Executive summary
AI visibility changes with the surface being observed
Unprompted commercial brand presence, brand divergence, aggregated rankings and differences in source ecosystems.
Market share, preference of the Italian population, sales, conversions or commercial causality.
September 3–5, 2026 · three complete runs · Europe/Rome.
In the observed panel, Google AI Mode introduces at least one commercial brand that was absent from the query in 74.9% of observations. For Google AI Overview, the value is 50.5% across the full base and rises to 67.3% when the generative surface is actually present.
Brand sets and attributed source domains also differ by surface. Across the available pairwise comparisons, mean brand divergence ranges from 55.88% to 62.20%; source divergence ranges from 82.77% to 91.27%.
Scope by surface
| Surface | Observations | Surface availability | Unprompted brand |
|---|---|---|---|
| ChatGPT | 1,440 | 100.00% | 54.93% |
| Gemini | 1,440 | 100.00% | 64.10% |
| Google AI Mode | 1,440 | 100.00% | 74.93% |
| Google AI Overview | 1,440 | 75.07% | 50.49% |
Base: main panel; 20 branded controls are excluded. For Google AI Overview, surface availability is the share of observations in which the generative response appears.
Study base. 6,000 observations in total: 5,760 in the main panel and 240 branded-control observations. “Brand present” is not the same as “brand recommended” or “site cited”.
01 · Brand presence
How often does an AI introduce a brand that was not named in the query?
Unprompted Commercial Brand Visibility is the share of observations in which at least one commercial brand appears even though it was absent from the query. It measures spontaneous presence in the response text, not user preference.
Presence of at least one unprompted commercial brand
Share of main-panel observations · 0–100%
Base: 480 queries × 3 runs = 1,440 observations per surface. The 20 controls are excluded.
Source: Telescop Research · September 3–5, 2026 · research summary v1.0.
AI Overview: two denominator choices
The same brand-presence concept measured on the full base or only when AIO is present
Across all observations
Base: 1,440 main-panel observations.
When AIO is present
Base: 1,081 observations with a generative response.
AIO appears in 75.07% of main-panel observations. Surface availability and brand presence inside the surface are separate metrics.
Average number of unprompted brands per observation
Main-panel average · scale 0–2.5 brands
Source: Telescop Research · brand metrics v1.2.
02 · Brand divergence
The same query produces different brand sets across surfaces
Divergence compares the brands found in responses for the same query/run combination. It is the complement of Jaccard similarity: 0% means identical sets; 100% means no brand overlap.
Mean brand divergence between pairs of surfaces
(1 − Jaccard) × 100 · comparisons where both responses contain at least one brand
The comparison base varies by pair and includes only observations where both responses contain brands. Brand-free responses are not treated as divergent brand choices.
No pair in the observed panel systematically returns the same brand set. This supports treating AI surfaces separately rather than collapsing them into a universal visibility ranking.
03 · Rankings
Google is the most frequently present brand in all four panel rankings
Second place differs by surface: Samsung on ChatGPT, Apple on Gemini, Facile.it on AI Mode and Segugio.it on AI Overview.
Top ten brands by surface
Individual-brand unprompted visibility · share of eligible observations
ChatGPT
| Pos. | Brand | Visibility |
|---|---|---|
| 1 | 14.44% | |
| 2 | Samsung | 6.74% |
| 3 | Meta | 4.86% |
| 4 | Amazon | 4.65% |
| 5 | MediaWorld | 4.03% |
| 6 | Apple | 3.75% |
| 7 | Microsoft | 3.40% |
| 8 | Shopify | 2.92% |
| 9 | Unieuro | 2.71% |
| 10 | Lenovo | 2.36% |
Gemini
| Pos. | Brand | Visibility |
|---|---|---|
| 1 | 16.88% | |
| 2 | Apple | 6.67% |
| 3 | Amazon | 5.69% |
| 4 | Facile.it | 5.69% |
| 5 | Samsung | 5.63% |
| 6 | Meta | 4.65% |
| 7 | Motorola | 3.20% |
| 8 | Luce-Gas.it | 2.71% |
| 9 | Segugio.it | 2.57% |
| 10 | CapCut | 2.36% |
Google AI Mode
| Pos. | Brand | Visibility |
|---|---|---|
| 1 | 16.39% | |
| 2 | Facile.it | 11.32% |
| 3 | Segugio.it | 7.85% |
| 4 | Amazon | 6.18% |
| 5 | Samsung | 6.18% |
| 6 | Apple | 5.97% |
| 7 | Meta | 5.56% |
| 8 | Microsoft | 4.17% |
| 9 | Motorola | 3.41% |
| 10 | Skyscanner | 3.26% |
Google AI Overview
| Pos. | Brand | Visibility |
|---|---|---|
| 1 | 10.28% | |
| 2 | Segugio.it | 7.15% |
| 3 | Facile.it | 6.25% |
| 4 | Microsoft | 4.24% |
| 5 | Amazon | 3.26% |
| 6 | CapCut | 2.71% |
| 7 | Shopify | 2.15% |
| 8 | Moto Guzzi | 2.10% |
| 9 | BPER | 2.08% |
| 10 | WooCommerce | 1.60% |
For each brand, the denominator excludes queries that already name that brand. Multiple brands can appear in the same response, so percentages do not sum to 100.
