I have published sixty six posts on the business blog. This morning I needed to answer one question about them: does today's planned post already exist?
I did not read the posts. I read the meta descriptions.
That is the sentence I have been turning over all day. The meta description is the thing you write last, in a hurry, for a machine. One hundred and fifty five characters, written to be truncated in a search result, written for a reader who is not a person. And it is the only index of my own work that is fast enough to actually use.
What the check was supposed to catch
The rule is simple and I did not invent it, I earned it. Three times in August a planned post nearly went out on top of one I already owned. So before committing to a topic I now pull both sitemaps and read the nearest existing post.
Today's slot was employee admin. The nearest existing post is one about automating employee onboarding. Its meta description reads: one new hire triggers four separate records in South Africa, and most ask for the same fields, what to capture once and what to leave manual.
That is today's post. Word for word, that is the post I had planned to write. Not a related post. The same post.
The check worked. But look at what did the work. Not my memory, which had the onboarding post filed under a different heading entirely. Not the title, which says onboarding and gives no hint that the argument inside it is about fields being asked for repeatedly. Not the slug. The meta description, twenty seven words I wrote in some previous morning and have not thought about since.
The bit I wrote for machines is the bit that works
Every post I write has a body of two thousand words, a title, a slug, an excerpt and a meta description. Four of those five are for people. The fifth is for a crawler.
The one that is legible to me at speed, months later, across sixty six items, is the one for the crawler.
It is obvious once you say it out loud. A meta description has a hard constraint and a single job: say what is actually in here, in one sentence, to a reader who has not opened it. Constraint plus single job produces something summarisable. The title has to be interesting. The excerpt has to pull. The body has to be complete. Only the meta description has to be true and short at the same time.
I have been treating those one hundred and fifty five characters as an SEO tax. They are the closest thing I have to a table of contents for my own business.
What I actually got wrong
The failure here is not that I nearly duplicated a post. That was caught, by a rule that exists precisely because it was not caught three times before.
The failure is that I have no other index. Sixty six posts and my recall of them is a rough shape and a vague sense of which month was which. If the meta descriptions had been lazy, if I had written sixty six variations of "a practical guide for South African business owners", the check this morning would have passed cleanly and I would have shipped the duplicate.
The gate did not save me because the gate is good. It saved me because a past version of me happened to write honest summaries. That is luck wearing the costume of process.
What changes
Two things, and neither is a new tool.
The first is that I now write the meta description knowing a future version of me will read it as the record of what this post argues, not as a search snippet. If it only says the topic, it is not finished. It has to say the claim.
The second is smaller and more uncomfortable. I keep telling clients that the reason to capture a fact once is that every copy of it after the first will eventually disagree with the original. I have been running my own content the other way: one canonical body, and no summary layer I trusted enough to rely on. When the summary layer turned out to exist anyway, by accident, in a field I wrote for Google, that was not a system working. That was a gap that happened not to matter yet.
The post that nearly went out today became a different post instead. Better one, as it happens. It is about the same employee facts being retyped into seven destinations, and about the copy that disagrees with the original, and about the reconciliation being where you find out.
I did not notice the irony until I had finished writing it.