98% of Quebec small businesses are missing from AI’s memory
We put to an AI the questions a customer types to find a plumber, an accountant or a restaurant in their town, for 300 Quebec small businesses. Of 226 usable readings, 222 businesses were named nowhere.
Who the AI names instead
When it does not know the local business, the AI falls back on the names it knows: big firms, national chains, franchises. A few answers from the study:
“best accountant in Québec City”
for a Québec City accounting firm: BDO, KPMG, PwC, Raymond Chabot Grant Thornton“best lawyer in Montréal”
for a Montréal immigration lawyer: Borden Ladner Gervais, Stikeman Elliott, Fasken“best running shop in Montréal”
for a shop open since 1985: Running Room, MEC“best financial adviser in Trois-Rivières”
for a local firm: Desjardins, BMO, RBC, Scotiabank“best fast food in Lévis”
for a neighbourhood snack bar: McDonald’s, Burger King, KFC, Subway
Questions were asked in French; translated here. The audited businesses are not named.
By sector
| Sector | Named nowhere |
|---|---|
| Construction and trades | 65 / 65 |
| Restaurants and food | 36 / 38 |
| Health and care | 26 / 27 |
| Professional services | 26 / 26 |
| Personal services | 22 / 22 |
| Automotive and moving | 19 / 20 |
| Other local services | 17 / 17 |
| Shops and retail | 11 / 11 |
By town
| Town asked about | Named nowhere | No answer at all |
|---|---|---|
| Montréal or Québec City | 60 / 63 | 0 |
| Other towns | 148 / 149 | 73 |
| No town (“in Quebec”) | 14 / 14 | 0 |
Outside the two big cities, the AI almost never names the business, and one time in three it simply knows no business in that trade there.
What it means for a small business
This study measures the model’s memory: what it knows without searching the web. ChatGPT, Perplexity and Google can also search live, and a business that is well represented online can then be found. But memory is where every answer starts, and it shows the gap: the big names are in it, small businesses are not.
Being found live depends on what the site and the listings tell machines. Yet 73% of the 124,769 small-business sites we audited do not even declare what kind of business they are. The 7 steps to get named.
And your business?
Tadaaah asks these questions to Google and the AI assistants and shows you who they name. Free, no account.
Analyze my siteMethodology
- Sample: 300 small businesses drawn at random (fixed seed) from 8,148 French-language Canadian sites audited by Tadaaah, excluding directories, public bodies, chains and franchises, at most 15% per sector; 29 sectors. It is not a draw of every business in Quebec.
- Questions: for each business, the four questions a customer types, in French, with its trade and town read off its site (“best electrician in Châteauguay”, “what are my options for…”, “where to find…”, “which… do you recommend”).
- Model: gpt-oss 120B (OpenAI, open-weight), asked without web search, temperature 0, on 27 October 2026. It measures a large model’s memory, not ChatGPT’s answers with live search.
- Reading: every name given was compared with the business, then checked by hand. 226 businesses got at least one usable answer; 73 got none (the AI named nobody) and 1 could not be classified. 95% confidence interval (Wilson): 95.5% to 99.3%.
- Limits: some names the AI gave instead of the businesses look invented; only names that clearly exist are quoted here. Another model or another date may answer differently.
Cite this study: Tadaaah, “98% of Quebec small businesses are missing from AI’s memory”, 27 October 2026, tadaaah.io/study/quebec-small-businesses-ai