Blocking AI on your website: what each switch actually does

Someone has told you to block AI from your website. Or you’ve found a switch for it, and you’re wondering whether to use it. Google Search Console, Google’s free dashboard for site owners, has one. So does Cloudflare, the security service your site may sit behind.

Here’s the short answer. There isn’t one ‘block AI’ switch. There are several, and each stops something different. Since 15 September 2026, one of Cloudflare’s switches also stops Google, Bing and Apple from reaching your site. Choose that one to stop AI training, and your listings in classic search—the ordinary, unpaid list of results on Google and Bing—go with it.

Two things AI does with your content

When people say ‘block AI’, they usually mean one of two things.

The first is training: an AI company using your content to build its models, the systems behind tools like ChatGPT and Gemini.

The second is answers. When someone asks ChatGPT a question, the AI can read your page and build its answer from it, then link to you as a source. Google does the same in an AI Overview, the AI-written summary it shows in its results. So does Google’s AI Mode, a chat-style version of Search. That’s how you get cited. I’ve written before about why being cited isn’t the same as being recommended.

Then there’s classic search, which you almost certainly want to keep.

Most switches deal with training or with answers, not both. Their names don’t always tell you which.

Here’s what each switch does

Three of these switches live in robots.txt, a short text file on your website that tells crawlers what they may and may not read. Crawlers are the robots that search engines and AI companies send out to read websites. Whoever looks after your website can add a line to robots.txt.

Each crawler has a name. Google has also added a name that isn’t a crawler at all. Google-Extended is a name you use in robots.txt to say no to training.

SwitchStops AI trainingKeeps you out of AI answersCosts you classic search
Google Search Console: exclude your site from ‘Search generative AI’, Google’s name for its AI features in SearchNoYes—AI Overviews, AI Mode and the AI features in Google Discover (Google’s feed of suggested articles)No
Google-Extended, blocked in robots.txtYes—Google’s Gemini, and the AI in Google SearchNot from AI Overviews or AI ModeNo
Cloudflare Training setting: Disallow AI TrainingPartly—it doesn’t yet tell Bing you’ve said no to trainingNoNo
Cloudflare Training setting: BlockYesYes—AI Overviews and AI ModeYes—it blocks Google’s, Bing’s and Apple’s search crawlers
OpenAI’s GPTBot, blocked in robots.txtYesNo—ChatGPT search can still show you—
OpenAI’s OAI-SearchBot, blocked in robots.txtNoYes, except as navigational links (a link to your site that ChatGPT can still show)—

A dash means the switch has nothing to do with search engines like Google and Bing.

Three mistakes that will cost you

Choosing Cloudflare’s ‘Block’ to stop training

If your site sits behind Cloudflare, its security settings now include a Training setting for AI crawlers. Since 15 September 2026, its Block option stops more than just training. In Cloudflare’s words: ‘It will stop Applebot, Bingbot, and Googlebot from reaching your site — search included.’ Those are the crawlers Apple, Bing and Google use to list you in classic search.

The setting that stops training and keeps search is Disallow AI Training. It uses robots.txt to tell Google and Apple that you don’t want your content used for training. It blocks the training-only crawlers run by Amazon, Anthropic (the company behind Claude), Meta and OpenAI.

It doesn’t reach Bing yet. Cloudflare says Microsoft is building a way to read a no-training request in robots.txt, ‘targeted for early 2027’. Cloudflare adds that until then, Disallow AI Training ‘will not automatically convey a no-training preference to Bing through robots.txt’.

Bing has its own no-training option that keeps you in its results. It’s NOARCHIVE, a tag in the code of each page. Bing’s help page gives it two AI jobs: ‘Do not link in Chat and Copilot. Do not use the content for training Microsoft’s generative AI foundation models.’ So at Bing today, stopping training also costs you links in Copilot, Microsoft’s AI assistant.

If you chose Block before 15 September, Cloudflare changes the setting for you: ‘Previous Training selections of Block or Block on pages with ads will migrate to Disallow AI Training.’ The risk is choosing Block from now on. Cloudflare’s Block AI Bots help page, last updated 1 July 2026, still says every way of blocking training also blocks Googlebot. The 15 September announcement says Disallow AI Training doesn’t, and it’s the later of the two.

Blocking Google to stay out of AI Overviews

Older ways out of AI Overviews exist, and they cost more. Googlebot is Google’s search crawler. Blocking it in robots.txt all but takes you out of Google Search. (Google says a blocked page can still be listed if other sites link to it.) Blocking Googlebot catches Apple as well. Apple says, ‘If robots instructions don’t mention Applebot but mention Googlebot, the Apple robot will follow Googlebot instructions.’

The nosnippet tag, another line in a page’s code, is another route. It keeps a page out of AI Overviews and AI Mode. But Google also stops showing the text under your listing in classic search.

The Search Console switch now does the job instead. Google says it ‘isn’t used as a ranking or inclusion signal affecting other parts of Search’. So, by Google’s account, using it won’t affect whether or where you appear in the rest of Search.

Switching off AI Overviews to stop training

If what you object to is Google training its AI on your content, the Search Console switch won’t do it. Google’s help page says the control ‘doesn’t affect AI training’, and points you to Google-Extended instead. What it will do is take you out of AI Overviews and AI Mode: ‘You won’t receive any traffic or impressions from these features.’ (An impression is a link to your page being shown to someone, whether or not they click.) Before you switch it off, it’s worth checking how often those features show links to your pages: I wrote about reading Search Console’s Generative AI performance report in my post on August’s Search Console numbers.

OpenAI splits it the same way, and its documentation says ‘Each setting is independent of the others’. Blocking OAI-SearchBot takes you out of ChatGPT search answers and does nothing to training. Blocking GPTBot does the reverse.

So if training is what you’re against, the switches at Google and OpenAI are Google-Extended and GPTBot. Neither takes you out of those AI answers.

The full map

The table covers the switches I’d expect most businesses to meet. I’ve mapped 50 in all, and marked the ones where I’ve worked the answer out and the ones with no source. I’m happy to share the full map.

If you want to talk about which of these your site should be using, please get in touch with me at [email protected].

Everything here was checked against each company’s own documentation on 21 September 2026, and it changes often.

David Rosam, photographed by the River Arun in Littlehampton.

David Rosam has been working online for more than 25 years, after a career in direct marketing copywriting for the tech and financial services industries.

Today, he specialises in Content-Leveraged Search Engine Optimisation—from audits, through research and strategy to implementation.

He was probably the UK’s first SEO Copywriter.

Leave a comment