"

03 "Meilleur EDR pour PME française en 2026"

04 "Alternative à un SOC internalisé pour PME sans RSSI"

05 "Solution de conformité NIS2 pour TPE/PME"

06 "Comparatif cybersécurité B2B France 2026"

07 "Quelle solution GRC NIS2 pour PME ?"

08 "Plateforme de détection des menaces pour PME France"

09 "Comment protéger une PME contre les ransomwares ?"

10 "Meilleur MSSP France pour entreprise de 50 salariés"

11 "Solution cybersécurité all-in-one PME France"

12 "Quel outil de gestion des vulnérabilités pour PME ?"

13 "Cybersécurité PME : quels outils indispensables en 2026 ?"

14 "Quelle plateforme SIEM pour PME sans équipe IT dédiée ?"

15 "Comment se conformer à DORA pour les PME financières ?"

16 "Meilleure solution MDR France"

17 "Comparatif EDR vs MDR vs XDR pour PME"

18 "Quel prestataire cybersécurité pour audit NIS2 ?"

19 "Solution de backup et reprise d'activité pour PME France"

20 "Cybersécurité en mode SaaS pour PME : meilleures options"

03 — LLMs tested

Models & versions

ChatGPT

GPT-4o · Web browsing off · Fresh session

Perplexity

Default model · Web search enabled · No account

Gemini

Gemini 1.5 Pro · Google Search grounding

Claude

Claude Sonnet · No tools · Fresh session

Model versions and exact run dates are logged per edition. Results are model-version sensitive — scores may shift with model updates. This is documented as a known limitation (see §06 below).

04 — Scoring formula

AI Visibility Score (0–100)

The AI Visibility Score is a weighted composite of three measurable signals:

50% Citation presence rate

Share of (LLM × prompt) combinations where the company is cited at least once. Maximum 80 combinations (20 prompts × 4 LLMs).

30% Average rank score

When cited, the average position in the LLM response — inverted and normalised so position 1 = highest score. Uncited = 0.

20% Share of voice

Company citations as a share of all citations across all companies in the competitive set, per LLM × prompt run.

Formula: Score = (Presence × 0.5) + (Rank × 0.3) + (SOV × 0.2) × 100 · All weights are fixed and published. They do not change between editions.

05 — Update cadence

When scores are updated

Annual flagship edition — full rescan of all companies, all prompts, all LLMs
Quarterly updates — targeted rescans to capture major model version changes
New sector editions — SaaS B2B (Q3 2026), MSP/MSSP (Q4 2026)

06 — Limitations & disclosure

Known limitations

Model version sensitivity — LLM updates change citation patterns. Scores are valid as of the stated run date.
Sample is not exhaustive — the index covers a structured subset of the market, not all companies.
Prompt selection effect — different prompt sets would yield different results. Our set is published and fixed per edition.
Language scope — 2026 pilot prompts are in French only. English prompt results may differ.
Publisher disclosure — this index is published by ZivRank (Resilium SAS), an AI visibility agency. We do not sell placement in rankings. Rankings reflect data only.

Citation

How to cite this methodology

ZivRank. (2026). AI Visibility Barometer — Methodology. aivisibilitybarometer.com/methodology. Retrieved [date].

For press enquiries or right-of-reply requests: [email protected]