We're fortunate to live in this era of computers, the internet, and AI.
On the day a redesigned website goes live, sign-off usually goes like this: the owner opens the homepage and looks at the new hero visual; picks up a phone and scrolls through to check nothing is broken; submits the contact form to see whether the message arrives. The design is right, the site is faster, the brand finally looks like who it is now. The checklist gets ticked box by box, and the project is closed.
Every box on that list is really asking the same thing: does it look good to people?
Our last three pieces were about how a brand says clearly who it is. This one goes a step further: once the message is clear, does the website — the container — deliver it to the reader intact? And some of today's readers aren't people.
01Freshly redesigned, and it scored 59
In August, Business Next (數位時代) ran an article that opens with a quote: "I told them to run the test. I never expected it to score only 59."
Here's what happened. The author suggested that a brand owner run their website through isitagentready.com, a free checkup tool launched by Cloudflare. The brand had spent years building itself up, and its website had just been redesigned. The result: 59.
The score measures Agent Readiness — whether AI agents can find, enter, and read a website without trouble — out of 100. The article splits the score into four bands: 80 and above is the leading group; 60 to 79 means the basics are in place but key gaps remain; 40 to 59 means AI can get into the site but reads it inefficiently, and some content may not be correctly recognized; 39 and below is the high-risk zone. By the grading scale most people in Taiwan know, 59 is one point short of a pass.
They then spent an afternoon fixing the items one by one, rescanned, and got 83. The author also notes that this is a single site's case; with a different architecture, both the time it takes and the change in score would be different.
To be clear: 59 and 83 come from the case in Business Next's report, not from our own measurements. The report covers in detail how those 24 points were recovered; what we care about is the premise up front: the site had just been redesigned.
A website that had just used up its budget and just been signed off was one point short of a pass in the eyes of a different kind of reader. The design itself wasn't necessarily wrong. The problem was more likely that on sign-off day, nobody looked at it on that reader's behalf.
02AI isn't browsing your website. It's parsing it.
One line in that report is worth copying down: "AI agents don't browse your website. They parse it."
When a person opens a web page, the first thing they take in is the feel of the colors, the photos, and the typefaces, and only then do they look for the information they came for. AI agents have no "take it in first" stage. They rely on the site's structure, markup, and protocols: which pages they're allowed into, what this page is about, where the important information is written. The color schemes and photos a redesign pours the most thought into are almost no use to them; the markup and structure hardly anyone checks, they read line by line.
The report also cites an analysis Cloudflare published in April this year. Of 200,000 popular domains scanned, 78% had a robots.txt, but most of those were still written for traditional search engines; only 4% declared their AI usage preferences through Content Signals; and only 3.9% could serve a Markdown version of their content.
The author separates two things that often get lumped together. GEO (generative engine optimization) deals with the content layer: is it written well enough, precisely enough. Agent Readiness deals with the foundation layer: does the site let AI get in, find things, and read them all the way through. However good the content is, if it can't be read, its impact still takes a hit.
This is the other side of what we said in "SEO Isn't Dead. It Became the Ticket to Getting Cited by AI." That piece said technology lets AI find you, content lets AI trust you, and structure lets AI recognize who you are. A redesign often lavishes care on content and visuals and leaves technology and structure to luck.
03The sign-off checklist needs one more box
So the argument is simple: a redesign's sign-off checklist needs one more box — can machines read it?
This box doesn't need to be written up as a technical spec. For a business owner, three questions before launch are enough:
01 Is the door open? Does the site tell crawlers where they can go (robots.txt), and does it provide a list of every page on the site (a sitemap)?
02 Has the business card been handed over? Does every page spell out, in a format machines can read (structured data such as JSON-LD), who this is, what they do, and what this page is about?
03 Are the words actually on the page? Is the key information about your products and services there as soon as the page is delivered, rather than only appearing after JavaScript finishes running in the browser?
You can put all three directly to your engineers or your outsourced team. The answers should be "yes" or "no," not "probably."
This box also doesn't mean "open everything to every AI." One set of notes on how Cloudflare classifies traffic sorts AI traffic into three buckets: Search indexes content and later uses it to answer questions; Agent reads pages on the spot on someone's behalf to get information; Training scrapes content to train models. The three carry benefits and risks on very different scales and should be decided separately. What a brand's website usually needs most is the Search bucket: being indexed, being found. Agent depends on frequency — whether it crowds out real human visitors; whether your content should go into training data is another question altogether. What the sign-off box needs to confirm is that someone has actually made these decisions, rather than leaving everything to the defaults.
04Our own sign-off checklist
Talking about other people is easy. Here's ours.
In July this year, we moved all 35 pages of oakorc.com onto a new architecture and rebuilt the site. Since then we've set launch gates for our own website — twelve so far — and if any one of them fails, nothing goes live. Several of them check what machines read, which has nothing to do with how the screen looks.
Every one of those gates corresponds to a mistake that really happened. One is especially worth telling: the sitemap and the page list meant for AI were at one point both generated automatically by code from the same data, so comparing them against each other always came out identical. That check was permanently green as a result, even when both lists missed the same page. Only after we switched the baseline to "the pages actually deployed and in existence" could it catch that.
Before that, we had also audited our own website once and changed three things: the definition sentence on the homepage, duplicate entries in the sitemap, and the author structured data on our journal articles (see "How Long Until ChatGPT Cites a New Page? We Changed Three Things on Our Own Site"). None of the three were on screen. After the changes, the website looked exactly the same.
That's what makes this box tricky: done right, the human eye can't see the difference; skipped, the human eye can't see that either. So it's the easiest box to drop off the checklist.
05The score isn't the goal
Finally, we want to dial this back a little.
The isitagentready.com score is very useful for finding a starting point. The report itself flags a few things: the tool can't guarantee a brand will be cited or recommended by AI; the fix suggestions attached to each failed item are AI-generated and should be checked by a professional before you act on them; llms.txt isn't scored by default, and as of August the agentic commerce checks weren't scored yet either. Chase the score and it's easy to reshape your website to please one particular tool.
Our argument goes this far and no further: one more box on the redesign checklist, so that before launch someone has looked at the site once on behalf of the machine reader. Whatever score you get is a by-product of that box.
Two readers, one sign-off checklist
Before A website redesign was signed off on whether it looked good to people.
Now The same checklist needs one more question: can machines read it?
Design that can be trusted. Technology that can be understood. Applied to a website redesign, that line maps neatly onto the two readers: people need to be able to trust how you look, and machines need to be able to understand what you've written down.
If your website is about to be redesigned, or was redesigned recently, our website audit starts by laying out where things stand, and one of its axes is technical SEO: can crawlers get in, are your pages indexed. A full audit is NT$ 3,000, fully credited toward a redesign if you decide to go ahead. See clearly first, then decide where to spend.
Rooted in craft. Built for the new wild.