# What 26 buyer questions in ChatGPT told us about Berlitz Canada

> Field notes from a live programme. We asked ChatGPT 26 questions a language learner in Canada would ask before choosing a school. Berlitz Canada was named in 5 answers and cited in 4, and only one answer did both. Here is what the baseline showed, what we built from it, and how to run the same check on your own brand.

**Héloïse Mazingarbe**, GEO Strategist, CrawlerPal · November 13, 2026 · Field notes

Canonical: https://crawlerpal.com/blog/berlitz-canada-chatgpt-baseline-field-notes

---

We asked ChatGPT 26 questions a language learner in Canada would ask before choosing a school. Berlitz Canada was named in 5 answers and cited as a source in 4. Only one answer did both. Here is what that baseline showed, what we built from it, and how to run the same check on your own brand.

## What we measured

In May 2026, before the comparison content for Berlitz Canada went live, we wrote down 28 prompts: 26 that do not contain the brand name, and 2 head-to-head comparisons that do. We ran each one once in ChatGPT, in a temporary chat, in English. For every answer we logged four things: whether Berlitz was named, whether a Berlitz URL sat in the sources, which schools were recommended instead, and which pages the answer leaned on.

The two branded prompts are left out of every figure below, because [a prompt that contains your name flatters the score](/blog/branded-vs-unbranded-ai-visibility). Read the rest with its limit in mind: one engine, one run per prompt. That is enough to steer the work. It is not a market share figure.

## What did the baseline show?

Berlitz Canada was named in 5 of 26 answers and cited in 4. The split by type of question matters more than the total.

| Type of question | Prompts | Berlitz named | Berlitz cited |
| --- | --- | --- | --- |
| Best school or class in a city | 9 | 3 | 1 |
| Exam preparation (TEF, TEFAQ, SLE) | 6 | 1 | 2 |
| Work and corporate training | 6 | 1 | 1 |
| Immigration and newcomers | 4 | 0 | 0 |
| Alternative to a language app | 1 | 0 | 0 |
| All unbranded prompts | 26 | 5 (19%) | 4 (15%) |

Source: CrawlerPal baseline for Berlitz Canada, ChatGPT, one run per prompt, May 2026.

## Three things the sheet made obvious

**Being cited and being named are separate events.** Only 1 of the 26 answers did both. In three answers ChatGPT read a Berlitz page, listed it eighth, ninth or last among its sources, and recommended other schools. In four answers it named Berlitz with no Berlitz page in the sources at all. We track the two in [separate columns](/blog/citation-vs-mention-measuring-geo) for this reason.

**The home market was the blind spot.** Six prompts were about Montreal or Quebec: the best French schools, TEF preparation, TEFAQ, French training for Bill 96. Berlitz was named in none of them. It was named in Vancouver, Edmonton and Calgary. A large brand can be absent from the market where it has the most to lose.

**The specific page beat the big name.** Across the three TEF and TEFAQ prompts, the answers went to providers with a programme built for that exam. One of the Berlitz pages that did get cited was an article written for exactly the question asked, French courses in Ottawa for government employees.

## What did we build from it?

The baseline became three working lists, in this order.

1.  **A keyword map.** 49 bottom-of-funnel queries, each tied to one target URL and one action: 26 new articles to write, 23 existing pages to rework. Montreal and exam preparation came first, because that is where the zeros were.
2.  **A source list.** The directories and comparison pages ChatGPT leaned on. Of the first four directories we logged, Berlitz Canada was listed on one.
3.  **A conversation list.** 15 Reddit threads that showed up in the sources, where a useful answer from the school belongs.

The reasoning behind that order is in [our piece on bottom-of-funnel pages](/blog/bottom-of-funnel-geo): engines recommend from comparisons and buying guides, so those get written first.

## What has moved so far?

The comparison articles are live and they earn search traffic. Berlitz Canada's September 2026 monthly report, read from Search Console, counts 115 Google clicks in three months to the comparison articles written for AI answers. Magic Pulpo, our parent studio, publishes that figure in [its case study](https://magicpulpo.com/works/berlitz-ca-seo-blog).

That is search, and it undercounts. The measure that settles it is the same 26 prompts, run again in the same conditions. We will publish that re-test here once Berlitz has signed off the figures.

**Same method, three programmes.** This is the first of a series. The same sequence is running for Berlitz UK and for a French consumer brand that prefers not to be named: baseline, keyword map, comparison content, source outreach, re-test.

## What this means for you

Write 20 to 30 questions your buyer asks before choosing, without your name in them, and include your home city. Run them in a clean session and log named and cited in two columns. [The one-hour method is here](/blog/check-if-your-brand-appears-in-chatgpt).

Then look for two patterns. Rows where you are cited and not named are pages the engine reads and passes over. Groups where you score zero are questions nobody has written your answer to. Start with the group closest to revenue.

## Frequently asked questions

### Why only ChatGPT, and why one run per prompt?

Because this was a first baseline to decide where to work, and ChatGPT carries most of the consumer traffic for this category. One run on one engine is enough to see which whole groups of questions return nothing. It is too thin to report as a market share. A Sprint diagnostic samples four engines, three runs each.

### Why are the two Berlitz comparison prompts left out of the figures?

A prompt that already contains the brand name will almost always return the brand, so it flatters the score. We ran the two head-to-head comparisons to read what ChatGPT says about Berlitz, and we kept them out of the 26. The number worth tracking is the unbranded one.

### Is being named in 19% of answers good or bad?

On its own the figure says little, because there is no benchmark for this exact set of questions. What matters is where the zeros are and whether the same set moves on the next run. Here the zeros sat in the home market and in exam preparation, which told us where to start.

### Which clients does CrawlerPal run this for?

Four clients have paid for this work: Berlitz Canada in English and French, Berlitz UK, the International Academy of Football, and a French consumer brand that prefers not to be named. The method is the same each time, from baseline to re-test. Notes are published once a client has signed off the figures.

---

CrawlerPal · https://crawlerpal.com
