AI Search Optimization: How Canadian Businesses Get Recommended by ChatGPT, Gemini and Perplexity

What the Canadian numbers actually say

CIRA's survey found roughly 46 percent of Canadians used a generative AI tool in the past year, up from 33 percent in 2025 and 16 percent in 2024, and 52 percent of those users say they use it as a search engine. (Canadian Internet Registration Authority CIRA) No Canadian survey measures how many people ask an assistant to find a local business, so treat it as a proxy, not proven demand in your category.

The click effect is better measured south of the border, where Pew Research Center tracked 68,879 Google searches by more than 900 consenting American adults in March 2025. On searches producing an AI summary, people clicked a traditional result 8 percent of the time against 15 percent without one. (Pew Research Center) That is US panel data; no Canadian equivalent exists.

Industry datasets agree on direction and argue about size. Seer measured 0.61 percent click-through with an AI Overview present against 1.62 percent without, across 5.47 million queries; Ahrefs measured roughly a 58 percent drop for the top organic position. Seer's own figure then rebounded from 1.3 percent in December 2025 to 2.4 percent in February 2026, so treat any single compression number as a snapshot. AI Overviews appeared on roughly 48 percent of queries in early 2026 by US and global tracking, skewed heavily to informational queries and thinner on local ones.

How AI assistants actually choose a business

Two mechanisms decide whether your name comes up, and they answer to different work.

Mechanism	What it is	What moves it	How fast it moves
Recall	Model names you from training data	Spread of brand mentions	Slow, tied to training cycles
Retrieval	Searches at answer time, cites what it finds	Crawlability, indexation, snippet eligibility	Days after a recrawl

AI Overviews and AI Mode ground on the ordinary Search index Googlebot builds, plus Google Business Profile and Merchant Center feeds. AI Mode uses query fan-out, turning one question into many background searches, so you can be cited for a sub-question you never targeted. ChatGPT Search and Perplexity retrieve from their own crawled and licensed indexes plus review platforms and directories; Claude and Copilot ground on live results at answer time. Retrieval is the lever you move this quarter. Recall is a multi-year brand project.

Grade every recommendation before you pay for it

Tier 1 is platform-documented: crawler access, indexation, snippet eligibility, Google Business Profile completeness, entity consistency. Google's AI optimization guidance states that no special schema.org markup is required for AI Overviews or AI Mode, and that ordinary search fundamentals are the requirement. (Google Search Central) Fundamentals are not ranking, though: measured overlap between AI Overview citations and the organic top ten fell from roughly 75 percent in late 2024 to between 17 and 38 percent by early 2026.

Tier 2 is correlated but unproven: brand mentions, review volume and recency, content structure, freshness. Tier 3 is speculative or disproven: llms.txt, FAQ schema as an AI signal, and most claims beginning "GEO schema". Ask any agency which tier each line of its proposal sits in, as our guide to choosing an agency sets out.

Three things you have been told to do that do not work

Publishing an llms.txt file

llms.txt is a proposed convention, not an adopted standard, with no W3C or IETF backing and no major provider committed to using it. Google's Gary Illyes said in July 2025 that Google does not support it and has no plans to, John Mueller compared it to the old keywords meta tag, and Google's guidance now states that Search ignores the file. (Google Search Central) An SE Ranking scan of about 300,000 domains found roughly 10 percent had published one, while reported log analysis found only 408 of more than 500 million AI bot visits over 90 days touched it. Adoption exists. Consumption does not.

Adding FAQPage schema for rich results

Google restricted FAQ rich results to well-known government and health sites in August 2023, then retired them from Search on 7 May 2026, dropping the Search Console report that June and API support that August. (Google Search Central) (Google Search Central Blog) FAQPage remains valid schema.org and unused markup is not penalized, but it produces no rich result and is not a known AI citation factor. Keep the FAQ content and question-style headings, which are easy for a model to quote. The markup itself is optional.

Blocking Google-Extended to stay out of AI Overviews

Google-Extended is not a crawler. It is a robots.txt token governing whether content Googlebot already fetched may be used for Gemini training and grounding, and never appears in your logs. It has no effect on AI Overviews or AI Mode, which are served from the standard Search index, and there is no AI Overviews user agent to block. Blocking it will not remove you and unblocking it will not get you in. The only real exits are blocking Googlebot or using nosnippet, max-snippet or data-nosnippet, which cost ordinary search visibility.

Which AI bots to allow, and which you can block

Token	Type	Effect of blocking	Default
Googlebot	Retrieval	Out of Search and AI answers	Allow
OAI-SearchBot, ChatGPT-User	Retrieval	Out of ChatGPT answers	Allow
Claude-SearchBot, Claude-User	Retrieval	Out of Claude answers	Allow
PerplexityBot, Perplexity-User	Retrieval	Out of Perplexity answers	Allow
GPTBot, ClaudeBot	Training	Little citation downside	Your call
Google-Extended, Applebot-Extended	Training token	No effect on retrieval	Your call
CCBot	Training corpus	Weakens long-term recall	Your call
Bytespider, meta-externalagent	Training	Little citation downside	Your call

Blocking a retrieval bot removes you from that assistant's answers, almost always by accident. Blocking pure training crawlers is a legitimate business choice with little measured citation cost. One caveat: OpenAI and Perplexity document their user-initiated fetchers as tools that may not follow robots.txt, so a disallow there is a request, not a guarantee.

Be an unambiguous entity

An assistant must be certain which business you are before it names you, and in most audits we run the sources disagree.

• Use one name, address and phone number across your site, Google Business Profile, directories and social profiles.
• Publish an About page stating what you do, where you do it and who you serve.
• Add Organization and LocalBusiness schema with sameAs links to every official profile.
• List in authoritative Canadian sources: Yelp.ca, the Better Business Bureau, Canada411, your provincial association and your local chamber of commerce.

