A Montreal company is rarely searched by one kind of person. The same firm turns up in an English query typed in Los Angeles, a French query typed in Longueuil, and a third query nobody in the office would have predicted — and a single report averages all three into one line.
A verified property hands you four figures a day: how often you were shown, how often somebody clicked, the ratio between them, and roughly where you sat. None arrives with an instruction attached. Turning them into one means knowing which source produced the figure, which audience it came from, and which language that audience types in.
Two sources do most of the work and they are not interchangeable. A panel holding both removes the reconciliation, but not the reading — that part stays yours.
Two instruments, and the blind spot built into each
Search Console reports on a property you have verified as yours. It logs the searches that put your domain in front of a person, the URL shown, and whether that person clicked. What it cannot see is structural: a search where you never appeared leaves no trace. If a Paris studio owns the first page for a term your Mile End shop wants, the view stays empty — as far as your property knows, nothing occurred.
Rank tracking stands outside the property. A query goes out from a set location with no personal history behind it, and whatever comes back is recorded — your domain somewhere in the list, or nowhere in it. That gives you the full field, competitors you have never heard of included, and nothing about clicks.
| The question | Search Console | Rank tracking |
|---|---|---|
| Did anyone click through | Broken out by search, URL, country and device | Not measured, ever |
| Who else occupies the first page | Silent about every domain except yours | Rival domains, their authority, the terms in common |
| Which URL was shown to the searcher | Listed in the pages view | Only URLs that place well enough to appear |
| Searches you have never once appeared for | Cannot exist in the data | Reported as soon as you add the term |
In the panel that arrives as eight views on the Search Console side and six on the ranking side. Treat them as specialised instruments: each is precise about one thing and vague about the rest. A post-production house wanting to know why enquiries dried up looks at pages; the same house asking who replaced it looks at competing domains, which no Search Console report supplies.
Fourteen views, two vantage points
What each module covers, so you stop asking one for the other's answer.
- Search Console, eight views. A dashboard across every verified site, then site overview, keywords with position history, pages, devices, countries and traffic.
- Keyword dynamics. Which terms entered or fell out of the top three, ten and thirty — movement recorded before the traffic reflects it.
- Ranking, six views. Ranking score, average position, keyword count, top terms by search volume, best pages, and competing domains with their authority.
- Presets that account for the lag. Twenty-eight and ninety days, with the two-day delay already built into the range.
Four numbers, four separate ways of being wrong
Four figures dominate every status meeting and each is routinely stretched past what it can carry. Taken alone, any one can describe a business that is not the one you run.
Exposure counts drift for causes unconnected to anything you did — a layout with more organic slots, a rival disappearing, a launch announcement keeping your studio's name in circulation. Clicks are unambiguous and equally uninformative about who clicked: forty visits from students researching a career in software are indistinguishable from forty by a producer drawing up a shortlist.
The rate hides how large its denominator is. Fourteen per cent of seven appearances is noise wearing a percentage sign; one and a half against five thousand is a title tag worth an afternoon. Sort by volume first.
The mean position is weighted by exposure, not by what a term is worth to you. Four thousand appearances at twenty-two pull it down harder than sixty at position two can pull it back — and the second term is the one named on the day a contract closes. Nor does the scale behave evenly. Eleven to six changes the traffic noticeably; thirty-four to twenty-nine changes none of it, and the distance travelled is the same.
Counting terms by band ends the argument, because a distribution can be compared and a mean cannot: nine terms in the first three places, thirty-eight in the first ten, a hundred and four in the first thirty. Those figures still mean the same thing next month.
Queries and pages describe different problems
The query view and the page view cover the same seven days and answer opposite questions. Queries tell you what was asked; pages tell you what was served. Where the two disagree, the disagreement is the finding.
The query view
What people typed, in whatever language, whether or not you have a page that deserves the visit.
- Reveals vocabulary you would not have chosen
- Shows both languages side by side, unlabelled
The page view
Which URL the engine considered your best answer — often not the one you built for the purpose.