These rankings describe the selected sectors and query panel. Google’s presence as a brand is not a measure of the performance of Google’s AI products, and observed differences do not by themselves prove intentional platform preference.
04 · Information sources
The surfaces draw on very different source sets
The source analysis compares domains attributable to responses, separately from brands mentioned in the text. Divergence describes overlap between domain sets; it does not score the authority or quality of any single source.
Comparability. The aggregated source comparison includes ChatGPT, Gemini and AI Overview. AI Mode is excluded from this part because citation metadata changed between provider runs; it remains included in the brand analysis.
Mean divergence of attributable domains
(1 − Jaccard) × 100
Examples of recurring source ecosystems
- ChatGPT
- ARERA, Treccani, IVASS, Shopify, support.google.com, IBM
- Gemini
- Facile.it, Papernest, Reddit, Aranzulla, Tom’s Hardware
- Google AI Overview
- YouTube, Facebook, Facile.it, Wikipedia, Reddit, Instagram, TikTok
Recurring sources in AI Overview
Incidence per observation · full panel including controls
Base: 1,500 AIO observations (500 queries × 3 runs). Each domain is counted at most once per observation.
05 · Run-to-run comparison
The ordering of the surfaces is unchanged across all three runs
For the presence of at least one unprompted brand, AI Mode remains highest, followed by Gemini, ChatGPT and AI Overview on the full denominator.
Unprompted visibility across the three runs
Share of observations · range in percentage points (pp)
| Surface | Run 1 | Run 2 | Run 3 | Max − min |
|---|---|---|---|---|
| ChatGPT | 56.67% | 54.79% | 53.33% | 3.34 pp |
| Gemini | 64.58% | 65.63% | 62.08% | 3.55 pp |
| Google AI Mode | 75.00% | 76.46% | 73.33% | 3.13 pp |
| Google AI Overview | 50.42% | 49.79% | 51.25% | 1.46 pp |
Base: 480 queries per surface and run; controls excluded. The range is descriptive, not a confidence interval.
The observed range is 1.46 to 3.55 percentage points. This stability applies to the aggregate metric during the study window; it does not imply identical individual answers or permanent future stability.
06 · Method & limitations
Study design and reading rules
The study is descriptive and uses a deterministic panel frozen before analysis. It is not a representative survey of the Italian population. The full process is documented in the public methodology.
- Panel
- 480 main-panel queries plus 20 branded controls across ten sectors.
- Surfaces
- ChatGPT, Gemini, Google AI Mode and Google AI Overview.
- Replicates
- Three complete, separate runs during September 3–5, 2026.
- Primary brand metric
- Unprompted Commercial Brand Visibility.
- Divergence
- Pairwise complement of Jaccard similarity on observed brand/source sets.
- Guardrail
- Observed presence is not market share, recommendation, sales or causal impact.
Limitations that should travel with the findings
- The panel represents the designed query set, not all Italian searches or all user behavior.
- AI systems and provider interfaces can change after the observation window.
- Brand and source extraction uses operational definitions documented in the methodology.
- AI Mode source metadata was not comparable across all runs and is excluded from the aggregate source-divergence analysis.
- The three runs describe short-window variability; they are not a longitudinal forecast.
07 · Applications
Who can use this research
The study is useful for teams that need to understand how brand and source representation changes by AI surface, and to turn that observation into more precise questions for further analysis.
Brand teams
Compare whether brand presence changes across ChatGPT, Gemini and Google AI surfaces.
PR & authority
Observe which third-party domains recur in answers and identify source territories worth investigating.
Competitive intelligence
See where brands are introduced spontaneously and where competitive sets diverge across surfaces.
Measurement
Establish a comparable baseline without compressing different surfaces into one number.
The research does not automatically assign causes to observed gaps. It narrows the field: surface, query, brand and source patterns to investigate next.
08 · Materials & citation
Inspect and reuse the public materials
The links below provide access to the frozen panel, aggregated results and methodology documentation. Reuse conditions are governed by the record deposited on Zenodo.
How to cite
English web report
Gentian Hajdaraj (2026). AI Search Visibility in Italy: brands and sources in ChatGPT, Gemini and Google AI responses. Telescop Research. English web edition. https://telescop.it/en/research/ai-search-visibility-italy/
Frozen dataset and methodology
Gentian Hajdaraj. (2026). AI Search Visibility in Italia — Dataset e metodologia v1.0 (Version v1.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.22640375
When citing a numerical result, keep the figure/surface, observation window and denominator together with the value. The DOI identifies the frozen v1.0 deposit; this English editorial edition does not create a new dataset.