We Ran 26 Siri AI Queries on Launch Day. Here's What Actually Gets Cited.

First-party data from the first 48 hours of Siri AI: what response types fire, which domains get cited, and what it means for Canadian agencies.

On 14 September 2026, Apple began rolling out Siri AI in English beta on iOS 27. We published what the launch means for agency GEO work the same day. This piece is what we found when we stopped reading the announcements and ran the queries.

Across 26 queries on an Apple Intelligence-eligible iPhone in English (Canada), our study returned zero AI-generated paragraph answers. Siri AI routed. It cited. It opened Maps. It did not synthesise. That is the working brief for Canadian agencies this week. It is directional first-party data from a small sample, not a census of the product.

How we ran the study

Between 14 and 15 September 2026, the AI Search Rank research team ran 26 queries on an iPhone eligible for Apple Intelligence, on iOS 27, with the device language set to English (Canada). The device was physically on Hornby Island, British Columbia. Query categories covered home services, legal, healthcare, financial, food and dining, digital marketing, real estate, travel, e-commerce, branded SaaS, local Canadian, and comparison.

This is first-party proprietary research. Twenty-six queries is a directional sample. It is not a statistically definitive study of Siri AI. Rural GPS location almost certainly shaped the Maps results. Urban agencies should treat the distances below as a warning about location bias, not as a prediction of downtown Vancouver behaviour.

Siri AI did not generate answers

The largest surprise in our study was the simplest: 0 of 26 queries returned an AI-generated paragraph answer.

Apple's 8 June 2026 Newsroom announcement describes Siri AI as powered by Apple Intelligence and able to help users by "answering questions from the web on virtually any topic." Launch-day coverage treated that language as a ChatGPT-style generative answer engine. Our device observations did not match that expectation.

In our data, Siri AI behaved as a routing and citation surface. Across 26 queries, we recorded four response types:

• Apple Maps: 14 of 26 (54%)
• Web link list: 10 of 26 (38%)
• App Store: 1 of 26 (4%)
• Direct URL load: 1 of 26 (4%)

The App Store result was the query "Slack app download." We did not observe a synthesised prose answer on any query in the set. If a GEO playbook assumes Siri will lift a paragraph from a blog post the way Perplexity or ChatGPT often does, our study says that playbook does not describe launch-week Siri AI on this device, locale, and query mix.

Apple Maps owned 54% of our results

Apple Maps owned 14 of 26 queries (54%) in our study. Every query we classified as local or transactional routed to Maps, not to web results. Yelp ratings surfaced directly in Maps cards.

Listing quality varied sharply. For "near me" queries run from rural British Columbia, our study returned results 139 km and 3,466 km from the device. That is not a round-the-corner local pack. It is an index matching whatever it can find, at distances that would be unusable for a client who thinks "near me" means their town.

Web GEO work does not fire if Siri never opens the web. For local and transactional intent, the Siri AI surface we observed is Apple Maps listing quality: name, category, location, reviews. Agencies still selling a Siri blog post for plumber, restaurant, intake, and "near me" queries are, on this evidence, working the wrong inventory.

GPS overrode every city name we tested

In our study, GPS location overrode explicit city names in all four queries we designed to test that behaviour.

• "best real estate agent Kelowna" returned RE/MAX Denman Hornby, 11 km from the device.
• "best poutine Montreal" returned a restaurant named Montreal Poutine, 3,790 km from the device. It was the nearest result with "Montreal" in the name, not a Montreal restaurant.
• "yoga studio Kelowna" returned Body And Soul Therapy in Courtenay, B.C., 29 km from the device.
• "best SEO agency Vancouver" returned a Best SEO Vancouver listing 115 km from the device.

City names in those queries had no observable effect on Apple Maps ranking relative to GPS. Every local query in this set behaved as a "near me" query regardless of phrasing.

The implication is operational. Apple Maps listing optimisation for a business's actual location matters more than the city an agency targeted in content. If you are testing Siri for a Kelowna client from an office in Vancouver, or from a cottage on Hornby Island, you are not testing the client's market. You are testing your GPS.

GEO does not exist in Siri's index yet

Generative engine optimisation (GEO) is the practice of earning citations in AI answer engines. In our study, that category does not exist yet as a marketing organisation in Siri's business index.

The query "GEO agency Canada" returned Geo Tours (a travel agency, 3,776 km away), Di-Geo International, and Geo Escapade. Siri matched "Geo" as a string in business names, not as a marketing acronym.

"AI search optimisation agency" returned flow-agency.com (positioned around LLM optimisation), optimizon.co.uk (an Amazon agency), and a Semrush agency directory. "What is GEO in marketing" returned reddit.com/DigitalMarketing (December 2025), contentful.com (June 2025), and adsmurai.com (October 2025). Our own GEO explainers on aisearchrank.ai were not cited. resources.aisearchrank.ai did not appear in any of the 26 results.

