<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Sergei Ivantsov</title>
    <description>The latest articles on DEV Community by Sergei Ivantsov (@sergei-swh).</description>
    <link>https://dev.to/sergei-swh</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4119596%2Fc72813a8-b266-489d-8ce0-d9847b1e81d5.jpg</url>
      <title>DEV Community: Sergei Ivantsov</title>
      <link>https://dev.to/sergei-swh</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/sergei-swh"/>
    <language>en</language>
    <item>
      <title>Hating "perfect" websites, I built my own — it's the little things</title>
      <dc:creator>Sergei Ivantsov</dc:creator>
      <pubDate>Sat, 14 Nov 2026 13:45:17 +0000</pubDate>
      <link>https://dev.to/sergei-swh/hating-perfect-websites-i-built-my-own-its-the-little-things-19da</link>
      <guid>https://dev.to/sergei-swh/hating-perfect-websites-i-built-my-own-its-the-little-things-19da</guid>
      <description>&lt;p&gt;I run a small site that sells four properties in North Cyprus: three flats and a villa, five languages, no agency. There's no hired developer either: I build it myself with an AI agent, and on my own project I don't need to justify a feature to anyone. "I want it to look nice" is enough.&lt;/p&gt;

&lt;p&gt;So here are five small things that ended up on the site for exactly that reason. None of them will sell a flat. I still like them more than most of the "important" work.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. The language buttons pulse on your first visit
&lt;/h2&gt;

&lt;p&gt;I'm tired of hunting for the language switcher on other people's sites. It's always somewhere: a tiny flag in the footer, a globe icon in a burger menu, a dropdown that says "EN" in grey on grey.&lt;/p&gt;

&lt;p&gt;On my site the switcher is a row of plain buttons at the top: RU, EN, TR, PL, DE. And the first time you land, they pulse three times, softly, and stop.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpdnnyl6t7t1ry2luo9km.gif" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fpdnnyl6t7t1ry2luo9km.gif" alt="Language buttons RU, EN, TR, PL, DE pulsing three times on first visit" width="640" height="134"&gt;&lt;/a&gt;&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;var&lt;/span&gt; &lt;span class="n"&gt;isFirstVisit&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;localStorage&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;getItem&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"lang"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="n"&gt;el&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;classList&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;toggle&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"attn"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;isFirstVisit&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="c1"&gt;// ...each button saves the choice on click:&lt;/span&gt;
&lt;span class="n"&gt;link&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;addEventListener&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"click"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="kd"&gt;function&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;localStorage&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;setItem&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"lang"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;code&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="c1"&gt;// ...and the hint switches itself off:&lt;/span&gt;
&lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;isFirstVisit&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nb"&gt;setTimeout&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="kd"&gt;function&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="n"&gt;el&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;classList&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;remove&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"attn"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;6000&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;





&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight css"&gt;&lt;code&gt;&lt;span class="nc"&gt;.lang-switch.attn&lt;/span&gt; &lt;span class="nt"&gt;a&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;animation&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;lang-attn-pulse&lt;/span&gt; &lt;span class="mf"&gt;1.6s&lt;/span&gt; &lt;span class="n"&gt;ease-in-out&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;

&lt;span class="k"&gt;@keyframes&lt;/span&gt; &lt;span class="nt"&gt;lang-attn-pulse&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="kt"&gt;%&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;100&lt;/span&gt;&lt;span class="kt"&gt;%&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;box-shadow&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="nf"&gt;rgba&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;47&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;111&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;94&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
  &lt;span class="mi"&gt;50&lt;/span&gt;&lt;span class="kt"&gt;%&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;box-shadow&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt; &lt;span class="mi"&gt;4px&lt;/span&gt; &lt;span class="nf"&gt;rgba&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;47&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;111&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;94&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mf"&gt;0.35&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;

&lt;span class="k"&gt;@media&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nt"&gt;prefers-reduced-motion&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="nt"&gt;reduce&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nc"&gt;.lang-switch.attn&lt;/span&gt; &lt;span class="n"&gt;a&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt; &lt;span class="nl"&gt;animation&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="n"&gt;none&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="p"&gt;}&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Three pulses of 1.6 seconds, then the class is gone. Once you've picked a language, the site remembers it and the buttons stay still. If your system asks for reduced motion, there's no animation at all. The current language is filled with the accent colour, so you can see where you are without reading.&lt;/p&gt;

&lt;p&gt;That's the whole feature. A developer would have asked me why. Claude got "make it pretty, please" — and that was enough.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. The WhatsApp message is always in Russian
&lt;/h2&gt;

&lt;p&gt;There is no contact form on the site. Each listing has a WhatsApp button, and the button opens a chat with a message already typed:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Здравствуйте! Интересует объявление La Isla (страница на польском): &lt;a href="https://girne-apartment-sale.com/" rel="noopener noreferrer"&gt;https://girne-apartment-sale.com/&lt;/a&gt;...&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;In English: "Hello! I'm interested in the La Isla listing (the page in Polish)", plus the link.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fncnhspfv0mfw9734bbjr.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fncnhspfv0mfw9734bbjr.png" alt="The WhatsApp draft as a buyer sees it, from the English Sea Pearl page" width="458" height="370"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Why Russian, if the visitor came from the Polish page? Because both owners read Russian. The draft is written for the person who receives it, not for the person who sends it. The note in brackets tells the owner which version of the page the buyer was reading, so they know what the buyer has already seen and which language to answer in.&lt;/p&gt;

&lt;p&gt;When we set this up, Claude objected. A Polish visitor gets a message they can't read, and that's odd. Fair point. The fix was one line under the button, in the visitor's own language:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Szkic wiadomości jest po rosyjsku, w języku właściciela, i podaje, z której wersji językowej piszesz. Możesz go zmienić przed wysłaniem.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;("The draft is in Russian, the owner's language, and says which language version you're writing from. You can edit it before sending.")&lt;/p&gt;

&lt;p&gt;And the owner will reply through a translator anyway.&lt;/p&gt;

