We read the robots.txt, llms.txt and home page of 13,000 randomly drawn .pt domains. The results describe 10,160 active sites, weighted to correct sampling bias, and measure how prepared they are for search engines and AI assistants.
Four technical elements that an AI agent or search engine finds at a site's entrance. The study measures what the site declares, not results in the engines.
What it is: File at the domain root that tells each automated agent what it may read.
Why it matters: Each AI agent identifies itself by name. The file lets you allow or block training agents and search agents separately.
How it was measured: By-name blocking of 12 agents (Disallow: / in the agent group) and whether rules exist.
What it is: Text file at the root with a summary of the site and links to the main pages, designed for language models.
Why it matters: It is a proposal, not a standard. It gives a model a short description of the site without parsing the HTML.
How it was measured: A 200 response at /llms.txt with text (not HTML) and more than 20 characters.
What it is: Schema.org structured data inside the page that describes the organisation, services or products.
Why it matters: It states explicitly who the company is, where it is and what it offers, without depending on page text.
How it was measured: Presence of JSON-LD blocks on the home page and declared types (@type).
What it is: Sentence on the home page that summarises what the site offers.
Why it matters: It is the shortest text a search or answer engine has to describe the site. Without it, the summary depends on what the engine extracts.
How it was measured: Existence of <meta name="description"> with content.
21.7% of sites have no robots.txt with rules. Among those that do, 93.3% have no rule for AI agents, and 3.1% explicitly allow at least one. The chart shows the percentage of sites blocking each agent by name (Disallow: /).
Training agents: Bytespider, GPTBot, Amazonbot, meta-externalagent, ClaudeBot, CCBot, Google-Extended, Applebot-Extended, anthropic-ai. Search and user-request agents: ChatGPT-User, PerplexityBot, OAI-SearchBot.
Among sites that block at least one agent. The average is 6.7 agents per site.
11.5% of sites that block also stop at least one search or user-request agent. The other 88.5% only block training agents.
None of the 156 Wix sites block AI agents. On sites with no detected platform (custom code or unidentified platforms), blocking is 4.1%. PrestaShop, Joomla and Wix have 131 to 195 sites in the sample, so the differences are indicative.
11.0% of sites have a valid llms.txt. The distribution depends on the platform: Wix and Shopify generate the file by default.
| Platform | Share of sites | Valid llms.txt |
|---|---|---|
| WordPress | 48.9% | 8.0% |
| WooCommerce | 18.2% | 7.5% |
| Shopify | 2.6% | 92.8% |
| Wix | 1.6% | 92.3% |
| Joomla | 2.0% | 0.5% |
| PrestaShop | 1.3% | 5.8% |
| None detected | 41.2% | 6.4% |
A WooCommerce shop also counts as WordPress. SEO plugins for WordPress may generate the file, which the study does not distinguish.
E-commerce drops from 29.0% to 12.3% without Shopify. SaaS and technology leads with 14.2% adoption outside platforms that generate the file. 9.5% of sites answer /llms.txt with a generic HTML page, which was discarded.
40.4% of sites declare JSON-LD on the home page. The most common types describe the page and the site, and few describe the business.
Only the home page was read. Product usually sits on product pages, so the 0.3% figure does not describe shops as a whole.
| Sector | Organization | LocalBusiness | FAQPage |
|---|---|---|---|
| E-commerce | 36.7% | 1.8% | 0.8% |
| Real estate and automotive | 25.7% | 1.6% | 1.3% |
| Media, culture and personal sites | 27.8% | 0.6% | 0.4% |
| SaaS and technology | 33.6% | 2.5% | 5.0% |
| Tourism, food service and leisure | 25.9% | 2.1% | 1.2% |
| Education and training | 28.2% | 3.2% | 1.0% |
| Health | 23.0% | 2.6% | 1.3% |
| Institutions and associations | 17.8% | 0.8% | 0.1% |
| Industry and B2B trade | 29.8% | 1.5% | 0.3% |
| Professional services | 30.1% | 2.7% | 2.2% |
| Other identified sectors | 26.4% | 1.1% | 3.2% |
| Not enough information | 18.1% | 0.7% | 0.1% |
SaaS and technology has the most FAQPage (5.0%). Health and professional services, sectors with local customer service, declare LocalBusiness on 2.6% and 2.7% of sites.
