We measured the baseline before we measured ourselves.
In July we drew 400 Malaysian electrical contractors at random from the public Suruhanjaya Tenaga register and went looking for their websites. Two search passes each, and one later correction after a company supplied a site we had missed.
Eight in ten had no website we could find at all. Of the 78 live sites we did verify, 32 were doing real commercial work — findable, clear about what the company does, showing evidence of past jobs, and contactable in one step. That is 8% of the register. The weakest of the five dimensions we scored, across every site, was project proof: 10.1 out of 20.
Publishing that study created an obligation. Having told 400 companies what a working website looks like, we could hardly decline to score our own by the same rubric — or to say what happened when we did.
What we actually did to our own site.
None of this is clever. It is the unglamorous list, and we are writing it out because the interesting thing about it is that so few sites in our own sample had done it.
One real question per page, answered in the first paragraph
Every article exists because someone asks that question before buying. The answer goes at the top, before the argument, so a reader who wants only the answer can leave with it.
Everything indexed, and checked that it is
A sitemap that lists real pages, and Search Console checked page by page rather than assumed. A page outside the index cannot be found by anyone, however well written.
Structured data that varies by page type
Not one blanket block everywhere. Articles carry Article and WebPage markup; structural pages carry breadcrumbs; service pages carry Service; the company identity and address sit on every page. Different pages are different things and the markup says so.
Dates only where a date is real
Our sitemap lists 68 pages. Forty-six carry a last-modified date and every one of those is an article; the twenty-two structural pages carry none. A date that moves when the content has not is a signal search engines discount, so we only claim one where something genuinely changed.
Old URLs redirected, not abandoned
An article we retired over a year ago still draws 38 impressions a month at position 8.8 — on a URL that no longer exists. It returns a permanent redirect to the page that replaced it rather than a 404, so the position keeps working. Most site rebuilds throw this away without noticing.
Prices published, not hidden behind a form
Every custody price is on one page, line by line, including what each tier does not include. It is the single thing that most reliably separates our pages from the ones we scored.
Original work rather than restated advice
The study above is ours: our sample, our scoring, our method published with its limits. It is the only thing on the site nobody else could have written.
What three weeks of data shows.
The figures below are from Google Search Console, covering 30 July to 18 August 2026 — 20 days with data on a site that had just been rebuilt. Search results take months to settle, so treat this as an early reading rather than a result.
| Measure | 30 Jul – 18 Aug 2026 |
|---|---|
| Impressions | 931 |
| Clicks | 19 |
| Click-through rate | 2.04% |
| Average position | 21.8 |
Nineteen clicks, published as nineteen clicks rather than as a percentage increase from a base of nothing, which is how a number that small is usually dressed up. It is worth reading against the baseline we measured: in a random sample of 400 companies in this trade, eight in ten had no website that could be found at all. A three-week-old site holding first-page positions on specific commercial questions is early, not impressive — but the distance between it and the sample is the point of the exercise.
The distribution is more useful than the total. A single Academy article — how to read a thermal image of an electrical panel — carries 117 of those impressions at an average position of 11.3. Our thermography cost article sits at position 7.9. The service page for thermography inspection sits at 21.9. Most of the rest of the site has no impressions at all yet.
One further thing happened that we did not plan for. Searching thermography malaysia now produces an AI-composed summary of the Malaysian market, and it names us among the companies doing industrial electrical scanning. We checked in incognito, repeated across several days. That is not a ranking and not an endorsement — these summaries are generated fresh each time and can name different companies on consecutive runs. What it does show is that the content sits in the retrieval set for a competitive commercial query.
One thing we assumed and got wrong is worth a line. Our site carries an llms.txt, a
brief written for AI systems, and when the citation appeared we took the two to be connected. They
are not: Google states in its own documentation that files of that kind neither help nor harm
visibility in Search. We set out what does
and does not affect this separately.
What the summary got wrong about us, and why.
That same summary described our pricing as tailored. That is the courteous way of saying we do not publish any. On the identical query it quoted a competitor’s actual figures, in ringgit bands from a small building to a large site, because that company had published them and we had not.
That was not a surprise to us. Our own competitor research had already identified the domain leading on price intent, scored four variants of the price query at 92 out of 100, and written the correction down: ringgit ranges, price drivers, inclusions. The gap was found by our own process before a customer found it, which is what the process is for — and it is now the first thing we check on any site we take on, because a company can hold the right answer internally and still publish a page that does not say it.
The lesson has nothing to do with AI. Anything summarising your company reads what is on the page. If the page avoids the question, the summary reports that you avoid the question — and then quotes whoever answered it.
Our Digital division publishes every price. Our Engineering division does not. We had not noticed the inconsistency until something outside the company described us back to ourselves.
What this is not.
It is not a client case study. AMKA Digital is in launch-cohort delivery and our policy is that client results are published only after they have been measured and approved by the client — so the absence of client case studies here is deliberate, not an oversight. When there is a measured client result, it will appear with its method attached, exactly as this one does.
It is not proof that the same work produces the same numbers on your site. It is one site, three weeks, in one market. We cannot attribute a single enquiry to any of it: traffic from AI summaries arrives in analytics as ordinary organic search and cannot be separated out.
What it is, is the method written down with its evidence attached — including a gap we found in our own work while writing it. If a company tells you it can improve your visibility, the reasonable question is what it did to its own site and what the numbers were. This is our answer, at 19 clicks, and we will publish the next one whatever it says.
Retrieved from https://amkatechnologies.com/insights/our-own-website-measured
Published 21 August 2026. AMKA Technologies Sdn Bhd, SSM 202301041763 (1535682-T).