&lt;p&gt;The link itself is not built in the browser. The generator that writes the static pages puts the finished &lt;code&gt;wa.me&lt;/code&gt; link into the HTML, and it refuses to build if a listing has no number or a language has no template:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;function&lt;/span&gt; &lt;span class="nf"&gt;whatsappBlock&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;id&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;lang&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;whatsapp&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`listings-data.js: listing ${id} has no whatsapp field`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="o"&gt;!&lt;/span&gt;&lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ui&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;whatsappMessage&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;throw&lt;/span&gt; &lt;span class="k"&gt;new&lt;/span&gt; &lt;span class="nb"&gt;Error&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="sb"&gt;`i18n.js: no ui.whatsappMessage for ${lang}`&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;t&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;ui&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;whatsappMessage&lt;/span&gt;
    &lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"{name}"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;data&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;name&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
    &lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"{url}"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pageUrl&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="kd"&gt;const&lt;/span&gt; &lt;span class="n"&gt;href&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="sb"&gt;`https://wa.me/${data.whatsapp}?text=${encodeURIComponent(text)}`&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt;
  &lt;span class="c1"&gt;// ...button + the hint, if this language has one&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The owners have different numbers, so there's no default number to fall back on. A listing with a missing number shouldn't quietly send the buyer to someone else.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. At night, the button asks first
&lt;/h2&gt;

&lt;p&gt;Press that WhatsApp button between 11 pm and 9 am Kyrenia time, and the chat doesn't open straight away. A window comes up first:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;It's nighttime in North Cyprus right now&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;It's currently 23:40 in Kyrenia (North Cyprus). The owner may already be asleep — he has two kids. Are you sure you want to message right now?&lt;/p&gt;

&lt;p&gt;[I'll wait until morning] [Message anyway]&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;A buyer in another country doesn't think about the time difference. Why would they? But WhatsApp is not email — a message at two in the morning buzzes a phone on someone's bedside table.&lt;/p&gt;

&lt;p&gt;From a conversion point of view this window is nonsense. It stands between the buyer and the owner at the exact moment the buyer is ready to write. Nobody is locked out, though: "Message anyway" carries the same link with the same Russian draft.&lt;/p&gt;

&lt;p&gt;I'd like to say I did it out of kindness. Mostly I did it because I could, and because nobody was there to cut it.&lt;/p&gt;

&lt;p&gt;And while writing this, I looked at the code and found a bug. The window counted the night by Istanbul time. Turkey doesn't change its clocks. North Cyprus does, together with the south. Since 25 October the north has been an hour behind Turkey, so the window was popping up at 10 pm and telling the buyer it was already 11. In the morning it let them through at 8 instead of 9.&lt;/p&gt;

&lt;p&gt;Claude wrote it that way back in September. And it didn't start from an empty folder: I gave it my own engine, WebCare, to build on, and that engine has had &lt;code&gt;Europe/Nicosia&lt;/code&gt; in its config all along. The agent took the multilingual part from the engine and picked the time zone on its own.&lt;/p&gt;

&lt;p&gt;The fix is one line:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;var&lt;/span&gt; &lt;span class="no"&gt;NICOSIA_TZ&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"Europe/Nicosia"&lt;/span&gt;&lt;span class="p"&gt;;&lt;/span&gt; &lt;span class="c1"&gt;// was "Europe/Istanbul"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A small digression. I've read and watched plenty of posts saying AI will now build you a great site from a single prompt. Maybe it will. Mine has been in the works since September, and I'm still finding things like this. Turns out there's a lot of finishing to do after the "great site" is ready.&lt;/p&gt;

&lt;p&gt;Maybe I'm just a nitpicker who sees problems that aren't there. This one was there, though — for almost three weeks the site was telling buyers the wrong time.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. The article cards are clickable all over
&lt;/h2&gt;

&lt;p&gt;At the bottom of the home page there are three cards under "Worth reading before you buy": the first questions about North Cyprus, the climate month by month, Girne or Iskele.&lt;/p&gt;

&lt;p&gt;You know the other way of doing it. "Important information about the climate can be found here", and only the word "here" is a link. You aim at four letters, miss, and select text instead.&lt;/p&gt;

&lt;p&gt;Here the whole card is the link:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;a&lt;/span&gt; &lt;span class="na"&gt;class&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"reading-card"&lt;/span&gt; &lt;span class="na"&gt;href&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"/climate"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;h3&amp;gt;&lt;/span&gt;Climate month by month: the heat is dry and the humidity comes in winter&lt;span class="nt"&gt;&amp;lt;/h3&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;p&amp;gt;&lt;/span&gt;Ten years of monthly figures for Iskele and Girne...&lt;span class="nt"&gt;&amp;lt;/p&amp;gt;&lt;/span&gt;
  &lt;span class="nt"&gt;&amp;lt;span&lt;/span&gt; &lt;span class="na"&gt;class&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"reading-more"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;Read it →&lt;span class="nt"&gt;&amp;lt;/span&amp;gt;&lt;/span&gt;
&lt;span class="nt"&gt;&amp;lt;/a&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;A block-level &lt;code&gt;&amp;lt;a&amp;gt;&lt;/code&gt; around a heading and a paragraph is valid HTML5. The heading, the description and the "Read it →" all lead to the same place. The bigger the target, the fewer the misses, especially on a phone.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Please don't translate my translations
&lt;/h2&gt;

&lt;p&gt;A friend told me he switched languages on the site and everything stayed in Russian. EN, PL, DE, still Russian.&lt;/p&gt;

&lt;p&gt;My guess is the browser. If Chrome has "always translate to Russian" switched on, it doesn't care that the page is a hand-made Polish version. It sees Polish and turns it into machine Russian. Five language versions, written and checked by a person — and the browser quietly throws four of them away.&lt;/p&gt;

&lt;p&gt;Auto-translation is a fine thing on a page that exists in one language only. I use it myself. On a page someone has already translated, it doesn't quite help. You get the gist, and you lose the rest. I'm a translator by training, so the rest is the part I care about.&lt;/p&gt;

