The figures describe the businesses Kin-etiq.ai assessed. They are not a share of all businesses, all Canadian businesses or any industry. Kin-etiq.ai sells AI-visibility services.
What did the study find?
The main figures cover 305 businesses with saved AI answers and 617 with a usable website scan. They are sales prospects and the clients of one reseller partner. Where a figure says "ChatGPT", it means OpenAI's models called through the API (the engine behind ChatGPT, not the ChatGPT app). "Named" counts any mention, not only a recommendation.
- Of 305 businesses whose buyer questions were put to OpenAI's models through the API, and for 204 of them also to Claude, between 9 July 2026 and 7 October 2026, 198 (64.9%) were not named in a single answer.
- Counting only answers where the AI searched the web live, 165 of 251 businesses (65.7%) were never named.
- OpenAI's models called through the API (the engine behind ChatGPT, not the ChatGPT app) did not name 212 of the 305 businesses (69.5%) in any answer to questions about their own category. 545 of those 2,762 answers were made without a web search.
- Only 49 of the 305 businesses (16.1%) were named in at least half of the answers about their category.
- Of 266 businesses whose homepage was readable and open to the crawlers AI assistants use to answer, 165 (62.0%) were named in no answers, close to the 64.9% for all 305 businesses tested.
- Blocking AI crawlers was rare. 21 of 617 websites (3.4%) blocked at least one of eight common AI crawlers site-wide in robots.txt, and only 4 (0.6%) blocked one of the crawlers among those eight that ChatGPT or Perplexity uses to answer questions.
- 263 of 617 homepages (42.6%) had no JSON-LD structured data, as served before JavaScript, telling machines what the business is, such as an Organization, LocalBusiness or Dentist record.
- 125 of 617 homepages (20.3%) had no meta description, the short summary search engines can show for a page.
- 51 of 617 homepages (8.3%) showed fewer than 150 words before JavaScript ran. A browser test from 5 August 2026 to 7 August 2026 confirmed at least 10 websites that showed their content only after JavaScript ran. Crawlers that do not run JavaScript cannot see that content.
What else did the website scans show?
These counts come from the same 617 homepages, or from the 531 scans that recorded the page basics.
- 174 of 617 homepages (28.2%) had no JSON-LD structured data at all.
- 47 of 617 sites (7.6%) had no robots.txt that the scanner could find.
- 17 of 617 sites (2.8%) blocked only crawlers that collect training data, and 2 of 617 (0.3%) blocked all eight AI crawlers checked.
- 84 of 531 homepages (15.8%) had no Open Graph tags, and 71 of 531 (13.4%) had no canonical link.
- 6 of 531 homepages (1.1%) were set to noindex, which asks search engines to leave the page out.
- 4 of 617 homepages (0.6%) had no title tag.
Who was in the sample?
1,217 unique website domains appear in at least one figure. They came from three sources, and the main figures use the first two.
- Sales prospects: 593 scanned domains and 283 businesses with saved AI answers. Almost all came from one business-owner association's member contact list.
- Client sites assessed for one reseller partner: 24 scanned and 22 with saved AI answers. They are reported only in total.
- AI-sourced businesses: 584 scanned and 334 with saved AI answers. Kin-etiq.ai first found them because an AI answer named them for another business's question.
The main scope holds 633 domains: 617 with a usable scan, 305 with saved AI answers and 290 with both. 215 of the 617 scanned domains end in .ca. Of the 305 businesses with answers, the business's own site placed 157 in Canada and 8 in the United States. The other 140 have no location recorded.
AI-sourced businesses are reported apart and are not compared with the main figures. They were picked because an engine had already named them, and their websites were matched from the business name before a name-matching fix on 10 September 2026.
How was it measured?
AI answers. For each business, Kin-etiq.ai wrote buyer questions about its category and, for most, its area. It put them to OpenAI's models through the API, and for 204 of the 305 businesses in the main figures also to Anthropic's Claude models. Gemini answered a handful. Each question was usually asked twice per engine, and every answer is stored word for word with a timestamp. In the main figures, the 305 businesses have 4,494 answers between them, a median of 15 each from a median of 5 questions. Questions that contained the business's own name were left out.
In the main figures, 545 of 2,762 answers from OpenAI's models and 1,671 of 1,683 Claude answers carry no record of a web search. The web-search figure counts only answers whose record proves a live search. July answers from Claude Opus carry no search marker, so they are left out of that figure. Counting them as searches changes 165 of 251 to 159 of 251.
The name rule. A business counts as named only on one of three matches, all on whole words: a form of its name of two or more words, a single distinctive word that is the first word of its name and part of its own domain, or its own domain in the answer or its sources. A generic single word never counts.
Websites. Kin-etiq.ai's scanner fetched each homepage without running JavaScript, read robots.txt and saved what it found for every site. The figures come from that saved record. Each scan is dated by the file that holds it, from 30 June 2026 to 7 October 2026.
Not checked. Sites the scanner could not read, and captures whose questions were malformed or built on a wrong category, are marked "not checked" and are never counted as a failure. That covers 40 scanned domains with no usable scan and 84 domains whose every fetch failed.
What are the limits?
These are businesses chosen for outreach, not a random sample: almost all outreach prospects came from one business-owner association's member contact list, plus 24 client sites of one reseller partner. They should be read as "businesses Kin-etiq.ai assessed", never as a share of all businesses, all Canadian businesses or any industry.
Questions came from each business's category, set by hand in older captures and read from the site in newer ones; many older questions name no place, and answers to those can list firms from other countries. The engines' models and methods changed over the period. A different question set gives different counts. The test measures whether these answers named the business on the test dates, not market standing.
The strict name rule can still miss a mention under a very different name, and the site-match exclusion removes some legitimate businesses. "Not checked" covers 40 scanned domains with no usable scan, 84 domains whose every fetch failed, and every capture left out; none is counted as a failure.
Website checks read the homepage only, before JavaScript, as served to Kin-etiq.ai's scanner. A firewall can treat AI crawlers differently; that was not measured reliably. A missing robots.txt was not told apart from one the server refused to send.
Industry splits are not given: categories are free text with no checked coding, so no group of 30 or more could be formed without guessing.
What does this data not show?
- Whether a fix such as structured data, robots.txt or a meta description causes an AI to name a business. These are counts side by side.
- Claude against OpenAI's models, Canada against the United States, or prospects against AI-sourced businesses. The groups were picked or tested differently.
- Which firms the AI named instead, or how often.
- Letter grades or an average readiness score.
- Results from the ChatGPT app itself, which can differ from OpenAI's models called through the API.
How to cite this research
Kin-etiq.ai AI visibility dataset, built 21 October 2026. AI answers 9 July 2026 to 7 October 2026; website scans 30 June 2026 to 7 October 2026. Published at kin-etiq.ai/research/ai-visibility-study.
What does this mean for a business?
In this group, most businesses were not named in a single answer. Businesses whose homepage was readable and open to the crawlers AI assistants use to answer were named about as rarely as the rest. The AI visibility audit runs the same kind of test on one business's buyer questions and reads the pages the engines cited. The answer on why ChatGPT does not recommend a business covers what that test looks for, and generative engine optimization is the ongoing work.