Siri AI Answers From the Web Now. Nobody Can Tell You If You’re In It.

Siri AI starts rolling out to the public on September 14, 2026, when iOS 27 ships, and it does something the old Siri never did: it goes to the web, reads pages, and writes an answer. Apple’s June 8 announcement describes an assistant that can “get up-to-date information from the web on virtually any topic and generate a helpful answer,” sitting inside Spotlight on iPad and Mac, one swipe down from the Dynamic Island on iPhone, and in a dedicated Siri app that syncs conversations across devices. Apple reported more than 2.5 billion active devices in January 2026. So a new answer engine is about to be switched on inside the default search box of a very large installed base, and if you run a website, here is the uncomfortable part: Apple has published exactly one sentence about how your pages might be credited, and no tool at all for finding out whether they were.

What is Siri AI and where does it show up?

Siri AI is the rebuilt version of Siri that Apple announced at WWDC on June 8, 2026, and it is a conversational assistant with three sources of knowledge: what is on your screen, what is in your personal apps, and the live web. That third source is the one that matters for marketers, because it turns Siri from a command line for your phone into an answer engine that competes with Google AI Mode, ChatGPT and Perplexity for the same questions.

The surfaces are the story. On iPad and Mac, Siri AI lives inside Spotlight, the search box people already use to open apps and find files, so a question typed there can now come back as a generated answer instead of a list of documents. On iPhone, a swipe down from the Dynamic Island opens a conversation. A control-click on any image, file or block of text offers “Ask Siri.” There is a dedicated Siri app that keeps your conversation history and syncs it through iCloud, so a question started on a Mac can be picked up on an iPhone. It reaches Apple Watch, CarPlay, AirPods and Vision Pro. And Siri mode in the Camera app lets someone point a phone at a product and ask about it.

The rollout is narrower than the announcement sounds. Apple’s iOS 27 page says the software arrives September 14 with Siri AI “rolling out in English,” and the June press release says Apple will “quickly expand support for more languages.” It runs only on Apple Intelligence hardware, which in the iPhone line means the 15 Pro and everything from the 16 onward. It is not available on iPhone, iPad or Apple Watch in the European Union at launch, and not in China at all. And the fine print on Apple’s own page notes that server-side features including Siri AI are “subject to daily usage limits,” with “expanded access to such features” to be sold “for a fee in the future.” Read that as Apple metering the expensive queries. It is a beta with a cap, launching in one language on newer hardware, and it is still a brand-new answer surface sitting on an installed base Apple puts at more than 2.5 billion devices.

Does Siri AI use Google Gemini?

Partly, and the part that is public is smaller than the headlines suggest. On January 12, 2026, Google and Apple issued a joint statement saying “the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology,” and that those models would “help power future Apple Intelligence features, including a more personalized Siri.” The same statement says Apple Intelligence “will continue to run on Apple devices and Private Cloud Compute.” Apple’s June press release, for its part, credits Siri AI to “the next generation of Apple Foundation Models that run on device and on servers using Private Cloud Compute” and does not name Google once.

So the model lineage runs through Gemini. What does not follow, and what nobody outside the two companies has shown, is that Google’s search index or Google’s ranking carries over into Siri’s web answers. The retrieval side, the part that goes out to the web and picks pages, is described by Apple as Applebot’s job. Apple’s crawler documentation, updated September 4, 2026, says Applebot’s crawled data “may be used to provide additional context and up-to-date content when AI models are used to generate output for display in Apple products and services,” and gives “answering broad world knowledge questions in Siri” as the example.

That is the honest version of the answer. The language model has Gemini in its family tree. The pages it is handed appear to come from Apple’s own crawl. Whether a page that ranks well on Google is more likely to be pulled into a Siri answer is a reasonable guess and nothing more, and anyone selling you a “Siri optimization playbook” this month is selling the guess. What we do know about how language models choose their sources in general still applies: clear, self-contained passages that answer a specific question get picked more often than pages that make the model work for it.

How does Siri AI decide which websites to use?

Through Applebot, using a crawl Apple has run for years to power Spotlight, Siri Suggestions and Safari, and Apple has published a short list of what its search ranking considers. The September 4 version of the Applebot page says Apple Search “may take the following factors into account”: aggregated user engagement with search results, relevancy and matching of search terms to webpage topics and content, the number and quality of links from other pages on the web, approximate user location signals, and webpage design characteristics. It adds that these factors carry “no (pre-determined) importance,” which is Apple’s way of saying it will not tell you the weights.

Two things in that list deserve a second look. “Webpage design characteristics” is unusual; Google does not publish anything with that wording, and Apple does not define it. And “aggregated user engagement” means Apple is watching what people click and tap after a Siri or Spotlight result, which is a signal you cannot buy and can only earn.

The same page tells you how the crawler behaves. Applebot renders pages in a browser, so JavaScript, CSS and XHR that are blocked in robots.txt can stop it from seeing your content. It ignores crawl-delay. It caches aggressively to avoid re-crawling. And this line is the one to underline: “If robots instructions don’t mention Applebot but mention Googlebot, the Apple robot will follow Googlebot instructions.” If you have ever written a Googlebot-specific rule and never wrote an Applebot one, Apple has been inheriting your Google rules the whole time, and now those rules decide whether your pages can feed a Siri answer.