&lt;p&gt;So I try to avoid it. One line in the head of every page:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;meta&lt;/span&gt; &lt;span class="na"&gt;name&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"google"&lt;/span&gt; &lt;span class="na"&gt;content&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="s"&gt;"notranslate"&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;"Try" is the right word. This is a request, not a lock. Chrome is supposed to listen, other browsers do what they like, and the visitor can still translate the page by hand if they want to.&lt;/p&gt;

&lt;p&gt;The line isn't typed into 48 files by hand. The same generator that writes the titles and canonical tags adds it, so a new page can't forget it.&lt;/p&gt;

&lt;p&gt;The site already has a translation switcher. It's those five pulsing buttons from part 1.&lt;/p&gt;




&lt;p&gt;I didn't do any of this for conversion — the night window even works against it. I did it because I could, and because on my own project nobody cuts it in a planning meeting.&lt;/p&gt;

&lt;p&gt;The site, if you want to try the buttons: &lt;a href="https://girne-apartment-sale.com/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article13" rel="noopener noreferrer"&gt;girne-apartment-sale.com&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;P.S. Yes, there's an em dash in the title, and a few more in the text. While we were writing this, Claude kept taking them out. The rule "em dash means AI" isn't made up: most people are too lazy to find the dash on a keyboard, and a machine never is. But it's a crude sign, not a verdict, and my AI seems to have taken it as one. I'm a translator by training. I know where a dash goes, and I put it there. One more little thing.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>ux</category>
      <category>javascript</category>
      <category>i18n</category>
    </item>
    <item>
      <title>SEO and copywriting are two jobs. Clients want four.</title>
      <dc:creator>Sergei Ivantsov</dc:creator>
      <pubDate>Mon, 26 Oct 2026 14:48:39 +0000</pubDate>
      <link>https://dev.to/sergei-swh/seo-and-copywriting-are-two-jobs-clients-want-four-43di</link>
      <guid>https://dev.to/sergei-swh/seo-and-copywriting-are-two-jobs-clients-want-four-43di</guid>
      <description>&lt;p&gt;I have done copy and search my whole adult life, and I work with an SEO/GEO agency. An agency can ship ninety articles a month for one client. My own property site got four articles in its first month, all on other platforms, and no bought links.&lt;/p&gt;

&lt;p&gt;The site went live on 3 September. As of 23 October, Google Search Console shows 4 clicks and 151 impressions over 28 days, and 42 pages in the index. All four clicks went to one listing. Small numbers in an almost empty niche, so they prove nothing about a crowded topic. But they came from a site where the order of work was strict, and I want to talk about that order.&lt;/p&gt;

&lt;p&gt;Put plainly: SEO is the more technical job. Copywriting is technical too, but it is more literary. They are different trades, and the market keeps mixing them up. Clients often expect two more on top: SMM and marketing.&lt;/p&gt;

&lt;p&gt;Yes, I do SEO and copy. But when I explain that to a client, it is simpler to say "I promote websites". In Russian that is one word, and it covers everything from SEO to ads. And right there the client's brain short-circuits: so you do marketing too, funnels and all?&lt;/p&gt;

&lt;h2&gt;
  
  
  First the frame, then the words
&lt;/h2&gt;

&lt;p&gt;In the first week the site had no articles at all. It had titles, meta descriptions, Open Graph, a canonical URL and hreflang for every language, clean addresses and a generator that builds every page's head from one data file. I wrote about that generator &lt;a href="https://dev.to/sergei-swh/my-ai-agent-copy-pasted-seo-tags-across-20-files-instead-of-generating-them-4lg2"&gt;here&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;The first article went up on 14 September, when the frame was already there. The site was a listing from day one, so there was always some text. But it already had structure: headings, body, footer, schema. The copy went into a ready frame, not the other way round.&lt;/p&gt;

&lt;p&gt;The first listing is the clearest case. I drew it as blocks before there was much to put in them: key facts, location, what's nearby, why it's good value, finish, the developer. Each block answers a question a buyer asks, in the order they ask it. Lists where people check facts, prose where something has to be argued. Later listings were built from that one. As the site grew, the template started to drift: one listing moved the value block to the top, another dropped it.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnrrjwj3r5eioy82pkbyn.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fnrrjwj3r5eioy82pkbyn.png" alt="The first listing: Key facts, Location, What's nearby, Why it's good value, each a block with its own heading" width="470" height="790"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Once the technical work is done, you can play with the copy. That is the order I believe in. And then I broke it myself.&lt;/p&gt;

&lt;h2&gt;
  
  
  The day the order broke
&lt;/h2&gt;

&lt;p&gt;On 24 October I wanted short answers to the questions people ask about North Cyprus first: visa, time zone, currency, airports, Kyrenia or Girne. I didn't phrase the task precisely enough. I meant a separate article. My AI assistant understood "a block of answers" and put it on the hub page that lists my articles — and there is the result: two rounds of fixes.&lt;/p&gt;

&lt;p&gt;The copywriting part of that day was real work, and none of it was SEO:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Topics came from demand, wording did not.&lt;/strong&gt; Google Trends gave the questions. But the heading "Kyrenia or Girne: what's the difference?" has almost no search volume as a question. People search the pair of names. We kept the question, because it is how a reader thinks.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Every fact was checked.&lt;/strong&gt; Distances to the airports we checked in Google Maps. For visas we only had secondary sources, so the note sends readers to the current rules on the ministry's website instead of listing exceptions.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;We cut more than we wrote.&lt;/strong&gt; Residence permits, the 90/180 rule, cost of living and a few more topics did not go in. One note went from six sentences to four after I said I was confused by it myself.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Words for the reader, not for the dictionary.&lt;/strong&gt; On one listing in Turkish, "ensuite" became "ebeveyn banyosu". I was against it at first. Then we counted 20 listings on a big local property site: "ebeveyn" seven times, "ensuite" zero.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The block went live. Then I asked one question: what goes into the title and the description of this page now?&lt;/p&gt;

&lt;p&gt;The answer: they still described the list of articles. Ten answers about visas and time zones were invisible in search. So the SEO had to be done as a separate pass the same evening: new title, new h1, new description, then a push to IndexNow.&lt;/p&gt;

