Research 8 min read

I Checked 75 SA Steel and Engineering Websites (2026 Research)

By ·

I checked the websites of every member of the Southern African Institute of Steel Construction (SAISC) on 6 October 2026: fabricators, engineers, suppliers and erectors. Of 101 members, 84 list a website and 75 loaded. The short answer is that AI tools and search engines can reach these sites, but most are slow on phones and give buyers little to go on. The median homepage took 9.1 seconds to show its main content in Google’s mobile test, 44% had a heading structure search engines find confusing, and 60% had no tap-to-call or WhatsApp link on the homepage.

This is the industry I build websites for, so I wanted numbers rather than impressions. No company is named here, and nothing in this post comes from any client’s audit.

Key Takeaways

  • Phones: only 6 of 72 homepages met Google’s “good” loading threshold of 2.5 seconds. The median was 9.1 seconds.
  • Search basics: 32% had no meta description, 36% had no structured data at all, and 44% did not have exactly one main heading.
  • Proof: 61% linked to projects or case studies from the homepage; 41% named a certification; 23% did neither.
  • Contact: 60% had neither a tap-to-call link nor a WhatsApp link on the homepage.
  • AI search: none blocked AI crawlers. Access is not the problem; what the crawlers find is.

How I Did the Study

The sample is the full public member list on saisc.co.za/members as it stood on 6 October 2026. I chose it because it is a complete, published list of firms in one industry, rather than a hand-picked set.

For each member I opened the SAISC listing to find the company’s website, then made one request each for the homepage, robots.txt and llms.txt, about a second apart, with a user agent that names DigiBannister. I then:

  • ran automated checks on the homepage HTML (titles, descriptions, headings, structured data, links, word count);
  • read robots.txt for rules aimed at the main AI crawlers;
  • ran one Google Lighthouse mobile performance test per site, which simulates a mid-range phone on a slow mobile connection. 72 of the 75 tests completed.
StepCount
SAISC members listed101
Members listing a website84
Homepages that loaded75
Lighthouse mobile tests completed72

Limits. These are homepages only, checked once, by script. A Lighthouse lab test is not the same as what real visitors experience, and a pattern check can miss a button built in an unusual way. The figures describe the group, not any one company.

Finding 1: Most Sites Are Slow on a Phone

Google treats a page as loading well when its largest piece of content appears within 2.5 seconds, and as poor beyond 4 seconds (Google Search Central: Core Web Vitals). Measured that way, very few of these homepages pass.

Phone test, 72 homepagesResult
Median time to main content9.1 s
Good (2.5 s or less)6 (8%)
Poor (over 4 s)57 (79%)
Over 10 s33 (46%)
Median download3.4 MB
Over 5 MB25 (35%)
Lighthouse score 90+9 (12%)

The pattern behind the slow sites is weight. A quarter of the homepages downloaded more than 5.7 MB. Images were the largest part of the download on 49 of 73 homepages, and on the median homepage they made up 42% of the bytes, against 29% for scripts. Of the five heaviest homepages, between 16 MB and 91 MB, three were mostly video and two mostly images. 65% of the sites run on WordPress.

This matters more for an engineering firm than it seems. A site manager or engineer checking you between meetings is on a phone, often on a site with patchy signal. If the page is still blank after five seconds, the next search result gets the call.

Finding 2: Search Engines Get Mixed Signals

These are the basics every search engine and AI assistant reads first. They are cheap to fix, and many sites miss them.

Check, 75 homepagesSites
No meta description24 (32%)
No main heading (H1)13 (17%)
Several main headings20 (27%)
No structured data27 (36%)
No business data35 (47%)
No https redirect11 (15%)
No mobile viewport tag6 (8%)
Copyright before 202510 of 38 (26%)

“No business data” means no structured data naming the business itself. “No https redirect” means the http:// address does not forward to https://.

Every site was served over HTTPS and every one had a page title, which is good. The gaps are in what tells a search engine what the page is about: a description, one clear main heading, and structured data naming the business, its location and what it does. An out-of-date copyright year is minor on its own, but to a buyer it suggests nobody is looking after the site.

Finding 3: Proof Is Hit and Miss

Procurement teams and engineers shortlist suppliers on evidence: projects like theirs, certifications, and enough detail to judge capability.