By page language: 44.9% of Portuguese-language sites declare JSON-LD, against 34.4% of English-language sites.
62.5% of sites have a meta description. Sites without a description are also the hardest to classify: among the 2,225 without enough information to assign a sector, only 21.5% have a meta description, against 58% to 87% in the other sectors.
| Element | Sites |
|---|---|
| Final URL on HTTPS | 98.8% |
| Title on the home page | 99.7% |
| viewport | 92.8% |
| Language declared in HTML | 91.9% |
| Meta description | 62.5% |
The basic elements of a page are almost universal. The difference is in elements that depend on a decision: description, structured data and llms.txt.
Sum of 4 elements, 1 point each: robots.txt with rules and without AI agent blocking, valid llms.txt, JSON-LD on the home page and meta description. It is a simple index with equal weights, not validated against results in AI engines.
The index stays stable across the two sample groups (32.1% in group 1 and 33.2% in group 2).
The sector was assigned by a language model from the title and description of the home page. The 95% confidence interval refers to AI agent blocking.
| Sector | Sites | AI blocking | llms.txt | JSON-LD | Meta description | Mean index |
|---|---|---|---|---|---|---|
| E-commerce | 948 | 7.5% (5.9 - 9.0) | 29.0% | 50.9% | 87.0% | 2.47 |
| Real estate and automotive | 416 | 5.1% (3.5 - 7.3) | 10.2% | 40.6% | 83.0% | 2.12 |
| Media, culture and personal sites | 717 | 4.6% (3.2 - 6.0) | 8.2% | 44.7% | 70.8% | 1.99 |
| SaaS and technology | 463 | 3.9% (2.7 - 5.7) | 14.9% | 48.3% | 82.1% | 2.17 |
| Tourism, food service and leisure | 868 | 3.1% (2.0 - 4.4) | 14.2% | 46.7% | 71.7% | 2.09 |
| Education and training | 331 | 3.1% (1.5 - 4.9) | 11.5% | 45.4% | 65.5% | 1.99 |
| Health | 480 | 2.4% (1.4 - 3.5) | 11.5% | 48.3% | 73.0% | 2.14 |
| Institutions and associations | 952 | 2.2% (1.5 - 3.1) | 5.9% | 27.5% | 58.0% | 1.66 |
| Industry and B2B trade | 905 | 2.1% (1.4 - 3.0) | 8.9% | 42.7% | 76.8% | 2.06 |
| Professional services | 1,459 | 1.8% (1.2 - 2.4) | 11.9% | 47.0% | 74.6% | 2.13 |
| Other identified sectors | 396 | 4.6% (2.9 - 6.9) | 9.6% | 42.0% | 76.5% | 2.03 |
| Not enough information | 2,225 | 1.8% (1.3 - 2.3) | 5.4% | 28.2% | 21.5% | 1.23 |
E-commerce. It has the highest index (2.47) and the most agent blocking (7.5%, interval 5.9 to 9.0%, with no overlap with professional services). llms.txt drops from 29.0% to 12.3% without Shopify.
SaaS and technology. It leads in own llms.txt adoption (14.2%) and FAQPage (5.0%), but agent blocking (3.9%) is higher than in professional services and health.
Professional services. They have the lowest blocking (1.8%) and 42.1% with 3 or 4 elements. LocalBusiness appears on only 2.7% of sites, even though these are businesses with a local presence.
Health. 41.8% have 3 or 4 elements and 48.3% declare JSON-LD, above the overall average. llms.txt is on 11.5% of sites.
Institutions and associations. It is the most behind: 5.9% with llms.txt, 27.5% with JSON-LD, 58.0% with a meta description and 49.9% with 0 or 1 index element.
Five checks matching the elements measured. They do not guarantee citations by AI engines: they make the site readable to them.