&lt;p&gt;Next day I said the block was impossible to find. The only way in was through "All articles". So the answers moved to their own page, and the title and description moved with them — by hand. The hub got its old title back.&lt;/p&gt;

&lt;p&gt;What we had skipped on the first day was not "the SEO". It was one technical decision, the very first one — a separate address, or a block on somebody else's page. Everything else had to be bent around that decision twice.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the two jobs touch
&lt;/h2&gt;

&lt;p&gt;There is one place where the line gets blurry, and it is headings.&lt;/p&gt;

&lt;p&gt;The page title "North Cyprus: visa, time, currency and other short answers" is made of words, so it looks like copy. But we chose those words by what people search. "Ebeveyn banyosu" goes the other way: we picked it by counting, which is a technical method, and it works for the reader.&lt;/p&gt;

&lt;p&gt;So the jobs meet at the headings. Below them, they split. Trends chose the questions, not the answers. The title was fitted to queries, the text of the notes was not.&lt;/p&gt;

&lt;p&gt;Copy and marketing touch too, once. In that first listing, the "why it's good value" block compares the price per square metre with Spain's Costa Blanca. The block itself is copy: a selling argument in the text. Choosing Spain was a marketing decision. It says who the buyer is: someone weighing a flat by the sea in Spain. I made that decision once, when I drew the frame. On this site I gave myself carte blanche, so I could make that call myself. For a client it would be somebody else's job, and somebody else's invoice. No funnels, no ads, no quizzes, no landing pages.&lt;/p&gt;

&lt;p&gt;So the trades overlap and lie in the same plane. But they are clearly not the same thing.&lt;/p&gt;

&lt;h2&gt;
  
  
  The generator does not decide
&lt;/h2&gt;

&lt;p&gt;All of this happened on a site with a working SEO generator. It keeps titles, descriptions, hreflang and markup consistent across five languages. After the block went live, it built the page faithfully, with the old description, because nobody had changed the data file.&lt;/p&gt;

&lt;p&gt;The generator takes away the routine. It does not take away the decision about what this page is for in search. A person still makes that one. And in my data file titles belong to addresses, not to texts. When the text moves, the SEO does not follow it on its own.&lt;/p&gt;

&lt;h2&gt;
  
  
  The client wants four jobs in one
&lt;/h2&gt;

&lt;p&gt;Back to that short circuit. Building funnels is another profession, and it is paid separately. What clients often want is four jobs for the price of one. That is normal: any business wants to spend less and get more. So I do believe that we professionals should act as a business as well, and not take on extra work for the same money. Which means I have to know where my own line is, and say it before the work starts.&lt;/p&gt;

&lt;p&gt;I can combine SEO and copy. I am no marketer, and I am not an SMM person at all. Some of the agency's clients even ask about Threads. Social media is not in the contract. But contracts are rarely read to the end, and to the client, if you do the website, you surely set up their social accounts too. Figuratively speaking: where exactly would I put JSON-LD in an Instagram post?&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fasqazmhkenuqkknjstt7.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fasqazmhkenuqkknjstt7.png" alt="Comic: a client asks for Instagram and Threads too, and the SEO guy tries to push a JSON-LD block into a phone" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;This October a network of clinics, one of the agency's clients, ordered five opinion columns for a national business outlet. The outlet has hard rules. A headline is a full sentence with a verb, 73 characters at most. The format is expert opinion, not instructions: an earlier batch of columns was rejected for exactly that. No products, no prices, no calls to action.&lt;/p&gt;

&lt;p&gt;The topics came from the client's head of SMM. I cut the headlines to the limit and turned them into opinions, mostly "Why…" sentences. The client accepted the first one and sent back their own versions of the other four. One of them, in English: "Boosting clinic profits through a personal brand on social media. A step-by-step plan." 78 characters in the original, and a step-by-step plan is the one thing this format does not take.&lt;/p&gt;

&lt;p&gt;We found a compromise. We took their wording, fixed the grammar and cut the "step-by-step plan". The columns were then written to those headlines. I told the client up front that the outlet has its own limits and moderation, so the final version may still change. So far, silence.&lt;/p&gt;

&lt;p&gt;Two weeks earlier, the same side had asked for more clickbait headlines on a blogging platform, because the articles felt "too standard, too educational". That is feed logic, and in a feed it may well work. A feed pays for the click. Search and a business outlet pay for the match between the headline and the text. That is why I need a good text, and a clickbait headline works against me. Correct me if I'm wrong, but an expert, almost academic tone and a clickbait headline are a bit at odds.&lt;/p&gt;

&lt;p&gt;To the client a headline is a headline. But a headline for Instagram and a headline for a business outlet are written for different places, and this one went through the person who works for the feed. The absurdity is in the setup, not in the person.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the client should know
&lt;/h2&gt;

&lt;p&gt;SEO is not done once for the whole site. Every new section needs its own pass: title, h1, description, links, a request to recrawl. So does every move. If that pass is not in the plan and in the budget, the new text goes into search under an old description.&lt;/p&gt;

&lt;p&gt;When I say two jobs, I don't mean two people. That day it was me and an AI assistant in one session. I mean two pieces of work, each with its own scope and its own price. Ordering a text does not include SEO. And the other way round.&lt;/p&gt;

&lt;p&gt;Even on my own site, with carte blanche, I knew where my line was. I can make things look good and neat. But I am no marketer and no SMM person, and I have no wish to sit in Instagram or Facebook, whatever tools they offer. I genuinely respect good SMM people, like the teams behind Aviasales and Vizit (a Russian flight-search service and a Russian condom brand, both known for their social media). I can't name a third. I just can't picture myself racking my brain over posts, let alone answering comments there. The numbers in Search Console and an article now and then on a platform like this one are enough for me.&lt;/p&gt;

&lt;p&gt;The page from this story: &lt;a href="https://girne-apartment-sale.com/north-cyprus-first-questions?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article09" rel="noopener noreferrer"&gt;girne-apartment-sale.com/north-cyprus-first-questions&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The standalone, genericized version of the generator, if you want the code rather than the description: &lt;a href="https://github.com/sergei-swh/seo-generator" rel="noopener noreferrer"&gt;github.com/sergei-swh/seo-generator&lt;/a&gt;&lt;/p&gt;

