Is your factory's website readable by ChatGPT? A 10-point check

Only if the text is in the HTML the server sends, the AI search bots are allowed in, and the pages are indexed in Bing and Google. When GETMILK checked 20 Portuguese manufacturer websites on 28 September 2026, 5 home pages did not say what the company makes without JavaScript and only 5 showed product specifications as text.

A factory website is readable by ChatGPT when three things are true: what you make is written in the HTML the server sends, the search bots of OpenAI, Anthropic, Perplexity and Microsoft are allowed in, and the pages are indexed in Bing and Google. When GETMILK checked 20 Portuguese manufacturer websites on 28 September 2026, 5 of 20 home pages did not say in plain text what the company makes, only 5 of 20 showed product specifications as text, and 1 blocked every crawler except Google's. This matters because AI crawlers read raw HTML: in Vercel's measurement they did not run JavaScript [11]The rise of the AI crawlerVercel, 17 December 2024.

Key facts

What did we find on 20 Portuguese manufacturer websites?

The table is the 10-point check. The last column is what we found on the 20 sites (aggregate only; we do not name companies). Three checks need access to the site owner's tools, so we did not score them.

Checks 1 and 2 matter most. A procurement engineer who asks an assistant for "injection mould makers in Portugal with two-component experience" gets answers built from text the assistant could read. If your processes, materials and capacities sit in a slider, an image or a PDF, you are not in the pool it chooses from.

How did we choose and test the 20 sites?

We drew the sample from GETMILK's atlas of manufacturers: active companies in Portugal with a website, two per sector across metals, plastics, moulds, textiles, cork and wood, ceramics and glass, footwear, electrical, and transport, picked in a fixed pseudo-random order. We skipped candidates that were unreachable, foreign groups' sales offices, distributors or not manufacturers, and took the next one. For each site we downloaded three pages the way a simple browser would, on 28 September 2026: the home page, the instructions for bots (robots.txt) and one product or capability page. No logins, no forms, no scripts run. That is deliberate: it is roughly what an AI crawler sees [11]The rise of the AI crawlerVercel, 17 December 2024. Twenty sites is a small sample: read the counts as a signal, not a national statistic.

Why can't ChatGPT read some factory websites?

Most manufacturer websites were built for people using browsers. An AI crawler downloads the HTML and reads what is there. If the product range is drawn onto the page by a script after it loads, the crawler receives an empty frame. Vercel found that ChatGPT's and Claude's crawlers fetch JavaScript files (11.50% and 23.84% of their requests) but do not execute them [11]The rise of the AI crawlerVercel, 17 December 2024. In our sample, one home page contained a title, a meta description and an empty body: a person sees a full site, a crawler sees one word.

The more common failure is softer. Four home pages in our sample had text, but only slogans, collection names or an address. "Tradition, commitment and quality" does not tell anyone that you make natural cork stoppers or technical textiles. Write one plain sentence near the top: what you make, from which materials, for which industries, and where.

Which bots should robots.txt allow?

Separate search from training. OAI-SearchBot, Claude-SearchBot and PerplexityBot index pages so they can appear in answers; Anthropic says blocking its search bot "may reduce your site's visibility" [4]Does Anthropic crawl data from the web, and how can site owners block the crawler?Anthropic, 7 April 2026, accessed 28 September 2026, and Perplexity says PerplexityBot surfaces sites in its results and is not used for training [5]Perplexity CrawlersPerplexity, accessed 28 September 2026. GPTBot collects training data; blocking it does not take you out of ChatGPT search [1]Overview of OpenAI crawlersOpenAI, accessed 28 September 2026. For Google there is no separate AI crawler to allow: AI Overviews and AI Mode follow the normal Googlebot and snippet rules [6]AI features and your websiteGoogle Search Central, 10 December 2025.

The one site in our sample that failed this check allowed Googlebot and disallowed everyone else. That rule, often left by a web agency years ago, also blocks Bingbot, which cuts the company out of Microsoft Copilot and weakens it in ChatGPT, since ChatGPT search draws on Microsoft [2]Searching the web with ChatGPTOpenAI Help Center, updated August 2026, accessed 27 September 2026. If you want to keep pages out of ChatGPT, OpenAI's advice is to mark those pages "do not index", not to block the bot in robots.txt [3]Publishers and Developers FAQOpenAI Help Center, updated August 2026, accessed 27 September 2026.

Why does Bing matter if the buyer uses ChatGPT?