- Exposes the wrong page winning a term
- Doubles as a language filter when directories are clean
A worked case makes it concrete. A visual effects studio finds a service term at position eleven with respectable exposure, which looks like a content problem. Open the page view and the URL being served is the careers page, which describes the same pipeline in the same words and has more links because people share job postings. Rewriting the service page fixes nothing; the fix is separating two pages that say the same thing to different audiences.
The pattern repeats wherever the work and the recruitment describe the same craft. An advanced manufacturing shop's equipment list and its apprenticeship page compete for identical vocabulary, as do a fashion label's lookbook and its wholesale page. The pages view shows this in seconds; the query view never does.
The same firm, searched under different names
Here the standard playbook stops being useful. Most published advice assumes one company, one market, one language, with translation bolted on later. A Montreal firm often works the other way round: it runs internally in French, sells in English to buyers in California and London, recruits in both, and holds a local reputation read almost entirely in French.
These are not two versions of one audience. They have nothing in common — vocabulary, intent, rivals, devices. They arrive at one property and, in an unsegmented report, are added together.
- The buyer abroad. Types English, compares you against vendors in other countries, reads a services or portfolio page, and never opens the French tree.
- The candidate here. Types French or a mix, arrives on mobile, and forms an impression of the company that outlasts any campaign.
- The local partner. Searches your name in French and reads the absence of a French page as a statement about the company.
- The wrong page for all three. One bilingual page attempting to serve everyone usually ranks for none of them and converts none of them.
Thresholds therefore belong to a language segment, not a property. A French segment running at a fifth of the English volume never triggers a rule calibrated on the English total, so it is quietly abandoned — not by decision, but by arithmetic.
Segmentation is a URL question before it is an analytics question. When each language has its own directory, the pages view becomes a reliable language filter and band counts can be kept apart. Where the two share addresses, no filter separates them and the structural work comes first.
The country split, and the competitor you did not expect
The country dimension is housekeeping in most reports. Here it is diagnostic, because French-language results are not one market. What ranks for a French query in Quebec and what ranks for the same string in France are often different pages, and the second set is older and better linked.
The consequence is direct: a Montreal company competing on generic French terms is usually competing against French websites, not Quebec ones. You are not losing to the studio on Saint-Laurent but to a Lyon firm with a decade of French-language links, on a results page assembled for a country you do not sell to.
| What the country split shows | Likely reading | What to do about it |
|---|---|---|
| French terms with heavy impressions from France, almost no clicks | You are appearing in a market that has no reason to hire you | Judge those terms separately or exclude them from the average |
| Strong Quebec position, weak overall average | The mean has been dragged down by a foreign audience | Report position by country, never as one figure |
| English terms with heavy United States exposure | Your real sales market, invisible in a Canada-only view | Track it as its own line, with its own targets |
| Both languages blended into one Canadian row | Both are searched from Canada, so no country filter divides them | Segment by directory in the pages view instead |
None of this shows up in a total. It shows up when the country view and the keyword view are read together, which is the argument for keeping both under one roof. The competitor list adds the third piece: the domains sharing your keyword space appear with their authority and your terms in common, and where they are based often explains a position nothing else accounts for.
What people say is not what they type
Technical and creative fields here borrow English terms into French speech constantly. In a meeting you will hear le pipeline, le rendering, le lookbook, le level design, said without hesitation by people whose working language is French. That is genuine usage. The trap is assuming it transfers to the search box.
The language of the room
Borrowed English terms used fluently inside French sentences, by people whose working language is French.
- Fast, informal, unmeasured
- Varies by seniority and by team
- Never appears in a report
The language of the query
What the same person enters on a French keyboard, often a different form and sometimes a different word entirely.
- Recorded exactly, spelling included
- Splits across accent variants
- The only version that ranks
It transfers unevenly. Some borrowed terms are typed exactly as spoken, because no French equivalent has settled. Others are spoken in English and typed in French, because the searcher is unsure of the borrowed spelling.
- Do not write from the meeting transcript. The room and the query are different registers, and only one of them is measured.
- Let the query list decide. Search Console reports the typed form, spelling and all — the only unarguable evidence of what your audience enters.