Apple also lists a page-level structured data signal that is new to this category. If your JSON-LD marks a page with the schema.org property isAccessibleForFree set to false, Apple says the page is still eligible for search results but “Applebot will not use that content as additional context when AI models are used to generate output.” Section-level markup with hasPart is not supported. For a publisher with a paywall that is a clean opt-out. For anyone who copied a template that set the property to false by accident, it is a silent removal from Siri’s answers, and it is worth checking your markup this week.

Should you block Applebot or Applebot-Extended?

For almost every company that sells something, no to both, and the reason is that they do different jobs. Applebot is the crawler; block it and your pages leave Spotlight, Siri and Safari search results on every Apple device, along with any chance of being pulled into a Siri AI answer. Applebot-Extended is not a crawler at all. Apple’s page is explicit: “Applebot-Extended does not crawl webpages.” It is a switch that controls whether pages Applebot already fetched can be used “to train Apple’s general purpose foundation models.” Disallowing it keeps you out of training data and changes nothing about discovery. Apple says so directly: pages that disallow Applebot-Extended “can still be included in search results,” and site rules for Applebot-Extended “are not considered in ranking for Search.”

That gives a site owner four real choices, and Apple’s documentation supports each one with a different mechanism. Here they are without the vendor spin.

There is one scenario where the fourth row is right: a page you never want summarized because the summary itself does damage, such as a pricing page with conditions a model tends to flatten, or a page that only makes sense with a login. For that page use nosnippet, not a crawler block, so the title can still be suggested. What you should not do is copy a robots.txt template off a tool blog that blocks every user agent with “bot” in its name. The training-versus-answers distinction is the same one we drew for OpenAI’s training and search crawlers in our piece on gated content, and it bites harder here because Applebot already inherits your Googlebot rules by default.

Does Siri AI send traffic to websites, and can you measure it?

Nobody outside Apple can answer the first half yet, and the second half is currently no. Apple has published no report, no dashboard and no Search Console equivalent for Siri, and the one sentence about source links says they “may” appear. ChatGPT at least shows up as a named referrer in your analytics. For Siri AI, as of the week before launch, there is nothing to log into and nothing to filter on.

What you can do is set up the measurement now so that the first real data lands in a place you will notice. Filter your server logs for the Applebot user agent and confirm with reverse DNS in the applebot.apple.com domain, which is the verification method Apple documents. Watch for crawl frequency changes on your most-answered pages after September 14. In analytics, isolate traffic from Safari on iOS 27 and macOS 27 with no referrer and a landing page that is not your homepage; that is the shape a tap on a Siri source link would most likely take if it arrives without a referrer string. And keep a short list of the ten questions your buyers ask most, so someone on the team with a supported device can type them into Spotlight during the first week and screenshot what comes back, including whether any sources are listed at all.

Early hands-on reports should temper expectations about the answers themselves. TechCrunch’s Sarah Perez, writing about the public beta on July 14, 2026, said the developer build handled web-style questions like local events and news better than the old Siri, but also “sometimes threw error messages or got confused,” and once responded to a request for news about Iran by searching her contacts for someone with that name. That is a beta. It will also be the beta that 2.5 billion devices’ worth of people form their first impression on, and the impression that matters to you is whether your company is in the answer or absent from it.

What should a marketing team do about Siri AI this quarter?

Do the two things that are actually available, watch the one thing that is worth watching, and refuse to pay for the rest. The two available things are the crawler decision and the snippet decision, and both take an afternoon.

First, open robots.txt and search for the string “Applebot.” If it is not there, find every rule that names Googlebot, because Apple has been applying those to its own crawler. Decide on purpose whether that is what you want. Most companies will want an explicit Allow for Applebot on everything they would want in a Google result, with a Disallow for Applebot-Extended only if legal has a position on model training. Second, audit the pages that carry nosnippet, noindex or an isAccessibleForFree property. A gated asset, a legacy paywall test or a copied schema block can quietly opt your best explainer page out of every Siri answer, and nothing will tell you.

Then look at whether the pages you would most want quoted are built to be quoted. Siri AI is one more retrieval system reading the same web, and the same properties that get a passage lifted into a Google AI Overview or a ChatGPT answer apply: a direct answer in the first two sentences under a question-shaped heading, a table where the fact is tabular, a dated figure with its source in the sentence. We laid out the page-level version in our guide to designing a website AI can quote, and none of it is Apple-specific, which is the point. The work you did for the other engines is the work that positions you here.

The thing worth watching is source links. If the beta ships them and they carry a referrer, Siri becomes a channel you can count, and it would be the first assistant to reach people who never open a browser to ask. If it ships them without a referrer, you are back to log analysis and brand-search lift. If it ships without them, Apple has built an answer engine with no way in for the sites it reads, and your only defense is being the brand the model already knows, which is a longer game that runs through the rest of your AI search program. The word Apple chose was “may.” We will know which one it meant within a few weeks of September 14.