</description>
      <category>seo</category>
      <category>webdev</category>
      <category>writing</category>
      <category>ai</category>
    </item>
    <item>
      <title>What AI crawlers are blind to: the text your JavaScript renders</title>
      <dc:creator>Sergei Ivantsov</dc:creator>
      <pubDate>Wed, 14 Oct 2026 13:47:45 +0000</pubDate>
      <link>https://dev.to/sergei-swh/what-ai-crawlers-are-blind-to-the-text-your-javascript-renders-4oen</link>
      <guid>https://dev.to/sergei-swh/what-ai-crawlers-are-blind-to-the-text-your-javascript-renders-4oen</guid>
      <description>&lt;p&gt;I run a small property site. Four languages, plain HTML, no framework, no npm dependencies, hosted on Cloudflare Pages. I have spent my working life on copy and search, so I did the usual things: structured data, hreflang, a sitemap, a generator that keeps every page's head in sync from one file.&lt;/p&gt;

&lt;p&gt;Then I ran one command against my own listing page:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;curl &lt;span class="nt"&gt;-s&lt;/span&gt; &lt;span class="nt"&gt;-A&lt;/span&gt; &lt;span class="s1"&gt;'GPTBot/1.0'&lt;/span&gt; https://example.com/my-listing | &lt;span class="se"&gt;\&lt;/span&gt;
  python3 &lt;span class="nt"&gt;-c&lt;/span&gt; &lt;span class="s2"&gt;"import sys,re,html; b=sys.stdin.read(); b=b[b.index('&amp;lt;body'):]; &lt;/span&gt;&lt;span class="se"&gt;\&lt;/span&gt;&lt;span class="s2"&gt;
  b=re.sub(r'&amp;lt;script.*?&amp;lt;/script&amp;gt;','',b,flags=re.S); &lt;/span&gt;&lt;span class="se"&gt;\&lt;/span&gt;&lt;span class="s2"&gt;
  print(len(re.sub(r'&lt;/span&gt;&lt;span class="se"&gt;\s&lt;/span&gt;&lt;span class="s2"&gt;+',' ',html.unescape(re.sub(r'&amp;lt;[^&amp;gt;]+&amp;gt;',' ',b))).strip()))"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;It counts the visible text in &lt;code&gt;&amp;lt;body&amp;gt;&lt;/code&gt; after removing scripts. That is roughly what a crawler gets if it does not execute JavaScript.&lt;/p&gt;

&lt;p&gt;The answer for my listing page was &lt;strong&gt;23 characters&lt;/strong&gt;. For my homepage, 17.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the text had gone
&lt;/h2&gt;

&lt;p&gt;Nowhere surprising, once you look. The site is multilingual, and the copy for every language lives in &lt;code&gt;assets/js/i18n.js&lt;/code&gt;. On load, &lt;code&gt;app.js&lt;/code&gt; picks the language and fills the containers:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight javascript"&gt;&lt;code&gt;&lt;span class="kd"&gt;var&lt;/span&gt; &lt;span class="nx"&gt;descEl&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nb"&gt;document&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;querySelector&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="s2"&gt;[data-listing-description]&lt;/span&gt;&lt;span class="dl"&gt;"&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;span class="k"&gt;if &lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;descEl&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="nx"&gt;descEl&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;innerHTML&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nx"&gt;l&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nx"&gt;description&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="nf"&gt;climateNoteFor&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nx"&gt;id&lt;/span&gt;&lt;span class="p"&gt;);&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That is a perfectly ordinary pattern. It also means the HTML I ship contains the layout and nothing to read.&lt;/p&gt;

&lt;p&gt;Googlebot renders JavaScript, so Google was fine. The question is everyone else.&lt;/p&gt;

&lt;h2&gt;
  
  
  What the crawler did get
&lt;/h2&gt;

&lt;p&gt;Not nothing. The head carried 2,891 characters of JSON-LD: name, price, number of rooms, floor area, address, coordinates, amenities. A machine could reconstruct a tidy product card from that.&lt;/p&gt;

&lt;p&gt;And that is exactly what made it worse.&lt;/p&gt;

&lt;p&gt;My site exists for one reason: it is a private sale, direct from the owner, and every listing states plainly what the title deed situation is — the kind of sentence an agency leaves out. That sentence lives in the prose. So a crawler without JavaScript got the price and the square metres, identical to any agency listing, and none of the disclosure the site was built around. The one thing that made the page different was the one thing it could not read.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why no dashboard told me
&lt;/h2&gt;

&lt;p&gt;Analytics did not show this, and could not. My analytics runs on JavaScript. It counts visitors whose browsers execute scripts, which by definition excludes every client that does not. The blind spot is the measurement instrument, not the site.&lt;/p&gt;

&lt;p&gt;What did show it was Cloudflare's AI crawler log. In 24 hours up to 15 September 2026 my domain took 43 requests from AI crawlers: Applebot 20, Googlebot 14, Meta-ExternalAgent 8, GPTBot 1. They fetched pages, the sitemap, images, the CSS, both JS files, and the favicon I had published an hour earlier.&lt;/p&gt;

&lt;p&gt;Worth noting, since it is the fashionable question: &lt;code&gt;llms.txt&lt;/code&gt; sits in my site root, and across 23 distinct crawled paths in that window it was requested exactly zero times. Cloudflare's own readiness checklist did not mention the file either. On that day its items were robots.txt, a sitemap, content signals, markdown versions of pages, and WebMCP.&lt;/p&gt;

&lt;h2&gt;
  
  
  The fix, and why it was cheap
&lt;/h2&gt;

