Cross-engine consistency of AI visibility for Portuguese consumer-service brands, 2026

Fragmentation, not just absence

The central result is not merely absent visibility, it is fragmentation. A brand can be recommended by one assistant and be absent from the other six. AI visibility is not a fixed attribute of the brand: it is a relation between the brand, the engine, the search intent, the wording of the prompt, the sources the engine retrieves, the sector and the location. Measuring presence in a single engine therefore overstates a brand's real position.

The distribution across the 84 brands makes the fragmentation concrete: 26 brands (31.0%) were recommended by no engine, 20 (23.8%) by only one, and 46 (54.8%) were limited to zero or one. At the top, eight brands (9.5%) were recommended by five or more engines, and only two (2.4%) by all seven. The full distribution runs: 0 engines, 26 brands; 1 engine, 20; 2 engines, 14; 3 engines, 9; 4 engines, 7; 5 engines, 4; 6 engines, 2; 7 engines, 2. The mean was 1.77 engines per brand.

Engine-by-engine results

Recommendation rates vary considerably across engines. Perplexity, with live web search, recommends the most brands; DeepSeek and Mistral the fewest. All percentages are over the 84 brands: Perplexity recommended 32 brands (38.1%), ChatGPT 28 (33.3%), Gemini 25 (29.8%), Claude 22 (26.2%), Grok 18 (21.4%), Mistral 13 (15.5%) and DeepSeek 11 (13.1%).

Mentions are broader than recommendations in every engine, and the ordering is the same. Perplexity mentioned 45 brands (53.6%), ChatGPT 39 (46.4%), Gemini 36 (42.9%), Claude 33 (39.3%), Grok 28 (33.3%), Mistral 22 (26.2%) and DeepSeek 19 (22.6%). In each engine, only a subset of the brands mentioned went on to meet the recommendation criterion.

Citations narrow the funnel further. Perplexity cited 27 brands (32.1%), ChatGPT 17 (20.2%), Gemini 15 (17.9%), Claude 12 (14.3%), Grok 8 (9.5%), Mistral 6 (7.1%) and DeepSeek 4 (4.8%). At every level (mention, recommendation, citation) the gap between the strongest and the weakest engine is roughly a factor of two to eight, which is why single-engine measurement misleads.

Recommendation and citation do not always travel together

At the level of brand and engine combinations, 588 combinations were observed (84 brands by seven engines). Of these, 222 had a mention (37.8%), 149 had a recommendation (25.3%) and 89 had a citation (15.1%). Of the 149 recommendations, 74 (49.7%) were accompanied by a qualifying citation. In other words, about half of all recommendations are not backed by an explicit source. A brand can be recognised without being anchored to a reference, and a citation does not, by itself, represent trust, quality or endorsement.

The three levels are defined restrictively. Mentioned: the brand or its domain appears in the answer, whether or not it is framed as a recommendation. Recommended: the brand is explicitly presented as a suitable option in at least two of three runs of at least three distinct prompts, at least one of them a bottom-of-funnel or persona-and-location prompt. Cited: the answer includes an identifiable link or reference to the brand's site, or to an independent source supporting the recommendation, in at least two responses.

Sector-by-sector results

Consistency varies by sector. In dental clinics and residential solar energy, the average brand is recommended by about 2.4 engines; in residential building and renovation works, by one. Half of the building-works brands are recommended by no engine at all. Each sector contains only 12 brands, so these values are indicative, not conclusions about an entire sector.

Average engines recommending each brand, by sector: dental clinics 2.42, residential solar energy 2.42, real estate agencies 2.17, language schools 1.67, home care and senior residences 1.58, gyms and fitness 1.17, residential building and renovation works 1.00. Brands with no recommendation from any engine: dental clinics 2 of 12 (16.7%), residential solar 2 (16.7%), real estate 3 (25.0%), language schools 4 (33.3%), home care and senior residences 4 (33.3%), gyms and fitness 5 (41.7%), building works 6 (50.0%).

The share of brands recommended in at least three engines follows the same ranking: dental clinics 5 of 12 (41.7%), residential solar 5 (41.7%), real estate 4 (33.3%), language schools 3 (25.0%), home care and senior residences 3 (25.0%), gyms and fitness 2 (16.7%), building works 2 (16.7%). Brands recommended in at least one engine range from 10 of 12 (83.3%) in dental clinics and residential solar down to 6 of 12 (50.0%) in building works.

Three anonymised examples

Three brands from the dataset, with no commercial names, show how mention, recommendation and citation separate in practice. DEN 12, a dental clinic brand, was mentioned and recommended by all seven engines but cited by only one. It appeared in 137 responses (43.5%) as a mention, in 80 (25.4%) as a recommendation and in 7 (2.2%) with a citation, at an average position of 3.49. Even a brand recommended everywhere can be almost never anchored to a source.

IMO 10, a real estate brand, was mentioned and recommended in four engines and never cited. It registered mentions in 68 responses (21.6%), recommendations in 40 (12.7%) and citations in 0 (0.0%), with an average position of 3.84. This is the middle of the distribution: reasonable recognition in some assistants, complete absence in others, and no explicit source behind any of it.

OBR 06, a residential building and renovation brand, was mentioned in two engines, met the recommendation criterion in none, and was not cited. It appeared as a mention in 13 responses (4.1%), as a recommendation in only 4 responses (1.3%), too few to meet the threshold in any engine, and with 0 citations (0.0%), at an average position of 8.06. This profile, sporadic and low-ranked appearances, describes a large share of the sample.

How this differs from previous destaque.ai studies

This study is not directly comparable to the two earlier destaque.ai studies. The 45-company study analysed B2B SaaS across four engines with web search switched off, so it measured model recall. The 251-company barometer crossed organic Google strength with mention in a single engine. This study measures consistency across seven engines, in consumer services, with live web search active. Numbers from different studies should not be summed or compared directly: they use different samples, engine sets and retrieval regimes.