Turning it off

Having spotted a feature, you may want it gone — or you may want it and want to know what it does with your material. Either way, the switches.

Where they hide

There is no standard. In practice, look in this order:

The feature’s own menu. The three dots beside the suggestion, or a small settings icon on the panel itself. Fastest when it exists.

Account or profile settings, usually under Privacy, Data, or a section named after the feature family — “Smart features”, “Intelligence”, “Assistant”.

A separate privacy dashboard. Several large providers keep data controls somewhere entirely different from ordinary settings.

Per-document or per-conversation, for some tools.

Nowhere. This is a real answer, and the rest of the lesson deals with it.

Hiding is not disabling

The distinction that matters most, and the one interfaces are vaguest about.

Hiding removes the suggestion from your screen. The processing may continue — the content is still read, still sent, still analysed. You have changed your view, not the data flow.

Disabling stops the processing.

The wording is your only clue, and it is a weak one. “Show suggestions” is almost certainly a display toggle. “Allow [product] to analyse your messages” is closer to the real thing.

When it matters, verify from the privacy documentation rather than the toggle label. And where only a display toggle exists, treat the processing as ongoing.

The training opt-out is a separate switch

Frequently missed, because people reasonably assume one control covers it.

Whether your content is used to train future models is usually a different setting from whether a feature appears. You can switch off suggestions and still be contributing text.

Look for wording about improving our services, model training, human review, or quality review. That last one matters: some services allow staff to read samples of user content, and it is a distinct question from automated training.

Business and enterprise tiers frequently exclude training by default, which is part of what the price buys — E2·L9’s territory, and worth confirming rather than assuming for your specific plan.

When there is genuinely no off

Increasingly common, and worth being straight about: some features cannot be disabled. AI summaries in search. Prediction baked into a keyboard. Vendor processing described as integral to the service.

Your options are narrower and still real:

Use a different tool for that content. The sensitive draft goes in a different editor; the private search goes to a search engine that does not generate answers.

Use a local-only option. E11·L5’s argument, arriving as a privacy answer rather than a cost one.

Change what you put in. Names removed, figures altered, the shape of the problem rather than the thing itself.

Ask your organisation. If work data is involved, this is not your decision alone, and someone is accountable for it.

Choosing not to use a tool for certain content is a legitimate control, and sometimes the only one you have.

Do the sweep once

Setting these individually as you meet them means you never finish. Do one pass:

Phone keyboard. Phone assistant and photo features. Email. Documents. Messaging. Browser. Work tools. Account-level privacy dashboards for each major provider you use.

An hour, once, then a quick check when something is redesigned — which is when settings quietly return to their defaults.

Try it now (6 minutes)

Pick the tool holding your most sensitive material. Find its AI settings.

Determine two things: whether the toggle hides or disables, and whether a separate training opt-out exists. Most people discover the second one exists and was on.

Check your understanding

1. A toggle labelled 'Show suggestions' most likely:

Stops the processing entirely Only changes what you see, while the analysis continues Deletes previously processed content

2. Turning off an AI feature usually:

Also stops your content being used for training Leaves the training question to a separate setting Deletes your data from the provider

3. When a feature genuinely cannot be disabled, a legitimate control is:

Using a different tool for that particular content Accepting it, since nothing can be done Reporting the vendor

Check answers

Recap

Look in the feature’s own menu, then account privacy settings, then a separate privacy dashboard. Distinguish hiding from disabling, because the wording is vague and only one stops the processing. Check separately whether your content trains future models — that is usually its own switch and often on. Where there is no off, changing what you put in or which tool you use is a legitimate control. Sweep once, then re-check after redesigns.