Building keyword lists for English and French Canada

Statistics Canada's 2021 census puts English and French bilingualism in Canada at 18.0 percent, and in Quebec at 46.4 percent, against 47.3 percent who know French only and 5.3 percent who know English only. Those figures come from the census table on knowledge of official languages by province and territory.

Read the Quebec row as an operational fact rather than a demographic one. Close to half of that market can search in either language, which means language targeting does not split your audience into two groups. It splits your inventory into two sets of auctions that the same person can enter on different days.

That is why the common approach fails. Translating an English keyword list into French and running it in the same campaign produces a set of terms nobody searches, competing for budget with terms that work, in a structure where you cannot tell which half is which.

The version that works is unglamorous: two separately sourced keyword lists, two campaigns, two budgets, and reporting that keeps both readable. Everything below is the detail of that.

Two rows, two different problems. Outside Quebec the French-speaking market is real but small in most categories, and the decision is usually whether a French programme is worth running at all rather than how to run it.

Inside Quebec the decision is different, because the English-only share is 5.3 percent while the bilingual share is 46.4 percent. An English-only campaign in Quebec is not reaching 5.3 percent of the market. It is reaching whichever part of the 46.4 percent happened to search in English that day, which is an unstable and unmeasured slice.

Statistics Canada's release accompanying those figures notes that more than one in two people in Quebec could hold a conversation in English in 2021, the highest level recorded since the census began collecting this in 1901. The direction of travel matters for planning: this is not a static split you can set once.

None of which tells you which language a given customer will use for a given search. Nobody can tell you that, which is the argument for covering both properly rather than modelling the choice.

A translated keyword is a correct word that may not be a searched word. The gap shows up in three predictable places, and all three are findable without speaking the language well.

Register is the first. The formal term for a service and the term a customer types are often different, in both languages, and translation software reliably returns the formal one. Quebec French in particular has trade vocabulary that is neither the France French term nor the English one, and no dictionary lookup surfaces it.

Borrowed English is the second. In several trades the working vocabulary in Quebec includes English terms, so a keyword list that refuses to include an English word because the campaign is French will miss searches that French speakers actually make.

Accents are the third. People type without them constantly, and the accented and unaccented forms are not the same string. Broader match types do a lot of this work for you, because match type governs how closely a search has to match the keyword, but relying on it silently rather than deciding it is how a French exact-match campaign ends up with a fraction of the volume it should have. The precise close-variant behaviour is its own subject and worth reading before you commit to exact match in French.

So source the list rather than translate it. The client's own French website copy, their French-speaking staff, the way their French customers describe the problem on the phone, and a deliberately broad French campaign run for a few weeks purely to harvest real search terms. Google's own guidance on building a keyword list starts from how customers describe what you sell, and that instruction does not survive being applied to a translation of somebody else's list.

Language is a targeting criterion, and Google's API documentation lists it as positive targeting only, available at both campaign and ad group level. Because both levels are available, the question comes up in every build, and the answer is almost always campaign.

The reason is not matching, it is money and measurement. Two ad groups in one campaign share a budget and a bid strategy. If the French ad group is a tenth of the volume, it competes for budget against the English one every day and loses, and its results are averaged into a campaign number where nobody can see them.

Separate campaigns fix all three at once. The French programme gets a budget somebody decided on, a target appropriate to its own economics, and a line in the report that a client can point at. That is the same argument that applies whenever two things with different economics are sharing a budget, and structure is what sets the ceiling on what the account can return.

There is one honest counterargument. At very low volume, a separate French campaign may not gather enough conversion data to support an automated bid strategy, and splitting it out can make the learning problem worse. That is a real trade-off, and the answer is usually to run the French campaign on a simpler strategy rather than to hide it inside an English one.

French campaigns in Canada are usually smaller, and small campaigns produce thin data. A keyword with four conversions in a quarter tells you almost nothing, and the temptation is to judge it anyway because it is the only French keyword with any conversions at all.

Judge at the level where the sample supports a judgement. For most French programmes that is the campaign over a quarter, not the keyword over a month. Writing that rule down before the first review saves the argument where somebody proposes pausing the French campaign in week three.

Give the French programme its own targets from the start. Cost per acquisition in a smaller, less contested auction is frequently better than the English equivalent, and occasionally much worse, and neither result is visible if both are measured against one blended number.

For a multi-location business the two splits compound, because language and location are different axes and reporting has to hold both. That is the same discipline as reporting cost per acquisition per location rather than in aggregate, with one more dimension attached.

Which makes naming load-bearing rather than cosmetic. If language is not a parsable field in your campaign names, no report will ever separate the two programmes reliably, and a naming taxonomy that survives handovers is the cheapest thing on this list to get right at the start and the most expensive to retrofit.

A French ad pointing at an English landing page is a wasted click with a good click-through rate, which is the worst combination because it looks like the campaign is working. Check the destination before you check anything else.

Half-French pages are almost as bad. A translated hero section over an English form, English error messages, or an English confirmation email all break the experience at the point where the customer was about to convert, and none of it appears in the ads reporting.

Phone handling counts too. If the French campaign generates calls and the phone is answered in English, the conversion is recorded and the customer is lost, and your conversion data now says the French campaign works better than it does.

There are also French-language obligations attached to commercial communication in Quebec that are a legal question rather than a marketing one, and the right move is to get an answer from someone qualified rather than to infer it from what competitors do.

Separately, operating a programme aimed at Quebec residents brings privacy and consent obligations that have nothing to do with language and everything to do with how you collect data, which is its own operational workload and is worth settling before the campaigns launch rather than after.

Confirm the destination is genuinely French, end to end, including forms, confirmations and the phone. If it is not, stop. Everything else is spending money to deliver people to a page that will not serve them.

Harvest before you build. Run a small, deliberately broad French campaign against a short list of terms you are confident about, for long enough to collect real searches, and build the real list from what comes back rather than from what you predicted.

Split the campaigns, set language targeting at campaign level, and give the French campaign its own budget and its own target. Write both targets into the reporting definition on day one.

Then leave it alone longer than feels comfortable. A French campaign at a fifth of the English volume needs roughly five times as long to say anything, and most French programmes I have seen abandoned were abandoned on a sample too small to have meant anything.

If an account already runs both languages blended and nobody can say which is performing, that is not a keyword problem and no amount of list work fixes it. It is a structure and measurement problem, and an audit that reads structure and tracking before tactics is the shorter route to an answer.