What does it mean when users rip out your AI feature?

When users go to the trouble of removing your AI feature, the feature costs them more than it gives back. Treat the opt-out as a signal, log it like one, and fix the value or the default before you ship more AI into the product.

What happened with RemoveMacAI?

On Oct 5, MacRumors' Tim Hardwick covered RemoveMacAI, an open-source command line tool that disables Apple Intelligence on macOS 27, removes the downloaded models, and stops them from downloading again. MacRumors reports that Apple removed the toggle to turn Apple Intelligence off wholesale in macOS 27, and that switching off individual features leaves the models on the drive. The tool frees roughly 12GB, depending on which models are present, and it picked up hundreds of GitHub stars after its release last week.

Read that from the user's side. People are installing a configuration profile from a GitHub repo, at their own risk, to delete features Apple shipped for free. They give up Writing Tools, Genmoji, summaries, and more. MacRumors says plenty are happy with the trade, especially on 256GB Macs that can't be upgraded.

That's the most honest product feedback an AI team can get. Nobody pays a cost to remove something they use.

Why is an opt-out a Signals event?

An opt-out is a user telling you, with behavior, that the feature charges them something and returns less. Storage. Screen space. Attention. Trust. A survey answer can be polite. A user who digs through settings or a forum thread to switch your feature off is done being polite.

Most AI SaaS teams track only the happy side: prompts sent, outputs generated, "AI adoption." That count can rise while the people who matter quietly turn the thing off. If you only log usage, you never see the exit.

Activation for an AI product is the first kept output. The opt-out is the mirror image: the moment a user decides the output isn't worth keeping around at all. You need both events on the same dashboard.

What does the opt-out tell you?

It tells you which of three things broke, and each one has a different fix.

1. The default was wrong. The feature turned itself on for people who never asked for it. Fix: opt-in, or on only for the segment where it earns its keep.
2. The cost is visible and the value isn't. Storage, latency, a sidebar that eats the screen, a bill that went up. Fix: shrink the cost, or show the value right where the cost shows up.
3. The output isn't good enough for this user. They tried it, it missed, they removed it. Fix: narrow the job until it lands, or stop pushing it to that segment.

You can't tell which one without context, so log every opt-out with who, when, and what happened right before.

How do I track AI feature opt-outs?

Log every off switch as an event, with enough properties to read the reason.

• An ai_feature_disabled event with the feature name, plan, account, and days since signup
• Whether the user ever reached a kept output before turning it off
• What happened in the session before the disable: an error, a slow response, a help search
• One optional prompt on disable: "What made you turn this off?" in free text

In PostHog, that's one event and a four-step funnel: enabled, first output, kept output, disabled. Watch where people fall into "disabled." That step is your problem.

Then roll it up to the account. One user switching off a feature is a preference. Three seats at one account switching it off in the same week is a churn signal for whoever owns that account.

What if I hide the off switch instead?

Then users build their own, and you lose the data.

Apple's choice is the warning. Remove the clean switch and a third-party tool does the job anyway, with no telemetry back to you. In B2B, that tool looks like an admin who blocks your feature at the workspace level, a security team that strips the integration, or a buyer who picks a competitor because your AI couldn't be turned off for one team.

An honest off switch is cheaper than a workaround you can't see. It keeps the user in the product and keeps the reason in your data.

How does this change how I sell the AI feature?

Sell the job the feature does for a specific buyer, and stop selling AI as a bundle everyone gets whether they want it or not.

If your opt-out data shows the feature lands for one segment and annoys another, your message is too wide. Position the AI feature by the outcome for the people who keep it, and make it optional for everyone else. "Included for every user" reads as a cost line until the buyer sees it work.

What should I do this week?

1. Add a disable event for every AI feature you ship.
2. Build the four-step funnel: enabled, first output, kept output, disabled.
3. Add the one-question prompt on disable. Read every answer for two weeks.
4. Roll disables up to the account, and send multi-seat disables to the account owner.
5. If an AI feature has no off switch, ship one before you ship the next AI feature.

When users pay a cost to remove what you gave them for free, believe them.

Start Signals, Convert, Grow

FAQ

Is a high AI feature opt-out rate always bad?

Not if it comes from the wrong segment. If buyers outside your target customer turn it off and your best accounts keep it, the feature is fine and the default is wrong. Read opt-outs by segment before you judge the feature.

Should AI features be opt-in or on by default?

On by default only where your data shows most users reach a kept output. Everywhere else, opt-in, with a clear prompt at the moment the job comes up. A default most people reverse is a tax on every user.

What is a good disable prompt?

One question, optional, free text: "What made you turn this off?" Skip the multiple choice list of reasons you'd like to hear. Read the raw answers yourself.

Does this apply to B2B admins, too?

Yes. An admin disabling your AI feature for a whole workspace is the strongest version of the signal. Route it to the account owner the same week.