Check, 75 homepagesSites
Links to projects46 (61%)
Names a certification31 (41%)
Neither17 (23%)
Under 300 words20 (27%)

“Links to projects” counts links to projects, case studies or a portfolio. “Names a certification” counts ISO 9001, ISO 14001, ISO 45001, SABS, B-BBEE, ECSA or SAISC anywhere on the homepage.

A homepage with fewer than 300 words gives Google, ChatGPT and a buyer very little to work with. An AI assistant can only recommend a supplier for what its pages state plainly: what it makes, for which industries, to which standards, with named projects as proof.

Finding 4: Making Contact Takes Work

Check, 75 homepagesSites
Tap-to-call link29 (39%)
WhatsApp link4 (5%)
Neither of those45 (60%)
A form30 (40%)
A quote link8 (11%)

On a phone, a phone number that is not a link has to be copied by hand. Only four homepages offered a WhatsApp link at all. For a buyer ready to ask for a price, the easiest route is often missing.

Finding 5: AI Crawlers Are Welcome, but Find Little

None of the 75 sites blocked GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended or CCBot in robots.txt. Thirteen (17%) had an llms.txt file, and several of those were generated automatically by an SEO plugin rather than written on purpose. Google says it does not use llms.txt for Search or its AI features (Google’s AI optimization guide), so that file is not where the gains are.

So access is not what keeps these firms out of AI answers. The same guide says ranking in AI search is still SEO: pages must be indexed, useful and trustworthy. The gaps in Findings 1 to 4 (speed, clear signals, proof) are exactly what decides whether an assistant recommends a supplier.

What to Fix First

If you run an engineering, fabrication or industrial supply business, these five changes cover most of the gaps in this study:

  1. Get the homepage under 2.5 seconds on a phone. Resize and compress photographs, and remove page-builder sections that load scripts nobody uses.
  2. Give every page one main heading and a description that says what you do and where.
  3. Add structured data for your business (name, address, phone, services) that matches what the page says.
  4. Put proof on the homepage: two or three named projects and the certifications you hold.
  5. Make contact one tap: a tap-to-call number, a WhatsApp link and a clear “request a quote” button.

For the full list, see the SEO and GEO checklist for SA business websites. If you’d rather know exactly where your own site stands, I’ll check it for you: the free website check covers speed on a phone, the proof buyers look for, and whether Google and AI assistants can find you, and we go through it in 15 minutes. You can also read how I approach websites for engineering firms and SEO and AI search-ready websites.

Frequently Asked Questions

How fast are South African engineering company websites on a phone?

Slow. In this study the median homepage took 9.1 seconds to show its main content in Google’s mobile lab test, and only 6 of the 72 sites tested met Google’s “good” threshold of 2.5 seconds. 57 of the 72 (79%) were in Google’s “poor” range, slower than 4 seconds, and 33 took longer than 10 seconds. The usual cause is weight: the median homepage downloaded 3.4 MB and a third more than 5 MB. Images were the largest part of the download on two thirds of the homepages, and three of the five heaviest were mostly video.

Do engineering firms block ChatGPT and other AI crawlers?

No. None of the 75 websites in this study blocked any of the main AI crawlers in robots.txt, including GPTBot, ClaudeBot, PerplexityBot and Google-Extended. AI tools can read these sites. What holds most of them back is what those tools find when they get there: thin homepages, missing descriptions, no structured data about the business, and little proof such as named projects or certifications.

What do buyers look for on an engineering company website?

Proof and an easy way to make contact. Procurement teams and engineers look for past projects, certifications such as ISO 9001 or SABS, and a fast route to a quote or a person. In this study 61% of homepages linked to projects or case studies, 41% named a certification, and 23% did neither. Only 39% had a tap-to-call phone link and 5% a WhatsApp link, so 60% offered neither on the homepage.

How was this research done?

I took the public member list of the Southern African Institute of Steel Construction (SAISC) on 6 October 2026: 101 members, of which 84 list a website and 75 loaded. For each, I fetched the homepage, robots.txt and llms.txt once, ran automated checks on the HTML, and ran one Google Lighthouse mobile test (72 completed). No firm is named. The checks look at the homepage only and are automated, so treat the figures as a snapshot, not an audit of any one company.

Sources

Interested in what we’re building?

Get in touch to learn more about our projects or discuss how we can help your business.