Tracking versus your own server

"But they don't actually read my documents, do they?" I get this question often enough that I have a standard answer, and the standard answer annoys people because it is not the reassurance they were fishing for. The honest response is: probably not in the way you are imagining, and that is the wrong thing to be worried about.

Nobody is sitting in an office reading your invoices. That is not how any of this works, and the conspiratorial version of the story does real damage because when it collapses people conclude there was nothing to worry about at all. What actually happens is less dramatic, entirely automated, and much harder to opt out of.

The interesting data is not the contents

In my work I build models on data, and there is a lesson from that trade worth passing on: the contents of a document are usually the least useful thing about it. What carries signal is everything around the contents.

Who created it. When. Who it was shared with, and how quickly they opened it. How many documents that person makes in a week and at what hours. Which people appear together in shared folders. What was accessed immediately before and immediately after. None of that requires reading a single word, all of it is generated automatically as a by-product of the service functioning, and together it describes an organisation with uncomfortable precision.

This is the thing I try to get across: metadata is not a lesser category of data. For most analytical purposes it is the better one. It is structured, it is complete, it is cheap to process, and it does not need any of the legal and technical care that reading content would require. If I wanted to understand a company, I would ask for its access logs long before I asked for its files.

What your own server knows

Here is the part that surprises people, and I want to state it before anybody accuses me of selling a fantasy. Your own server generates exactly the same metadata. Every one of those records exists on a machine in your office too — who opened what, when, from where. Self-hosting does not make the data stop existing.

What changes is who holds them and what else they can be combined with. That second part is the whole argument. A log on my server describes my server. The same log at a provider that also operates a search engine, a mail service, a mobile operating system, a browser and an advertising network joins a set of other records about the same people. The value is not in any single stream. It is in the join, and the join is only available to whoever holds all the streams.

There is a second difference and it is legal rather than technical. Data on a machine in your office is reached through you — somebody has to ask you, or compel you, and you find out it happened. Data held by a provider is reached through the provider, under whatever jurisdiction they operate in, and you may never learn about it. Whether that matters depends entirely on who you are and what you do, and for most small businesses the honest answer is that it matters less than the marketing on either side suggests.

How to find out rather than guess

Two things are worth doing, and both are available to anyone.

The first is boring and nobody does it: request your data. Every major provider has an export function, mandated in Europe, and what comes back is frequently educational. Not the documents — you knew about those. The activity history. People are generally unbothered by the files and quietly unsettled by the timeline.

The second is to look at your own network. Point your router at a DNS server you control and watch what the devices in your home ask for while nobody is touching them. I did this expecting to learn something about my computers and learned considerably more about the television. The devices that talk most persistently are rarely the ones you would suspect.

What I actually recommend

Not withdrawal. I use plenty of hosted services and I am not about to pretend otherwise while typing this into a browser. What I recommend is deciding deliberately which categories of thing live where, instead of letting it be decided by whichever app was most convenient at the time.

My own split, for whatever it is worth: things with legal or personal weight — contracts, accounting, identity documents, family photographs, medical paperwork — live on hardware I control. Things where convenience genuinely wins and the stakes are low live wherever they work best. That is not ideological purity, it is a filing decision, and the useful part is that it was made on purpose.

And the answer to the original question, for the record: no, they are almost certainly not reading your documents. They do not need to. Ask instead who can see the shape of your week, and whether you would have chosen that if anybody had put it to you as a question.