&lt;p&gt;I already had a generator for the head. It takes one data file and writes it into every page between markers:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="c"&gt;&amp;lt;!-- SEO:START (generated by build/generate-seo.js — do not hand-edit between markers) --&amp;gt;&lt;/span&gt;
...
&lt;span class="c"&gt;&amp;lt;!-- SEO:END --&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;a href="https://dev.to/sergei-swh/four-layers-of-defense-for-a-multilingual-site-26b8"&gt;previous post&lt;/a&gt; is about it. The same idea works for the body. I put markers inside the containers the script already fills:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight html"&gt;&lt;code&gt;&lt;span class="nt"&gt;&amp;lt;div&lt;/span&gt; &lt;span class="na"&gt;class=&lt;/span&gt;&lt;span class="s"&gt;"listing-description"&lt;/span&gt; &lt;span class="na"&gt;data-listing-description&lt;/span&gt;&lt;span class="nt"&gt;&amp;gt;&lt;/span&gt;&lt;span class="c"&gt;&amp;lt;!-- BODY:listing-description --&amp;gt;&amp;lt;!-- /BODY:listing-description --&amp;gt;&lt;/span&gt;&lt;span class="nt"&gt;&amp;lt;/div&amp;gt;&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;and taught the generator to fill them from &lt;code&gt;i18n.js&lt;/code&gt; and &lt;code&gt;listings-data.js&lt;/code&gt; — the same files the browser uses. That detail is the whole point: the static text cannot drift from what the script renders, because there is one source and one build step. &lt;code&gt;app.js&lt;/code&gt; still redraws those containers on load, so the language switcher behaves exactly as before.&lt;/p&gt;

&lt;p&gt;Results, same command as above:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Page&lt;/th&gt;
&lt;th&gt;Before&lt;/th&gt;
&lt;th&gt;After (15 September 2026)&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Homepage&lt;/td&gt;
&lt;td&gt;17&lt;/td&gt;
&lt;td&gt;2,648&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Listing&lt;/td&gt;
&lt;td&gt;23&lt;/td&gt;
&lt;td&gt;3,127&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;About&lt;/td&gt;
&lt;td&gt;17&lt;/td&gt;
&lt;td&gt;995&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Total work: one script and a pass that inserted markers. No framework, no SSR runtime, no rewrite.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I would take from this
&lt;/h2&gt;

&lt;p&gt;A generator earns its keep twice. The first time when it saves you from editing twenty files by hand. The second time when a new requirement lands and the data is already in one place, so the fix is a function rather than a project.&lt;/p&gt;

&lt;p&gt;And the general version, which is older than any of this: check what the machine receives, not what your browser draws. The page looked finished in every browser I opened. It was empty to anything that did not run my JavaScript, and it stayed that way for weeks because every tool I used to check it ran JavaScript too.&lt;/p&gt;

&lt;p&gt;Run the command above on your own site. It takes ten seconds. If your stack renders on the server you will see a number in the thousands and can stop reading. If you see two digits, you now know something your analytics will never tell you.&lt;/p&gt;

&lt;p&gt;The site: &lt;a href="https://girne-apartment-sale.com/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article04" rel="noopener noreferrer"&gt;https://girne-apartment-sale.com/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article04&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Standalone, genericized version of the generator, if you want the code rather than the description: &lt;a href="https://github.com/sergei-swh/seo-generator" rel="noopener noreferrer"&gt;https://github.com/sergei-swh/seo-generator&lt;/a&gt;&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>seo</category>
      <category>javascript</category>
      <category>ai</category>
    </item>
    <item>
      <title>Four layers of defense for a multilingual site</title>
      <dc:creator>Sergei Ivantsov</dc:creator>
      <pubDate>Wed, 23 Sep 2026 13:25:25 +0000</pubDate>
      <link>https://dev.to/sergei-swh/four-layers-of-defense-for-a-multilingual-site-26b8</link>
      <guid>https://dev.to/sergei-swh/four-layers-of-defense-for-a-multilingual-site-26b8</guid>
      <description>&lt;p&gt;I keep building multilingual sites. Restaurants, shops, now apartment listings on Northern Cyprus. People assume the hard part is translation. It isn't. The hard part is keeping four languages in sync after you ship, when nobody is looking.&lt;/p&gt;

&lt;p&gt;Here is what actually holds a multilingual site together. Not one trick. Four layers, and each one catches a different kind of failure.&lt;/p&gt;

&lt;h2&gt;
  
  
  Layer 1: the generator
&lt;/h2&gt;

&lt;p&gt;Text used to live copy-pasted across twenty HTML files: five page types, four languages. On September 7 that broke. The title, the &lt;code&gt;og:title&lt;/code&gt;, and the JSON-LD &lt;code&gt;name&lt;/code&gt; for one listing were out of sync on all four languages, because someone (me, with an AI agent) copied the string by hand one too many times. The Polish version used a different phrase entirely.&lt;/p&gt;

&lt;p&gt;The fix wasn't "be more careful next time." It was &lt;code&gt;build/generate-seo.js&lt;/code&gt;: one data file, one generator, twenty files written from it. A &lt;code&gt;--check&lt;/code&gt; flag fails the build if a file drifted from the source instead of silently overwriting it. This is the layer that makes structural drift physically impossible, not just unlikely.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight console"&gt;&lt;code&gt;&lt;span class="gp"&gt;$&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;node build/generate-seo.js &lt;span class="nt"&gt;--check&lt;/span&gt;
&lt;span class="go"&gt;SEO head blocks are up to date.
Body blocks are up to date.
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;That's the whole contract: the generator either confirms the twenty files still match the source, or it fails the build. No middle state where a file is quietly out of sync.&lt;/p&gt;

&lt;h2&gt;
  
  
  Layer 2: the linter
&lt;/h2&gt;

&lt;p&gt;The generator can't see style. It writes what the data file says, correctly, forever. If the data file itself is inconsistent, the generator faithfully ships that inconsistency four times.&lt;/p&gt;

&lt;p&gt;So I wrote &lt;code&gt;build/lint-text.js&lt;/code&gt;. It caught British spelling mixed with American in my own English copy. It caught the same empty hedge word duplicated across languages by the same translation habit: "actually" in English, "на самом деле" in Russian, "gerçekte" in Turkish, "naprawdę" in Polish — one language picks up a verbal tic, and if you don't check, it rides along into every translation.&lt;/p&gt;

