Can you turn off generative AI from social media and your phone?

Big Tech companies like Google, LinkedIn, TikTok, Meta (which owns Instagram, Facebook, and WhatsApp), and X (formerly Twitter) have been training their AI models on user data, which they argue will improve users' experiences.

In absorbing the plethora of data online and user activity, companies are able to power the fast-growing industry of generative AI (gen-AI). The so-called 'machine-learning models' use all this information to create (depending on what the user asks it to do and what it has been trained on) images, text, and video, which has increasingly become difficult to distinguish from human-created content.

While there are obvious benefits for productivity and accessibility, the director of Waikato University's AI Institute, Albert Bifet, has concerns around “privacy, bias, misinformation, and the concentration of power among a few large tech companies”.

Related stories:

There are also fears about the environmental impact from the use of AI and plagiarism.

Personal users aren’t the only ones who may be wary. Some publishers, like the New York Times and The Guardian, are blocking AI web crawlers which can be used to collect data to train AI chatbots.

Bifet says many generative AI models have already been trained on large volumes of publicly available online data.

"Once content is out on the internet, it can be difficult to control how it’s used for training."

Diagram founder Tom Hovey, who helps organisations in NZ and the UK better understand AI, says developments in the industry are happening so quick that even the information he provides to clients is changing every day.

"Never before have we had such changeable (some would say "buggy" and inconsistent) software being used at such a grand scale. The speed of change is insane."

For example, the free version of ChatGPT initially did not let you opt out of it using your data to train its model. But that recently changed and you can now opt out of ChatGPT using your data to train the model.

However, big tech companies often have no incentive to provide opt-out options when we’re heading towards a future that’s increasingly all about personalisation, he says.

Plus, you won’t always be in control of your data and that’s where the ethical dilemma comes in, he says.

For example, if you leave a message on someone’s phone and they have AI enabled to summarise voicemails. “Have I given you permission to use my information to summarise it, to send it off to an AI provider to have it presented to you? Where did I consent to being recorded and processed in an AI data centre?”

There are also questions about the potential agendas in how chatbots may prompt us to make decisions, he says.

“You can see a future where, as a consumer, you need to be super aware ‘Am I getting neutral information or am I getting information which is being sponsored?’”

Consumer NZ product test writer Nick Gelling says there are some ways to toggle off generative AI functions, but not for everything and, even then, you might not be able opt out of your data being harvested for training the model.

“Consumers elsewhere in the world have better rights than we do for opting out of this kind of thing,” Gelling says.

“We only know about what companies are doing with some of our data because they have to inform European consumers [otherwise] that might not have made its way into the public domain at all.”

Some advice online suggests social media users tap ‘not interested’ or ‘don’t recommend’ on AI-generated posts, or resetting the suggested content in settings (TikTok, Instagram). This doesn’t completely remove them from your feeds, but it may lessen their frequency.

You may be able to identify these posts if they have ‘AI-generated’ labels, but X, Meta (Facebook and Instagram) and TikTok haven’t created a way to block this content completely, Bifet says.

Hovey believes AI-produced content will eventually inundate social media and lead to its downfall.

“They'll just become a ridiculous fake, no connection to humanity pile of trash.”

Earlier this year, Meta introduced a generative AI assistant on all its platforms by default. Users can ask questions or search for content as you would on search engines or chatbots, which can also be posted into a public ‘Discover’ feed. You can ensure your prompts are ‘visible only to you’ via privacy settings.

Gelling says there’s no way to remove Meta AI assistant from your accounts and anything you have posted publicly online on Instagram, Facebook and WhatsApp or searched through the bot can be used to train their models.

Unlike Europe where users have more privacy options, people in New Zealand can only ask Meta to delete data used by third parties to develop its AI, “if you have a concern about your personal information being included in a response from an AI at Meta model”.

Setting your account and posts to private may help limit the future data available for training.

Users can use a tool like Glaze or Nightshade before posting images to protect against AI mimicry by adding a layer of minor invisible changes that confuse AI algorithms, Gelling says. Alternatives include Cara, a platform with an explicitly anti-AI stance.

Launched in 2023 by Elon Musk’s X, the Grok AI chatbot is enabled as a default for users of the platform. It’s also using your activity on X and posts to train the model. You can disable this by heading to Settings > Privacy and Safety > Grok > Deselect the option under ‘Data Sharing’.

To opt out of having your data being used by third parties, head to the same privacy settings > Data sharing and personalisation > Data sharing with business partners.

The short-form video platform’s generative AI tool Symphony can turn images or text prompts to video and offers advertisers realistic digital avatars to showcase their products. There’s no option to opt out of seeing this content or the platform using your data to train its algorithms.

“I imagine [ByteDance is] working behind the scenes to understand what gen AI content is getting attention and who's creating it, so that it can recommend this 'winning' content to keep the platform so addictive. It's an addiction flywheel,” Hovey says.

Similarly to Meta, you can potentially limit the future data available for training by making your account private. There’s also the option of filtering video keywords, but still some AI-generated content may slip through the labelling system.

The Microsoft-owned social networking site trains on users’ public data by default, apart from those who appear to be in high school. You can toggle this off by heading to settings> data privacy> move the slider to off for ‘use my data for training content creation AI models’.

Google search results are often topped with AI Overviews – a generative AI summary based on information online - that might answer a question you’ve entered.

If you would rather do your own research, try other search engines or find browser extensions which automatically default to the web-links only page, Gelling says. Or you can make the web-links page the default yourself by following the instructions here. Alternatively, you can type ‘-ai’ after a search query, he says.

Hovey says it does appear that Google uses our data inputted into its gen-AI assistant Gemini or web search to train its AI models. However, if you have a paid account, you have some more privacy controls for Gemini.

Gemini can work with your phone apps, like Gmail, Google Docs, photos, camera, Google Maps, and Spotify. It may be automatically installed on an Android device when you receive it, or it may come through in system updates on your phone. You can usually switch back to the traditional 'voice assistant' system instead of Gemini through phone settings, if you prefer.

Apple initially offered its own assistant, called Apple Intelligence, as opt-in for users before enabling it by default (for new users or those upgrading to iOS 18.3, iPadOS 18.3, and macOS 15.3) earlier this year.

It can summarise calls, emails, messages, voicemails, prioritises your tasks, creates written content based on prompts and descriptions as well as emojis, images and short-form videos based on your gallery.

You can turn off Apple Intelligence through your phone settings >Apple Intelligence and Siri > toggle off, Gelling says.

“But Apple does say it doesn’t use any customer interactions to train its core models,” Gelling says.

“What I think that means is it will use your interactions to tweak your personalised model that is there on your phone to kind of make that more relevant to you and feel more personal, but those changes won’t get fed back into Apple’s servers to update the global version.”