Your AI Speaks English. Your Customers Speak French.

Your AI Speaks English. Your Customers Speak French

Most Quebec business owners I meet use AI the same way. They type their prompt in English. The AI answers in French. They copy it, paste it, send it.

Nobody re-reads it. That is the mistake.

Your customers read every word. They notice when your quote says "mail" instead of "courriel." They notice when your ad sounds like it was written in Paris. It costs you nothing to fix, and it costs you credibility every time you skip it.

So I tested it. For real.

The test I ran this morning

On October 13, 2026, I gave four typical Quebec business tasks to three major AI models: ChatGPT, Google Gemini, and Muse (Meta). Same four prompts, word for word, in English, each one asking explicitly for Quebec French.

The four tasks: a customer complaint reply email, the wording of a renovation quote, a short Facebook ad for a plumber, and a password reset support answer.

I scored each answer from 1 to 5 on Quebec French quality: québécois vocabulary versus France French, anglicisms, tone, and formality. Two more models never made it into the scoring. Meta AI was blocked by a network policy before I could even open it. Perplexity answered one prompt as a guest, then locked everything behind an account. I do not score incomplete tests, and I do not invent scores.

The scores

ChatGPT: 17 out of 20. Email 3, quote 5, ad 4, support 5.

Gemini: 17 out of 20. Email 3, quote 4, ad 5, support 5.

Muse: 16 out of 20. Email 3, quote 4, ad 4, support 5.

Three models, same pattern. Read that twice.

What the results actually mean

The good news first. When you ask explicitly for Quebec French, all three models deliver real québécois vocabulary. Not Paris French with a maple leaf on top.

The renovation quotes were the proof. Both ChatGPT and Gemini wrote "vanité" and "toilette," not the France French equivalents. Both included TPS and TVQ tax lines. Gemini even used the exact TVQ rate of 9.975 percent and wrote "tirage de joints" and "gypse," which is what drywall is actually called on a Quebec jobsite. ChatGPT added the RBQ licence line. These are documents you could almost send to a client.

The support answers were just as strong. All three models wrote "courriel," "pourriels," and "ça nous fera plaisir." That is exactly how a Quebec support team writes.

Now the weak spot. All three models scored 3 out of 5 on the complaint email. Every reply was correct and professional. Every reply was also neutral international French. Zero québécois character. Nothing wrong, nothing local. A customer in Gatineau would read it and feel nothing.

The ads landed in the middle. Good vocabulary ("fournaise," "89 $ + taxes"), but stuffed with emojis and written like every other Facebook ad on the internet.

The one habit that fixes most of it

Here is what the test really proved. The models know Quebec French. What they do not know is when you want it.

Every prompt in my test said "in Quebec French" explicitly. That one phrase is doing most of the work. Without it, you get France French by default, because that is most of the French on the internet.

So the habit is simple. Always write "en français québécois" in your prompt. Then re-read the output before it reaches a customer. Two minutes. Look for three things: "mail" instead of "courriel," "week-end" instead of "fin de semaine," and any sentence that sounds like it was translated from English word by word.

The models are not the problem. Copy-paste without review is the problem.

Your customers speak French. Make sure your AI speaks theirs.

If you want help putting AI to work in your actual processes, in French that sounds like your business, that is what I do. Neurotek AI builds automation systems and AI agents for Quebec SMBs. My discovery conference reviews your processes and shows you exactly where AI pays for itself. Write to me and we will set it up.