- Watch for accent variants. The same word arrives with and without accents, from different devices, and can appear as two rows with two positions.
- Do not settle it once. Vocabulary in games, AI and post-production moves faster than a site rebuild. A term unsearched two years ago may now carry volume.
This is where research that generates candidate terms rather than counting existing ones earns its place. Seed keywords, live results and query history feed a pool, and every candidate is approved, rejected or held back individually — which matters, because in a bilingual vocabulary half the suggestions are terms no client of yours has used.
Devices, date ranges and the reading over time
A number without a comparison is decoration, and three conditions decide whether the comparison is honest. The first is the lag: Search Console data lands roughly two days behind. The presets account for it, but people still open a dashboard the morning after a launch and read the thin right edge as failure.
The second is the shape of your year. Post-production work arrives in blocks tied to production schedules; game studios spike around announcements; fashion runs on seasons; independent retail has a December unlike any other month. October against July tells you about the calendar. This October against last October tells you about the work.
The third is the device split, rarely neutral in a bilingual property. Candidates and local readers skew heavily to mobile; buyers on the studio and procurement side work at a desk. A blended position smooths away a gap of several places, and that gap is often the whole justification for template work on a page that looks fine on a large screen.
Frequently asked questions
Should we track French and English keywords in one project or two?
One project is fine as long as the languages live in separate directories, because the pages view then acts as a language filter. What is not fine is one blended set of thresholds. Set a target per language, or the smaller segment never triggers anything and drifts out of the reporting within a quarter.
Our English pages outrank our French ones in Quebec. Is that a problem?
It depends who is searching. If the query is English and the searcher is comparing vendors internationally, the English page winning is correct. If a French query in Quebec returns your English page, the engine is telling you the French page is thin, translated rather than written, or too close to its counterpart. Check the query language first.
Why does the country view show so many impressions from France?
Because generic French terms are contested in a far larger market than yours. Appearing there costs nothing and converts nothing, but it distorts your average position and click-through rate. Report those terms separately and set position targets against Quebec figures rather than the global mean.
Some clicks in the totals do not appear in any keyword row. Where are they?
Searches with very little volume are held back from the detail rows so individual people cannot be identified. Those visits sit in the site total but attach to no search term. The held-back share is proportionally larger in whichever language segment is smaller, which is one more reason a French segment reads worse than it performs.
Where the numbers stop and the decisions start
Measurement runs out earlier than most people expect. Nothing in these views decides whether a French portfolio is worth building, whether a careers page should be left to outrank a service page, or whether a term used only in France is worth fighting for. Those are commercial calls, and what the reporting owes you is early warning, delivered to whoever can act on it.
Where the figures earn their keep is in pointing at the right place at the right moment — but only if the rules were agreed before the reading rather than assembled around it. How far the top-ten count falls before somebody reviews it; which French search with exposure and no visits gets rewritten. A rule invented afterwards is an opinion with a number attached.
Handing the weekly reading to somebody else
For teams unwilling to carry the segmentation and the weekly pass internally.
- AutoSEO at 149 USD per month, per domain. Terms discovered and ranked without your involvement, links built the same way, on-site recommendations, and every view described here. Near 200 CAD — treat that conversion as rough.
- FullSEO at 500 USD per month, per domain. Terms picked by hand with the automatic route as fallback, placement held to a domain-rating threshold, on-site edits read by a person before release, and specialists, developers and writers behind it.
- Stream. The assistant inside My SEO: one running feed of replies, reports, freshly placed links and open to-dos, tied to what the project actually measures, and it takes lists in bulk.
Set those rules against your own record, not a published template. Replay the last three months through the keyword dynamics view and note how often each would have gone off. A rule firing every Monday sits too low; one that has never fired is decoration. Keep a dated record of what you changed and when so the next comparison holds up, and there is more of this on our blog.
To see your own figures divided the way this city requires, connect a verified property to a dashboard with your own domain attached and give it a quarter of real data. What deserves attention here is seldom the average position. It is the handful of searches where the language of the query and the language of the page have quietly drifted apart — and no total shows you that.
