← All guides · Published November 14, 2026 · Reviewed November 14, 2026 · By the Ingenious AI team · 6 min read
How Does Applebot-Extended Opt You Out of Apple Intelligence Training?
The short answer: Applebot-Extended is Apple's training opt-out token, mirroring Google's arrangement: the fetching crawler is Applebot, which feeds Siri, Spotlight and Apple search features, and the Extended token governs whether fetched content may also train Apple Intelligence models. Blocking Applebot-Extended — a verified training-class entry in AgentBlock's database — declines the training use while leaving Applebot's indexing, and the visibility that comes with it, alone.
Build my block kit — free · Test a policy in the simulator first
What is Applebot-Extended?
Apple's AI training opt-out token, a verified training-class entry in AgentBlock's database. Like Google-Extended, it is a policy token rather than a fetching program: it tells Apple not to use your content to train Apple Intelligence, without affecting Applebot's crawling for Siri, Spotlight and search.
Apple's two-layer arrangement
Apple's documentation draws the same line Google's does. Applebot is the crawler that visits your pages and powers Apple's search surfaces — the results behind Siri answers and Spotlight suggestions. Applebot-Extended is the additional token a site uses to say: index if you like, but do not train on what you find. Owners routinely miss the second layer because the first is the name in their logs.
The practical consequence mirrors the Google case. Blocking Applebot itself to stop training would cost Apple-ecosystem visibility that many businesses — especially local services whose customers ask Siri for recommendations — cannot afford to lose. The Extended token exists so that nobody has to make that trade.
Why this opt-out is easy to forget
Apple's AI training is quieter than the headline vendors', and its crawler volumes on a small site are modest — a few hundred fetches a month where GPTBot brings thousands. Quiet entries are the ones that slip through a policy built from log impressions: the owner blocks the five names they recognise and ships, leaving Applebot-Extended, cohere-ai and anthropic-ai in the unexamined tail.
That is an argument for deciding by class and database, not by log familiarity. AgentBlock's builder presents all 25 entries with classes and evidence grades, so the training decision covers nine entries whether or not each one has personally visited you yet. A crawler that has not arrived is still worth a decision today — arrival is only ever a matter of time, and deciding late means deciding under pressure.
Expressing and evidencing the decision
As a token, Applebot-Extended is expressed in the robots and policy layer of the kit, version-stamped like every other decision. The evidence trail is the policy export plus the kit hash, verified by the kit verifier; there is no block log for a token, because nothing fetches under that name. Owners accustomed to counting refusals should file this one under 'declarative controls' alongside Google-Extended.
The enforcement-gap audit covers it: the audit compares live robots.txt and meta tags against the edge policy and reports drift. If a site migration drops the token from the production robots file, the audit is the mechanism that notices — nothing else will, because the failure is an absence.
The training class, completed
With Applebot-Extended understood, the training class reads as a complete set: three vendor fetching crawlers from OpenAI and Anthropic (GPTBot, ClaudeBot, anthropic-ai), Meta-ExternalAgent, CCBot's open corpus, Bytespider, cohere-ai, and the two opt-out tokens from Google and Apple. Blocking the set is the coherent 'no training use' posture, and each member is blocked through its own documented mechanism.
Owners who allow training — open documentation projects, public research, standards bodies — make the mirror decision per entry. AgentBlock's defaults block the class, but the builder treats every entry as an owner's flip, and the impact preview will show you each crawler's real visit volume before you decide either way.
Five actions to take today
- Do not block Applebot itself. It powers Siri and Spotlight visibility; the training decision belongs to the Extended token.
- Set Applebot-Extended to block. Express the Apple Intelligence training opt-out inside your AgentBlock policy.
- Decide by class, not by log fame. Work through all nine training entries in the builder, including the quiet ones that have not visited yet.
- Audit the robots layer. Run the enforcement-gap audit after any migration or CMS change to confirm the token is live.
- File the policy export. Tokens produce no block log, so the dated, hashed policy record is your evidence of the opt-out.
Worked example: a hotel keeps Siri visibility, declines training
Casa Mirador, a fictional 22-room hotel, gets a meaningful share of its direct bookings from guests who ask Siri for somewhere to stay nearby — visibility it traced to Applebot's indexing. Its owners nevertheless objected to their photography and room descriptions training models, and for a month the site blocked Applebot outright while they 'thought about it', costing an estimated 40 Siri-referred sessions.
The corrected policy allows Applebot's world entirely — it is not in AgentBlock's database — and blocks Applebot-Extended as the training opt-out. The owners' policy note records the split explicitly: indexing welcome, training declined, decision dated, kit hash filed.
In the following quarter the hotel's logs showed Applebot fetching on its usual pattern of roughly 300 requests a month, and direct bookings recovered to their earlier level, with the training posture the owners actually wanted now in force.
Frequently asked questions
Is Applebot-Extended a crawler I will see in my logs?
No — it is a policy token, like Google-Extended. Applebot does the fetching; the Extended token governs whether what Applebot fetches may be used for Apple Intelligence training. You express it in robots and policy, and evidence it with your policy record.
What do I lose if I block Applebot by mistake?
Apple-ecosystem visibility: Siri answers, Spotlight suggestions and Apple search features draw on Applebot's index. For local and travel businesses that can be real booking volume, as the worked example shows — and the loss is silent, with no error anywhere.
Does Applebot-Extended affect Siri citations of my content?
The token governs training use. Apple's documentation separates that from Applebot's indexing role behind Siri and Spotlight. AgentBlock's database records Applebot-Extended as training-class for exactly that reason.
Why is Apple in the training class if its crawler is small on my site?
Class follows function, not volume. The training decision should cover every vendor's training mechanism before it activates at scale on your site; deciding by class in the builder takes minutes and covers entries whose first visit is still ahead of you.
Related guides
- How to Block AI Scrapers and Bots from Your Website
- Which AI Crawlers Should You Allow?
- How Do You Block AI Bots on WordPress, Cloudflare, or Plain JS?
- Pre-deploy breakage simulator — test a policy before it ships
- Impact preview — your server log against a draft policy
- How Often Should You Review Your AI Crawler Policy?
- Does Blocking AI Crawlers Hurt Your SEO?
- What Is an AI Crawler?
- How Do You Block GPTBot Without Losing ChatGPT Citations?
- How Do You Block ClaudeBot and Keep Claude Citations?
- Block evidence — parse a log, build an evidence pack
- The dated agent database — 25 entries, graded