&lt;p&gt;Dash frequency needed per-language thresholds, not one global number. English prose runs light on em dashes. Russian carries more. Turkish and Polish sit somewhere in between. A linter tuned for one language would either flag good Russian prose as robotic or miss actual robotic English.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight console"&gt;&lt;code&gt;&lt;span class="gp"&gt;$&lt;/span&gt;&lt;span class="w"&gt; &lt;/span&gt;node build/lint-text.js
&lt;span class="c"&gt;...
&lt;/span&gt;&lt;span class="go"&gt;· listing reciaoglu    en  empty evaluative adjective   2
&lt;/span&gt;&lt;span class="c"&gt;...
&lt;/span&gt;&lt;span class="go"&gt;Notes: 0 warnings, 1 flag.
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;(trimmed; the real run checks every listing in all four languages and prints a full table even when it finds nothing — that flag above was a real one, not staged for this article.)&lt;/p&gt;

&lt;h2&gt;
  
  
  Layer 3: the trained eye
&lt;/h2&gt;

&lt;p&gt;Neither tool catches meaning. A flyer said "~50 km to both airports." True for Ercan. Off by 20 km for Larnaca. Grammatically fine, technically consistent, and wrong, because nobody checked the number against the actual reader driving to the actual airport. That's not a generator problem or a linter problem. That's a person who reads the sentence and asks "is this true for the person landing at Larnaca," and my background is translation, so noticing when a sentence says one thing in one language and something slightly different in another is close to the whole job.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight diff"&gt;&lt;code&gt;&lt;span class="gd"&gt;-           &amp;lt;li&amp;gt;About 50 km to Ercan and Larnaca airports&amp;lt;/li&amp;gt;
&lt;/span&gt;&lt;span class="gi"&gt;+           &amp;lt;li&amp;gt;About 50 minutes by car to Ercan airport&amp;lt;/li&amp;gt;
+           &amp;lt;li&amp;gt;About 70 minutes to Larnaca — the route crosses the checkpoint&amp;lt;/li&amp;gt;
&lt;/span&gt;&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;One commit, same change repeated across all four language blocks in the same file — because the flawed number wasn't a typo in one place, it was the same inherited assumption translated four times.&lt;/p&gt;

&lt;p&gt;The same instinct works before anything is written, not just after. Nobody asked for a disclaimer page or an "about" page on a private listing site. I added both on day two: the site is not an agency, takes no prepayment, and the deal is between the two sides. No checklist told me to. It's the same trained eye, pointed at trust instead of at a sentence — noticing what a reader needs to see before they'll believe the rest of the page.&lt;/p&gt;

&lt;h2&gt;
  
  
  Layer 4: going back and looking
&lt;/h2&gt;

&lt;p&gt;The three layers above run at build time. None of them catch a section that's simply missing from the template. &lt;code&gt;data-reading&lt;/code&gt;, the "worth reading before you buy" block linking to the climate article, was absent from three of four homepage templates entirely. Not empty. Not there. The generator can't fill a slot that doesn't exist, and no linter checks for missing HTML structure.&lt;/p&gt;

&lt;p&gt;I found it by doing the unglamorous thing: opening the live site periodically and reading it like a stranger would. Same pass caught the language button order drifting out of sync with the default language, and a footer note still describing two listings after two more had been added.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why all four, and not just the best one
&lt;/h2&gt;

&lt;p&gt;Each layer catches what the others structurally cannot. The generator kills copy-paste drift. The linter kills style drift the generator inherits from bad source data. The eye kills meaning-level errors no automated check understands. And periodic re-reading kills the class of bug where something simply isn't there to check.&lt;/p&gt;

&lt;p&gt;Skip any one layer and a specific kind of mistake gets through, quietly, in all four languages at once.&lt;/p&gt;

&lt;p&gt;This whole habit started with the same thought as the first post in this series: "let's just do it right." Turns out doing it right on a multilingual site means doing it right four times, once per language, every single time.&lt;/p&gt;

&lt;p&gt;The site is still here and growing: &lt;a href="https://girne-apartment-sale.com/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article07" rel="noopener noreferrer"&gt;girne-apartment-sale.com&lt;/a&gt;. The layer 1 generator is open source: &lt;a href="https://github.com/sergei-swh/seo-generator" rel="noopener noreferrer"&gt;seo-generator on GitHub&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>webdev</category>
      <category>javascript</category>
      <category>i18n</category>
      <category>seo</category>
    </item>
    <item>
      <title>My AI agent copy-pasted SEO tags across 20 files instead of generating them</title>
      <dc:creator>Sergei Ivantsov</dc:creator>
      <pubDate>Thu, 10 Sep 2026 16:01:55 +0000</pubDate>
      <link>https://dev.to/sergei-swh/my-ai-agent-copy-pasted-seo-tags-across-20-files-instead-of-generating-them-4lg2</link>
      <guid>https://dev.to/sergei-swh/my-ai-agent-copy-pasted-seo-tags-across-20-files-instead-of-generating-them-4lg2</guid>
      <description>&lt;p&gt;I wrote a small generator to stop an AI coding agent from hand-duplicating SEO metadata across a multilingual static site. It has a &lt;code&gt;--check&lt;/code&gt; flag that verifies the generated files still match their data source. The day after I shipped it, I found that when the generator silently corrupted all 20 pages, &lt;code&gt;--check&lt;/code&gt; reported it as "out of date" — the exact same message it prints when you legitimately edit the data file and haven't regenerated yet.&lt;/p&gt;

&lt;p&gt;That's the part worth writing down. A verification flag that can't distinguish "you changed something on purpose" from "the tool just wrote broken URLs into every page you have" isn't verifying much. Here's how I got there.&lt;/p&gt;

&lt;h2&gt;
  
  
  The site
&lt;/h2&gt;

&lt;p&gt;A free, direct-from-owner property listing site for North Cyprus — &lt;a href="https://girne-apartment-sale.com" rel="noopener noreferrer"&gt;girne-apartment-sale.com&lt;/a&gt;, no agency, no commission. Vanilla HTML/CSS/JS, zero npm dependencies, deployed to Cloudflare Pages. Four languages (EN/RU/TR/PL), each language a real indexable URL rather than a client-side toggle, so 5 page types × 4 languages = 20 HTML files. Written together with Claude Code.&lt;/p&gt;

