Meta pays to look over your shoulder. Work out what your work is worth.
Meta has a new model, Muse Spark. Use it and you pay almost nothing, on one condition. Your prompts and the model's answers go into the training of their next model. An average discount of about 95%. I'd call it something else.
What Meta is actually offering
TechCrunch AI puts it plainly. Muse Spark is meant for running coding agents and other agents. Meta attaches an explicit discount, averaging around 95%, for users who "contribute" to the development of future models. Contributing means one thing here: your prompts and the outputs go along.
That is the whole point. Meta has worked out what your usage is worth to them, and you get that amount back on your invoice. They call it a discount. It is a price on your work, set by the party buying it.
Do the sum for your own desk
You don't have to take my word for it. Run the numbers.
Take what you pay per month for a model like this. Take 95% off. What's left is what you save. It is also what the other side thinks your input is worth, every month.
Then look at what goes into that input. For a photographer it's pricing and client names. For a DJ it's the fee you ask and the fee your colleague asks. For an organiser it's the budget for an event that hasn't been announced yet. For a coach it's what someone wrote you in confidence. You don't type that in on purpose. It rides along in the email you have summarised and the quote you have drafted.
Two things are certain. You know what you save in advance. What you give away, you can't ask back. A model trained on your text can't be told to forget it knows your text.
What people type is 'safe AI assistant for freelancers'
Search that and you mostly get advertising. So ask three questions yourself, and ask them in writing:
- Is my input used to train models? Yes or no. "We value your privacy" is not an answer.
- Where is my data held and who can reach it? Country, party, and what happens when you stop.
- What is the full price? Ask for the price without data sharing too. If the gap is large, you now know what they earn on your data.
Wanting an AI assistant that doesn't share your data isn't paranoid. You wouldn't keep your client files in a stranger's attic because the storage is free.
What you can do with this tomorrow
Take twenty minutes. Open the tools you put client work into today and find one line per tool: is my input used for training, yes or no. With most of them it's written down somewhere near the bottom.
Note next to each tool what actually goes in. Names, rates, contracts, or just loose text. Then you can see at a glance where to be careful and where it doesn't matter.
If you find a "yes" on a tool that handles client data, that's your one job this week. Switch that tool off for that work, or pay the full rate.
Here's how the sum works out for me. I work for €35 a month on my own server in the Netherlands, and what a client sends me stays with that client. That's not a sales line, it's the same calculation as above: I'd rather pay full price than sell someone else's work. And if the sum comes out differently for you, I'll say so.
Frequently asked questions
I only put harmless things in. Can I just take the discount?
Probably yes. Just be honest about what "harmless" means. Summarising an email means pasting the whole email, name and rate included. If you genuinely only paste loose text with no client data, the discount is simply a discount.
Am I allowed to put client data into an AI assistant at all?
You are, as long as you know where it goes and your client could reasonably expect it. Ask the provider for a data processing agreement and for a clear yes or no on the training question. If you can't get that in writing, that's your answer.
How do I find out whether my current tool trains on my data?
Look in the settings under privacy or data controls; there's often a toggle. If it isn't there, email support with one line: "Is my input used to train models, yes or no?" Keep the reply with your records.
