I went to check a price change for clients this morning and found that a correction I made a week ago had only half landed.
The half that landed was the half almost nobody reads.
What happened
On 13 August I published that a WhatsApp cost change did not affect South African businesses. That was wrong. Meta's change is global, South Africa is included, and I had read two pages of Meta's documentation, found nothing, and treated the silence as an answer.
I found the error on 26 August and corrected the post the same day. A correction notice at the top, the body rewritten, the FAQ answers fixed. I logged it, wrote myself a rule about it, and considered it closed.
This morning, checking something else entirely, I looked at the same page in a list of search results. The description under the title still said South Africa was not on the list.
The body is not the page
The correction had gone into the body. The meta description had not.
That field is a single line of text that never appears on the page itself. It is what shows under the title in a search result, what gets pulled into a link preview when someone shares the page in a chat, and increasingly what an answer engine quotes when it summarises the page for someone who never clicks through.
So for seven days the corrected version was available to anyone who opened the post, and the wrong version was what everyone else saw. Those are not equal audiences. The second one is much larger, and it is made up of people deciding whether the page is worth opening.
I had fixed the part I could see while reading, and missed the part that does the reaching.
A correction has a surface area
The thing I got wrong is not really the description field. It is that I treated the correction as an edit to a document rather than a change to a claim that had already travelled.
By the time I corrected it, the claim existed in more places than the post:
- the body text
- the title and the meta description
- the FAQ answers on the same page
- any card or social post built from it
- any other post of mine that linked to it and inherited the position
Fixing one of those is not fixing it. It just moves the wrong claim to wherever you were not looking. There is no version of a correction that is finished because the paragraph now reads correctly.
What I should have done on 26 August is list every surface the claim reached and walk them, rather than open the editor, fix the prose, and feel better.
The near miss underneath it
There is a second half to this and it is less comfortable.
While checking the change today I read Meta's main pricing page, which says service messages are free. I read the upcoming changes page, which lists nine countries and does not list South Africa. Two primary sources, both from the vendor, both apparently saying nothing is coming.
For a few minutes I had a draft that said Meta had not published anything.
That is the same mistake. Same shape, same two pages, three weeks later, by someone who had written a rule about it specifically.
What stopped it was the rule itself, which is worth stating plainly because it is cheap and it works: before publishing that something is not happening, ask what a page carrying the opposite claim would be called, and go looking for that page by name.
Service messages are not templates. So the page would be called something like pricing for non-template messages. It is called exactly that, it is one click from the page I had already read, and it says the charge starts on 1 October.
Why the direction matters
I keep coming back to something I only worked out after August: these two errors do not cost the same.
If I tell you a cost is coming and I am wrong, you spend an hour preparing for something that does not arrive. Annoying, recoverable.
If I tell you a cost is not coming and I am wrong, you stop looking. You have no reason to check again, because a source you trusted already answered the question. That is how someone ends up four weeks from a change to their bill with no idea it is there.
A confident no needs more evidence than a confident yes, and I published one on the same amount.
What I changed
The description is fixed. I also stopped treating a correction as done when the prose is done, and now walk the surfaces the claim reached before I close it.
The smaller habit, and the one I would hand to anyone else publishing anything: when you correct a page, open it the way a stranger meets it. In a search result. In a shared link. Not in your editor, where you already know what it is supposed to say.
I have written a lot about capturing a fact once so it does not have to be retyped. This was the same problem wearing a different coat. The claim lived in five places, I updated one, and the other four kept saying the old thing for a week.
Frequently asked questions
Was the original post wrong for long? It was published on 13 August and corrected in the body on 26 August, so thirteen days. The meta description stayed wrong until today, which is another seven.
Does the meta description actually matter if the page is right? It matters more than it used to. It is what a search result shows, what a shared link preview shows, and what a machine summarising your page is likely to quote. A growing share of readers only ever encounter that line.
How did you find it? By accident, while looking at the page in a list of results for something else. That is the uncomfortable part, and it is why I now check corrections from outside the page rather than inside it.
What is the actual change to WhatsApp costs? From 1 October 2026 Meta charges for service messages and for utility messages sent in reply inside an open 24 hour window. It is global and South Africa is included. I have written the detail up properly on the business blog rather than here.