Where you stand
Tick what your site already has and compare with the 10,160 sites in the study.
Tick the elements your site already has.
21.7% have no rules. Among those that do, 93.3% do not mention AI agents.
The file decides, agent by agent, what can be read. Without rules, every agent has access. Training agents and search or user-request agents can have different rules.
How to check: Type your-domain.pt/robots.txt in the browser. Look for names such as GPTBot, ClaudeBot, ChatGPT-User, OAI-SearchBot and PerplexityBot followed by Disallow: /.
Example of per-agent rules: one training agent blocked and search agents allowed
User-agent: GPTBot
Disallow: /
User-agent: ChatGPT-User
Allow: /
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /11.0% have one. Without Wix and Shopify, 7.4%.
A text file at the root with the company name, a two or three sentence description and links to the main pages.
How to check: Open your-domain.pt/llms.txt. If a page of the site appears instead of text, the file does not exist. On Wix and Shopify the file is generated: confirm the text describes the business.
Example llms.txt
# Company name
> What the company does, for whom and where, in two sentences.
## Services
- [Service A](https://example.com/service-a): short description
- [Service B](https://example.com/service-b): short description
## Contact
- [Contacts](https://example.com/contacts): address, phone and email40.4% have JSON-LD. 26.9% declare Organization or LocalBusiness.
Organization or LocalBusiness on the home page (name, address, contacts), FAQPage on frequently asked questions and Product on product pages.
How to check: Paste the home page address into validator.schema.org and confirm the declared types.
Example JSON-LD for a local business
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "LocalBusiness",
"name": "Company name",
"description": "What it does, for whom and where.",
"url": "https://example.com",
"telephone": "+351 000 000 000",
"address": {
"@type": "PostalAddress",
"streetAddress": "1 Example Street",
"addressLocality": "Lisboa",
"addressCountry": "PT"
}
}
</script>62.5% have one. 37.5% do not.
One sentence saying what the company does, for whom and where.
How to check: Open the home page source (Ctrl+U or Cmd+Option+U) and search for name="description". On WordPress, the field is usually in the SEO plugin.
Example meta description
<meta name="description" content="Accounting for freelancers and small businesses in Braga. Tax returns and payroll processing.">All four points in one report.
The AEO audit checks these points on your domain and shows what is missing.
How to check: Enter your domain in the audit and compare the result with the figures in this study.
| Group 1 | Group 2 | |
|---|---|---|
| Domains drawn | 10,000 | 3,000 |
| Active sites analysed | 8,347 | 1,813 |
| Inactive or empty (suspended account, under construction, default page) | 823 | 222 |
| Gambling and spam | 142 | 91 |
| Unreachable or without HTML | 612 | 775 |
| Excluded because robots.txt blocks the bot | 76 | 99 |
A text file at the domain root (/llms.txt) with a summary of the site and links to the main pages, designed for language models. It is a proposal, not a standard. The study records whether the file exists and returns text, and does not assess its content.
robots.txt controls access agent by agent, by name. Blocking one agent does not block the others: 88.5% of sites that block only block training agents and leave search and user-request agents open. The study measures what sites declare, not whether AI cites them.
A language model read the home page title and description and assigned one of 12 sectors. Estimated accuracy is about 90%, from a manual review of 60 sites. The 2,225 sites without enough information sit in a separate group.
No. Only totals and aggregated percentages are published. There are no site names or per-domain data.
Yes, citing the source. Aggregated results are in CSV at the bottom of the page, with a suggested citation.
Add "User-agent: RodrigoSilvanoResearchBot" followed by "Disallow: /" to your robots.txt, or write to info@rodrigosilvano.pt with the domain. Data for that domain is deleted.
Aggregated results are available as CSV: totals, sectors, agents, platforms and readiness index. No site names or per-domain data are published.
Download the aggregated data (CSV)
Suggested citation
Silvano, R. (2026). AEO Portugal study 2026. rodrigosilvano.pt/en/aeo-study-portugal
If your domain was read and you want it excluded, see the research bot page. Research bot.