That is a first-mover gap, not a compliment. The citations that did appear are travel companies, adjacent agencies, and older definitional posts on other domains. Agencies that publish clear, entity-tied GEO and AEO definitions now are writing into an index that, in our sample, has not yet attached the acronym to a marketing firm.

Entity clarity failed on our own brand

The same index problem showed up on our own name. The query "AI Search Rank" cited rankscale.ai first, reddit.com/SaaS second, and useomnia.com third. aisearchrank.ai was not cited for its own brand name.

Our working diagnosis is weak entity signals: no Wikidata entry, Schema.org Organization markup not confirmed, and the brand name not yet strongly associated with the domain in Apple's index. We are stating that as our own gap, observed in this study, not as a general law about every brand.

The contrast in our set was branded SaaS with cleaner entity resolution. Queries for Hootsuite and FreshBooks returned own-domain results exclusively: two to three results, all from the brand's domain. FreshBooks returned an /en-ca subdirectory, which is evidence that Canadian locale is active for branded queries. "Ferry schedule BC Ferries" cited bcferries.com three times. The official domain swept that branded informational query.

Two other patterns were consistent enough to report. Reddit appeared as the third result for informational and branded pricing queries in our set. Directory sites (canadianlawyermag.com, bestlawyers.com, Lexpert) dominated "best X in Canada" legal queries. Individual firm websites were not cited in those legal results.

Entity clarity is the first Siri AI priority we would put on an agency checklist this week, because none of the content work matters if the index cannot attach the name to the domain.

What agencies should do this week

These actions follow the 26 queries. They are not a generic GEO checklist.

1. Split the roster by intent. Local and transactional clients need Apple Maps listing work (category, NAP consistency, reviews), not a Siri article. Our study sent those queries to Maps 14 times out of 26.
2. Run Siri tests from the client's actual location, or from a device location that matches it. City names in the query did not override GPS in any of the four tests we ran.
3. Claim the GEO and AEO category with definitional pages tied to a real organisation, not to the letters G-E-O. Siri treated "GEO agency" as a travel-company name match in our study.
4. Fix entity clarity before you brief more content: Wikidata, Schema.org Organization markup, identical brand name across the site and major profiles. We failed our own brand query. Hootsuite and FreshBooks did not.
5. Go where Siri already cites. For "best X in Canada" legal queries, that meant directories, not firm sites. For branded informational queries, that meant the official domain. BC Ferries did not split citation with a blog roundup in our sample; bcferries.com took all three slots.

If you want a baseline on the engines we audit today (ChatGPT, Perplexity, Claude, and Gemini), start with a free AI visibility audit. Siri AI monitoring is in development. The sequence we use for a first cross-engine pass is the same one in how to run your first AI search audit: build a prompt set, record who gets cited, then fix entity before you fix copy.

Siri AI is a new citation surface. In the first 48 hours, on 26 queries, it did not look like a generative answer engine. It looked like Maps, link lists, and entity matching. Agencies that spend this week writing Siri-optimised articles for local intent are, on this evidence, writing for a response type we did not observe.

Study methodology and limitations

This article reports first-party proprietary observations from AI Search Rank's Siri AI citation study, conducted 14-15 September 2026, the first two days of Siri AI's public availability.

• Queries: 26
• Device: iPhone (Apple Intelligence-eligible)
• OS: iOS 27 (launch day and day 2)
• Locale: English (Canada)
• Location: Hornby Island, British Columbia, Canada
• Dates: 14-15 September 2026
• Query mix: home services, legal, healthcare, financial, food and dining, digital marketing, real estate, travel, e-commerce, branded SaaS, local Canadian, comparison

Apple Newsroom (14 September 2026) states that Siri AI is rolling out in beta in English on iOS 27, iPadOS 27, macOS 27, watchOS 27, and visionOS 27, and that certain server-side Apple Intelligence features, including Siri AI, are subject to daily usage limits. Hardware eligibility for Apple Intelligence and Siri AI includes iPhone 16 models or later, iPhone 15 Pro, and iPhone 15 Pro Max, per Apple Newsroom (8 June 2026) and the 14 September update.

Limitations: 26 queries is a small sample. Findings are directional observations from one device, one rural GPS location, one locale, and a two-day window. They are not independently replicated, not a statistical census of Siri AI, and not a claim about how the product will behave in urban markets, other languages, or later software builds. Percentages describe this set of 26 only (14 Maps, 10 web link lists, 1 App Store, 1 direct URL). We did not observe an AI-generated paragraph answer in this set; we are not asserting that Siri AI can never produce one.