Wikidata is the principal feed for Google's Knowledge Graph and Wikipedia sits among the most-cited domains on every assistant, but notability rules exclude most small businesses. The list above is the realistic version, and it belongs with your local SEO.

Build local proof, because AI names one to three businesses

The classic map pack shows three results. AI local answers often name one to three and stop, so the concentration is sharper than you are used to, and Google Business Profile completeness, review volume and recency feed those answers directly.

Yelp reportedly drew 512,680 AI citations in Q4 2025, about 3.4 times the second-place Better Business Bureau, and OpenAI signed a deal reported in July 2026 bringing Yelp reviews into ChatGPT's local answers. Reviews naming a service and a place, such as "installed our furnace in Airdrie within 24 hours", give a model something specific to repeat. Ask for them systematically, as our review guide sets out.

Earn citations, not just links

Ahrefs studied roughly 75,000 brands and found AI visibility correlated with YouTube mentions at 0.737, unlinked web mentions at 0.664 and total backlinks at only 0.218. Those are correlations on observational data, not proof of cause, and backlinks still matter as the threshold for being indexed at all. An unlinked mention in a publication a model already trusts can beat a link from a site it has never cited.

Citations are concentrated to an uncomfortable degree. Reddit, Wikipedia, YouTube, LinkedIn and a short list of large publishers carry most of them, with a reported 15 domains accounting for about 68 percent, and Reddit's share swings hard by engine and by month. A study of 557 AI answers on Canadian policy questions found CBC News cited in 16.5 percent and The Hub in 12.2 percent, ahead of the Globe and Mail and CTV News. That scope was policy, not consumer services, so aim Canadian digital PR with it, not a media plan.

Canadian specifics most guides skip

If you serve Quebec, one rule outranks every tactic here. Under the reformed Charter of the French Language, since 1 June 2025 any commercial website reachable by Quebec residents must present French at least as prominently as any other language, at any company size. OQLF penalties are reported in the CAD 3,000 to CAD 30,000 range per violation, employers with 25 or more Quebec staff for six months must register for francization, and the OQLF has cautioned against machine translation. Our Bill 96 guide has the detail and our Quebec page covers sequencing.

The AI picture in French differs again, and is worth rechecking before you plan around it. Reporting citing Google Canada indicates AI Overviews still do not run in French, while AI Mode added French support in October 2025. If that split holds, French queries in Quebec are less click-compressed than their English equivalents, which argues for finishing French compliance and French SEO first.

Measure what you can, say plainly what you cannot

On 13 May 2026 Google added a native AI Assistant channel to GA4's Default Channel Group, tagging qualifying sessions with the medium ai-assistant. (Google Analytics Help) Google's documented examples include ChatGPT, Gemini, Microsoft Copilot, Claude and Perplexity, so those referrals are grouped automatically; a custom channel group is still worth building to split the channel by assistant and to catch any newcomer Google has not yet added. The old chat.openai.com domain is legacy.

Search Console added generative AI performance reporting in June 2026, fully rolled out by the end of August. It shows impressions, pages, countries and devices, with no clicks, no click-through rate and no query breakdown, so anyone promising AI click-through reporting today is estimating.

What does work is duller. Ask each assistant the same 20 to 30 buying questions once a month, log who was named and cited, and watch the trend rather than any single answer, while tracking branded search volume and brand mentions. Similarweb's vendor estimate puts AI-referred sessions at roughly 0.9 percent of web traffic, converting well above classic organic on figures that are vendor-reported, not audited.

Your first 90 days

1. Confirm no retrieval bot is blocked in robots.txt and that key pages are snippet-eligible, including stray nosnippet tags left by a redesign.
2. Complete and actively maintain your Google Business Profile: categories, services, hours, service areas, photos and posts.
3. Fix entity consistency and ship Organization and LocalBusiness schema with sameAs links.
4. Build reviews on Google first, then Yelp.ca and the Better Business Bureau, asking for detail that names the service and the city.
5. Rewrite your top ten pages answer-first, dated and author-attributed, with prices, timelines and service areas as quotable facts.
6. Pursue unlinked brand mentions in Canadian publications, trade press and relevant subreddits.
7. If you serve Quebec, bring the French site into compliance using a human translator.

Steps 1 to 3 take about a week. Steps 4 to 6 never finish. None of it sits outside ordinary search work, and the local SEO checklist still applies.

What changed in this update

Updated 19 September 2026. This revision dropped llms.txt from the priority list, corrected the FAQ schema and Google-Extended guidance, split training crawlers from retrieval crawlers, and added Canadian figures, Quebec compliance and a rebuilt measurement section. Every number here moved within twelve months, which is why the page carries a date.

Where Better Businesses fits

We do this work for Canadian businesses from Toronto to every province, and we tell you which tier each recommendation sits in before you approve it. If you want to know whether ChatGPT, Gemini or Perplexity names you or your competitors today, we will run the prompt set and show you the output first. Talk to us and we will start there.

Sources

1. 2026 Canadian Internet Trends – CIRA — Canadian Internet Registration Authority (CIRA)
2. Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center
3. AI Features and Your Website | Google Search Central | Documentation | Google for Developers — Google Search Central
4. Google's Guide to Optimizing for Generative AI Features on Google Search | Google Search Central | Documentati — Google Search Central
5. Changes to HowTo and FAQ rich results | Google Search Central Blog | Google for Developers — Google Search Central Blog
6. Latest Google Search Documentation Updates | Google Search Central | What's new | Google for Developers — Google Search Central
7. Default channel group - Analytics Help — Google Analytics Help

Sources checked September 2026.