The State of AI Visibility for Portuguese B2B SaaS, 2026

Technical readiness: open by default, not by decision

The dominant technical posture among the 45 companies is absence of decision, not blocking. 96% (43 of 45) have no explicit policy for AI crawlers (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and similar) in their robots.txt. The two exceptions both allow access, so where a decision exists at all, the direction is openness. In practice, AI crawlers can read almost every site in the sample, but almost no company has deliberately chosen that outcome.

Structured identity is missing in more than half of the sample. 58% (26 of 45) do not publish an llms.txt file, and 58% (26 of 45) do not expose structured Organization schema, the machine-readable identity that AI systems use to recognise who the company is. The honest reading is that these companies are not blocking AI, they are ignoring it: access is open by default, but the structured foundation that would make each brand legible and citable to models is absent in most cases.

Recommendation rate by engine

Claude recommends the largest share of Portuguese companies, at 40%, and is the engine that most often recognises Portuguese brands. ChatGPT (GPT-4o) recommends 30%. Gemini recommends 29% and tends to favour global players, making it the most international of the four. Grok also recommends 29%. The criterion behind these figures is deliberately strict: for each of the 31 categories, the buyer-style question was run in four engines, three times each, and a company only counts as appearing if it is genuinely recommended as an option, not merely mentioned in passing, in at least 2 of the 3 runs.

Citation with a linked source is close to nonexistent, at approximately 2% across all engines. With web search switched off, the models answer from memory (recall) and almost never cite sources with links. What this study measures is therefore recommendation by recognition, not citation with a source: whether the model knows the brand well enough to put it forward when asked about its category.

The distribution that matters: a visible core and an invisible majority

Only 29% of the 45 companies (13) are recommended by all four engines, the visible core of Portuguese B2B SaaS. At the other end, 56% (25 of 45) are recommended by none of the four assistants when asked about their own category, and 31% (14) are completely invisible: the AI does not even name them. The full distribution across the four engines is 29% recommended in all four, 16% recommended in 1 to 3 of the 4 engines, 24% mentioned but never recommended, and 31% completely invisible.

The middle band is thin. Between the 29% that every engine recommends and the 56% that no engine recommends, only 16% sit in partial visibility, recommended by some engines but not others. Visibility in this sample behaves less like a gradient and more like a divide: a company either belongs to the recognised core or it is effectively absent from the answer a buyer reads.

Results by category: where Portuguese SaaS wins and where it loses

Health and HR is the only category where every Portuguese company measured appears in every engine, a 100% recommendation rate, with Coverflex, Landing.jobs, Sword Health and Knok all surfacing. Below that peak, rates fall quickly: invoicing and ERP reaches 60%, contact center 50%, and developer tools, security and data approximately 25%.

Payments sits at 11%: nearly all Portuguese fintechs lose their place to Stripe and PayPal. Industry and operations sits at 7%, with Portuguese brands in the category invisible while the slot goes to Fiix, UpKeep and MaintainX. Hospitality and tourism is the floor at 0%: none of the Portuguese brands measured appears in any engine, despite being real and active companies, and the space belongs to Cloudbeds, SiteMinder and Mews. ChatGPT confirmed the pattern, also returning 0% in this category.

Who the AI recommends instead

Where Portuguese brands are absent, international players occupy the space, category by category: Stripe in payments, Cloudbeds, SiteMinder and Mews in hotel management, and Fiix, UpKeep and MaintainX in industrial maintenance. In almost every category, the Portuguese brand is pushed to the second or third sentence of the answer, or does not appear at all. The list a buyer reads is not empty, it is simply filled by companies from other markets with greater presence and authority in the material the models learned from.

What this means: invisibility is not chance

Crossing the technical audit with the visibility measurement shows that invisibility is not random. The companies that are not recommended tend to be the ones that also lack the technical base, with no schema, no AI policy and no extractable content, and/or brands that are too small, carry an ambiguous name, or have pivoted. AI recommends what it recognises with confidence; the rest is left out, and the space is filled by international brands with larger presence and authority. For a Portuguese B2B SaaS company, not being recommended by AI increasingly means not being in the purchase decision.

The convergence across engines strengthens the conclusion. That all four engines, including ChatGPT, the most widely used, land on such close numbers (29-40% recommendation) indicates this is not the quirk of one model: it is a cross-engine pattern in how AI, today, sees (or fails to see) Portuguese B2B SaaS.

This is not a judgement on the companies themselves: several are scale-ups and some are unicorns. It is a diagnosis of notability in the eyes of the models, a technical and editorial problem rather than a product-quality problem. And it is a solvable one.

Four questions to check today

The study reduces to four practical checks any company can run immediately. First, robots.txt: is there a conscious decision about AI crawlers? Allowing access is legitimate; not deciding means living with the default. Second, Organization schema with sameAs: does the site declare to AI who the company is and link its profiles? Third, llms.txt: is there a model-readable summary of the site? This is hygiene, not a guarantee. Fourth, the direct test: ask the four assistants about your own category. If they do not see you, you know where you stand.

The pattern also extends beyond B2B SaaS, with a caveat the study states explicitly. The same invisibility problem applies to local brands (clinics, restaurants, tourism, local retail), but there the visibility work rests on different signals: Google Business Profile, reviews, NAP consistency, LocalBusiness schema and local directories. Without public data of comparable rigour for those segments, the study avoids extrapolating its numbers; the method and the problem, however, are recognisably the same.