I built a content calendar for August with a target keyword on every line. Thirty one lines, one per day. I was pleased with it.
Five publishing days in, I have replaced the keyword on all five.
Not adjusted. Replaced, on the morning of publication, after a check that takes about two seconds to run.
What the check is
Before I write anything, I ask Google's autosuggest endpoint what it returns for the phrase I planned. It is the same list that drops down under the search box when you start typing.
That is not a ranking signal and I am not going to pretend it is. It answers a cruder question: has a human being typed this string often enough for Google to have noticed.
Five days, five swaps
On 1 August I planned "small business compliance checklist south africa". Nothing came back. I published against "business compliance checklist south africa".
On 2 August I planned "provisional tax south africa small business". Nothing came back. I published against "provisional tax first period".
On 3 August I planned "beneficial ownership cipc filing". Nothing came back for that exact string. I published against "cipc beneficial ownership".
On 4 August I planned "uif registration for employers south africa". Nothing came back. I published against "uif registration for employers".
This morning I planned "employment contract requirements south africa", and that one deserves its own section, because it did not fail the way the others did.
Today's was the worst one
The first four failed cleanly. I asked, an empty list came back, and an empty list is impossible to misread.
This morning the list was not empty. I asked for "employment contract requirements south africa" and got seven suggestions back. If I had glanced at the screen and moved on, I would have ticked the box feeling good about it.
Not one of the seven was the phrase I planned. Every one of them was shorter: "employment contract south africa", "employment contract law south africa", "employee contract south africa", and so on down the list. The endpoint was not confirming my phrase. It was politely telling me what people type instead, and the answer was three words with the country on the end.
A failure that looks like a pass is the expensive kind. I wrote almost exactly that sentence yesterday, about a delete call that returned success four times and deleted nothing. Two completely unrelated systems, two days apart, same shape: the thing did not refuse, it agreed with me and was wrong.
Why I got it wrong thirty one times
Sitting down to write the calendar, I was not choosing search terms. I was describing the articles to myself.
"Small business compliance checklist south africa" is an accurate description of that article. It is a terrible search. Nobody types their own market segment into a search box. They do not think of themselves as a small business at the moment they have a question, they think of themselves as a person who needs to know a thing.
People type the shortest string that will plausibly get them there and let the search engine do the rest. I was writing filenames and calling them keywords.
The counterintuitive part
There is a second finding underneath this one, and it surprised me more.
The country suffix is not uniformly helpful. On compliance terms it is essential: the obligation is South African, so people search it that way. On automation terms it kills the phrase outright. "Workflow automation for small business south africa", "automate onboarding south africa" and "small business automation south africa" each return nothing at all, while the same phrases without the suffix all return live suggestions.
That makes sense once you see it. A person asking about UIF knows it is a South African question, because UIF only exists here. A person who is tired of retyping the same invoice details four times does not experience that as a South African problem, because it isn't one. Adding the country narrows the phrase to something nobody says out loud.
So the rule is not "add South Africa" and it is not "drop South Africa". It is that the suffix belongs where the jurisdiction is part of the question.
What I actually changed
The check was never the problem. I was running it at the wrong end of the process.
It ran on the morning of publication, which is the last possible moment it could still be useful and the worst possible moment to discover anything. By then the angle is set, the internal links are chosen, and the swap is a patch rather than a decision.
For September it moves to the day the calendar is written. Every planned keyword gets validated before it is committed to a line, so the plan starts out made of things people type.
The planning unit changes too. I will plan the head term and let the specific angle live in the title and the slug, instead of planning a four word phrase and discovering on the day that it was a sentence I made up.
The part I keep relearning
I want to be careful not to turn this into a tidy lesson about humility, because that is not what it was.
The calendar was not wrong because I was careless. I put real thought into it. It was wrong because thinking hard about something is not the same as checking it, and I had built a process where the checking happened after the committing.
The check costs one command. I had it in the right place in the workflow and the wrong place in the order. Five days of swaps is a cheap way to find that out, and it is a lot cheaper than a month of posts aimed at phrases nobody has ever typed.
Two other times I have written up something similar, both about the same underlying habit: the day the geo redirect ate my Google rankings and fixing the SEO bugs that hid our site. In both cases the system was working exactly as built. It was built on an assumption nobody had gone and tested.