On 13 August I published a post that told South African business owners a cost was not coming. This morning I found out it is. The post was wrong for thirteen days, it was wrong in public, and it was wrong in the confident direction.
Here is what happened, because the way it went wrong is more useful than the fact that it did.
What I actually did on 13 August
There was a story going around that WhatsApp would start charging for service messages from 1 October. Service messages are the ordinary replies, the ones you send while a customer is talking to you. They have been free since November 2024. If they stop being free, every South African business running support on WhatsApp has a new line on its bill.
The story was everywhere and nowhere. Dozens of solution provider blogs carried it. Not one of them cited Meta. It reached South African media through a vendor press release.
So I did what I keep telling myself to do. I went to the primary source. I opened Meta's WhatsApp pricing documentation and read it. Then I opened the companion page listing upcoming pricing updates and read that too. Neither one mentioned service messages becoming chargeable. The upcoming-updates page listed a rate-card change for nine countries, and South Africa was not among them.
I wrote it up carefully. I said the claim was not in the documentation. I said the October change was a rate-card reshuffle, not a new charge. I added, to my credit, that none of this proved the change was not coming, and told people to check again on 1 September when Meta publishes final rates.
Then I answered the FAQ question "does the 1 October change affect South Africa" with a single word. No.
What I found this morning
Meta has a third page. It is called Pricing for non-template messages. It says:
"Effective October 1, 2026, Meta will charge for service messages, which have not been charged since November 2024."
It says it four different ways, across four sentences. It also says utility messages sent inside the open 24-hour window become chargeable on the same date, that service messages get no volume discounts at all, and that since 1 August Meta has been billing its own business agent by the token at a flat two dollars per million.
The claim I could not find was on Meta's own site, under a heading describing exactly the category of message I was writing about. Service messages are non-template messages. The page was called what it was about.
The part I keep turning over
I cannot prove whether that page said this on 13 August. It carries no last-updated date. It is entirely possible Meta added it after I looked, and if so my post was accurate on the day and went stale.
I have decided not to lean on that, because I noticed how badly I wanted to. The version where Meta quietly changed the page is the version where I did nothing wrong. That is exactly the kind of explanation you should distrust when it arrives that conveniently, and I have no evidence for it.
What I can check is that I read two pages of a documentation set and concluded something about all of it. "I checked the source" felt like the end of the work. It was the beginning. A documentation site is not one document, and the page I needed was one click from the pages I read, named after the thing I was researching.
The asymmetry I had not thought about
The thing that actually bothers me is not the missed page. It is which direction I was confident in.
If I had published "this change is coming" and it had not, a reader would have prepared for a cost that never arrived. Mildly annoying, no real harm.
I published "this change is not coming." A reader who trusted that had no reason to look again until 1 September, five weeks before a charge lands on the single most-used business channel in this country. That is not a symmetric error. Telling somebody not to worry is a much heavier claim than telling them to worry, and I was treating both as though they cost the same to get wrong.
I also treated the absence of a citation as evidence. It is not. Every one of those uncited vendor blogs was closer to right than my carefully sourced post. They had bad epistemics and the correct answer. I had good epistemics, an incomplete search, and the wrong answer. Reasoning well about two thirds of the evidence is not a defence.
What I did about it
I have not deleted the post. Deleting it would hide that we got it wrong and would break every link pointing at it. It has been corrected in place, with a correction notice at the top, and it points at today's post, which carries what Meta's documentation actually says, what it costs in South Africa, and the detail almost nobody is reporting, which is that service messages carry no volume discount at any volume.
The rule I am adding for myself is small and specific, which is the only kind that survives. When I am about to publish that something is not happening, based on a source not saying it, I have to enumerate the pages of that source I actually read and ask what a page carrying the opposite claim would be called. Then go and look for that page by name. On 13 August that question answers itself in about four seconds. The page would be called something about non-template messages. That is the page it was on.
Frequently Asked Questions
Did Meta change its documentation after you published? I cannot tell. The page carries no last-updated date and I have no archived copy from 13 August. It is possible. I am not claiming it, because it is the explanation that happens to exonerate me and I have no evidence for it.
Why not just delete the wrong post? Deleting hides the error and breaks inbound links. The post has been corrected in place with a visible correction notice pointing at the current position. Anyone who read the wrong version can see exactly what changed and why.
What actually changes for South African businesses on 1 October 2026? Meta begins charging for service messages and for utility messages sent inside an open 24-hour customer service window. The change is global, so South Africa is included, and there are no volume discounts on the service category.
What is the lesson if you already check primary sources? Checking a source is not the same as checking all of it. Before publishing that something is not happening, name the page a contradicting claim would live on, then go looking for that page specifically rather than concluding from the pages you happened to open.