&lt;p&gt;I have a professional translation background, so my bar for "done" here was a genuinely correct multilingual site, not a page that renders. That bar is the only reason the rest of this happened — nothing below was surfaced by the agent on its own.&lt;/p&gt;

&lt;h2&gt;
  
  
  The first bug: no single source of truth
&lt;/h2&gt;

&lt;p&gt;The agent put &lt;code&gt;&amp;lt;title&amp;gt;&lt;/code&gt;, &lt;code&gt;og:title&lt;/code&gt;, and the JSON-LD &lt;code&gt;name&lt;/code&gt; into each of those 20 files by hand, as literal strings, rather than deriving them from one place. That works right up until you update 19 files and miss one.&lt;/p&gt;

&lt;p&gt;I caught it by re-running the site against an old GEO/AEO audit checklist I had lying around, written for a completely unrelated niche (health, not real estate) and used purely as a generic methodology pass — hreflang, canonicals, structured data. The second listing's Polish version had different wording in its title and description than the other three languages. Not a typo: drift. Individually harmless, collectively the exact failure a site that advertises itself as multilingual cannot afford.&lt;/p&gt;

&lt;p&gt;Worth noting what &lt;em&gt;didn't&lt;/em&gt; happen: reference material of mine describing the practices I wanted — zero dependencies, one source of truth for data, structured data over keyword stuffing — was sitting in the working environment the whole time. None of it got applied unprompted. Every item had to be asked for explicitly, one at a time. Being available in context is not the same as being pulled forward.&lt;/p&gt;

&lt;h2&gt;
  
  
  The fix, and why it isn't Hugo
&lt;/h2&gt;

&lt;p&gt;The obvious objection first: this is a solved problem, and Eleventy, Hugo, Jinja, or any templating layer solves it out of the box. True. I didn't use one, deliberately.&lt;/p&gt;

&lt;p&gt;Adopting an SSG here means moving the entire page body — layout, listing copy, gallery markup, four languages of it — into a template system, to fix a problem that lives in roughly 15 lines of &lt;code&gt;&amp;lt;head&amp;gt;&lt;/code&gt;. The migration is the whole site; the bug is the header. So instead: &lt;code&gt;build/seo-data.js&lt;/code&gt; holds title, meta description, Open Graph, hreflang, and JSON-LD for every page in one place, and &lt;code&gt;build/generate-seo.js&lt;/code&gt; injects it into each file between &lt;code&gt;&amp;lt;!-- SEO:START --&amp;gt;&lt;/code&gt; / &lt;code&gt;&amp;lt;!-- SEO:END --&amp;gt;&lt;/code&gt; markers. Everything outside those markers is never touched — the HTML stays hand-written HTML, editable without knowing the generator exists. Plain Node, no dependencies. &lt;code&gt;--check&lt;/code&gt; verifies without writing, run before deploy.&lt;/p&gt;

&lt;p&gt;That closed the drift class structurally. There's no second copy of the string left anywhere for me or the agent to forget.&lt;/p&gt;

&lt;p&gt;Standalone, genericized version with an example data file, if you want the code rather than the description: &lt;a href="https://github.com/sergei-swh/seo-generator" rel="noopener noreferrer"&gt;https://github.com/sergei-swh/seo-generator&lt;/a&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  The second bug: the tool built to stop looking
&lt;/h2&gt;

&lt;p&gt;Then I tested adding a fifth language — German — to see what the generator did with it.&lt;/p&gt;

&lt;p&gt;Adding &lt;code&gt;"de"&lt;/code&gt; to the language list without the matching &lt;code&gt;paths&lt;/code&gt; / &lt;code&gt;perLang&lt;/code&gt; / &lt;code&gt;homeLabel&lt;/code&gt; data didn't error. The hreflang builder loops over every configured language for every page, found no path for German, and wrote &lt;code&gt;href="https://girne-apartment-sale.comundefined"&lt;/code&gt; into all 20 existing pages. Pages that were correct a minute earlier. And &lt;code&gt;--check&lt;/code&gt; dutifully reported: out of date. Same string as any ordinary pending edit.&lt;/p&gt;

&lt;p&gt;The fix was a &lt;code&gt;validateData()&lt;/code&gt; pass that runs before anything is written and fails loudly with an explicit list of what's missing. But the lesson generalizes past this one script: if your verification step reports "different from expected" without a notion of &lt;em&gt;how&lt;/em&gt; different, it is a diff, not a check. Mine was a diff pretending to be a check from the moment I wrote it, and I only found out because I went poking at a tool I had specifically built so I could stop poking.&lt;/p&gt;

&lt;h2&gt;
  
  
  Limits
&lt;/h2&gt;

&lt;p&gt;I didn't run this at max capability — Sonnet 5, reasoning effort medium, not higher — so none of this is a claim about what these agents do at their best, or in general. It's what happened in this setup, starting from an existing foundation rather than a cold start.&lt;/p&gt;

&lt;p&gt;What I'd take from it: an AI agent will reliably produce something that works, and it optimizes hardest for the layer you can see. The failure modes live one layer down, in duplication it introduces because duplication is locally cheaper than structure, and they surface as drift rather than as errors. Nothing about that is exotic — it's the same failure mode as any copy-paste — but it arrives faster and across more files than a human would manage, and it doesn't get flagged, because from the inside every individual file looks fine.&lt;/p&gt;

&lt;p&gt;The site is still here and growing: &lt;a href="https://girne-apartment-sale.com/?utm_source=devto&amp;amp;utm_medium=article&amp;amp;utm_campaign=article02" rel="noopener noreferrer"&gt;girne-apartment-sale.com&lt;/a&gt;. The generator from this post is open source: &lt;a href="https://github.com/sergei-swh/seo-generator" rel="noopener noreferrer"&gt;seo-generator on GitHub&lt;/a&gt;.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>webdev</category>
      <category>seo</category>
      <category>javascript</category>
    </item>
  </channel>
</rss>
