murattunalı.
Reading

MURATTUNALI. / JOURNAL — NOTEBOOK · ACTIVE FROM DAY ONE 23 ENTRIES 03 LANGUAGES LAST BUILD 2026-11-17 ISTANBUL --:--:--

Journal.

Short, evidence-backed writing on visibility in the GEO era.

SCHEMA: BLOG + PERSON · RSS · LLMS.TXT

Every claim has a measurement behind it, and every article names its method.

Measurement · Certification

We measured Turkish manufacturers' certificate pages

None of the 34 certificate pages that carried certificates gave a certificate’s expiry date in the page text, and none linked the buyer to a place where the certificate could be verified. On November 10, 2026, we measured 77 websites of exporting manufacturers in four provinces: we found their certificate pages, read them and coded each one with six questions.

11 November 2026 · 7′

When a new post is published, it enters the feed, the sitemap and the llms.txt file read by model crawlers at the same time — all three are generated from the same list, so they cannot diverge.

02 — CONTRACT

Every entry listed on this page was written under the same contract.
01

What this journal is for

We keep this Journal as a working notebook. We write each post to record a decision we actually made in this repository and the measurement behind it. That is why most posts open with a number and go on to explain how that number was obtained.

In practice this means: when a post is proven wrong, we correct it and record the correction itself. For example, the post describing the verification suite’s test count goes stale every time a new test is added to the suite, and a test catches it. The suite compares the number in the published HTML with its own real total; the moment they diverge it turns red, and updating the post becomes mandatory.

02

How it differs from the guides

The site has two separate content families, and their boundaries are clear so they are not confused. The posts describe what happens in this project: what we tried, what we measured, what we rejected. The guides, by contrast, teach a topic from start to finish and are not tied to a single project.

We protect the separation with code. Every guide and glossary entry belongs to a cluster, and the verification suite does not allow the same term to be defined on two different pages; that is how we prevent content that cannibalizes itself. You can enter the guides through the accessibility, answer engine optimization, performance and multilingual architecture pillars.

03

How it is written

The first sentence of every post has to be true without context. The reason is answer engines: the unit they quote is the paragraph, not the page, and a paragraph that cannot be understood without reading what came before cannot be quoted. An opening that starts with “as described above” therefore turns red in the verification suite.

The second rule is citing sources. If we can tie a claim to a primary source, we do: W3C, Google Search Central, web.dev, MDN. If we cannot, we either back it with a measurement or do not write it. We guard the source strip at the bottom of the page with a test: every post must link to at least two different authority domains. We explain how quotability is built separately.

04

Correction policy

We do not delete a post that turns out to be wrong; we correct it and record the correction itself. The reason is simple: when we correct a mistake and keep it on record, we learn from it. We also record the rejected approaches in this repository’s decision ledger, and that is often the most useful part: knowing why something was not done is as useful as knowing what was done.

Numbers going stale is a separate issue. When a post says “today it is this much”, that sentence may be wrong tomorrow. So there are two rules: a number about the site itself must either match today’s measurement, or the sentence must be dated. A sentence that does neither turns red in the verification suite, and the rule has genuinely earned its keep once: thirteen sentences went stale at once in a single release.

05

What is not written

No news, no industry commentary and no AI-generated filler text is written here. If there is nothing new to say about a topic, no post is written; instead of repeating what a source says, the post links to the source.

Posting frequency is not a goal either. A regular publishing schedule works well when there are things to say piling up; if it forces publishing without them, it thins the content. Thin content has a measurable cost: a search engine can choose to crawl a page and not index it, and that is not a crawl error but a quality verdict. We have written up how a claim is proven separately.

06

The measure of a post

A post’s length is set by its content. There is still a floor: on this site guide pages must carry at least eight hundred words, and we set that threshold by measurement. The reason was Search Console’s own verdict — thin pages were being crawled and not indexed, and an internal measurement taken the same day showed that zero of fifty-eight pages cleared this threshold.

The threshold is protected by code: a guide under eight hundred words cannot go live, and a glossary entry has to stay between three hundred and six hundred words. The second limit is needed too, because length alone is not quality; defining a term in six hundred words makes the definition harder to find.

07

Two readers

Every post is built for two readers: a human and a model. The visible layer addresses the human — heading hierarchy, readable line length, body text contrast. The structural layer addresses the machine — article markup, breadcrumb trail, publication date and author information. Both must tell the same truth; the moment they diverge, one of them is lying.

That is why we put only visible content into structured data. Marking up a frequently asked question that is not on the page, or declaring an author who did not write the post, may seem to work in the short term but costs trust in the long term. We explain what semantic markup is and the concept of an entity separately.

The whole journal is bound to this principle: for every number written, it is clear where it came from, on what date it was measured and how it can be repeated. All the posts listed on this page were written under the same contract.

08

How to follow it

Each language of the journal has its own RSS feed, and the feed address can be reached from the links at the bottom of the page. When a new post is published, it enters the feed, the sitemap and the llms.txt file read by model crawlers at the same time — all three are generated from the same list, so they cannot diverge.