Because ChatGPT search does not rely only on its own crawler. OpenAI says it sometimes sends queries to third-party search providers and names Microsoft [2]Searching the web with ChatGPTOpenAI Help Center, updated August 2026, accessed 27 September 2026. Bing's AI Performance report, launched in February 2026, shows how often your pages are cited in Copilot and Bing's AI summaries and which grounding queries led there [9]Introducing AI Performance in Bing Webmaster Tools Public PreviewMicrosoft Bing Webmaster Blog, 10 February 2026. Its advice on content is plain: clear headings, tables and FAQ sections help AI systems reference information accurately [9]Introducing AI Performance in Bing Webmaster Tools Public PreviewMicrosoft Bing Webmaster Blog, 10 February 2026.

Do llms.txt and schema markup help?

Less than people selling them say. llms.txt is a proposal published on its own site [13]The /llms.txt filellmstxt.org, accessed 28 September 2026, and Google says you do not need AI text files or special markup to appear in its AI features [7]Optimizing your website for generative AI features on Google SearchGoogle Search Central, last updated 10 July 2026. Schema.org Organization markup is still worth having, because Google uses it for your logo and company details [8]Organization structured dataGoogle Search Central, last updated 8 September 2026, and 9 of our 20 sites had it. But do not expect it to bring citations: when Ahrefs tracked 1,885 pages that added schema, AI citations "barely moved" [12]We tracked 1,885 pages adding schema. AI citations barely movedAhrefs, 11 May 2026. Fix checks 1 to 3 first; they are the ones that decide whether you can be read at all.

How does Portugal compare with the rest of the EU?

The gap starts before AI. In 2025, 62.9% of Portuguese enterprises with 10 or more employees had a website, against 79.0% across the EU [14]Websites and functionalities by size class of enterprise (isoc_ciweb)Eurostat, 2025 data, updated 27 February 2026. Only 48.6% of Portuguese enterprises had a website that described their goods or services or listed prices, against 66.4% in the EU [14]Websites and functionalities by size class of enterprise (isoc_ciweb)Eurostat, 2025 data, updated 27 February 2026 [15]Digital economy and society statistics: enterprisesEurostat, data extracted January 2026. On the other hand, 31.4% of Portuguese enterprises had website content in at least two languages, slightly above the EU's 29.5% [14]Websites and functionalities by size class of enterprise (isoc_ciweb)Eurostat, 2025 data, updated 27 February 2026, which fits our finding that 19 of 20 sites had an English version.

AI adoption follows the same pattern: 11.5% of Portuguese enterprises used AI technologies in 2025, against 19.95% in the EU [16]Use of artificial intelligence in enterprisesEurostat, data extracted December 2025, edited 2 June 2026 [17]Artificial intelligence by size class of enterprise (isoc_eb_ai)Eurostat, 2025 data, updated 15 June 2026. The European Commission's 2026 country report says Portugal lags its peers in enterprise uptake of cloud and AI [18]Portugal's 2026 Digital Decade Country ReportEuropean Commission, 2026, accessed 28 September 2026. Your buyers in Germany or the Netherlands are further along, and they search accordingly.

Which hidden faults do modern sites often have?

Passing the crawler check is not the end. Four faults are common even on recently built sites: pages that first send an empty placeholder and only show the real text after a script runs; a language switch that forwards visitors in a way search engines do not treat as a proper redirect; sitemap entries that claim a new "last changed" date every time they are checked; and no notice to Bing through IndexNow when a page changes. Each has a simple fix: send the text in the first response, use a proper permanent redirect, give real dates, and send an IndexNow notice on every publish [10]IndexNow FAQIndexNow, accessed 28 September 2026.

Once buyers can read you, the next bottleneck is answering them: see how to respond to an RFQ in 24 hours. For how buyers look for suppliers in the first place, see how Portuguese manufacturers find foreign buyers.

What to do next

1. Run checks 1 to 3 today. Use "View page source" on your home page and one product page, and open your site's robots.txt. If you cannot find a sentence saying what you make, or robots.txt blocks all bots except Google, that is the first fix, and your current web agency can usually make it in a day.
2. Move your key specifications out of PDFs. Materials, sizes, tolerances, standards, machine capacities and certificates, as text on the product page. Then verify the site in Bing Webmaster Tools and Google Search Console.
3. If the template itself is the problem, rebuild the pages that matter. GETMILK builds manufacturer websites from €2,400 excluding VAT: up to 12 pages in English and Portuguese, capability or catalogue pages, and these ten checks done at handover (fixed price, 30% to start, 70% when it works). If your site passes checks 1 to 3 already, you probably do not need a rebuild; you need better product pages.