Settling the disagreement
The instrument, the origin and the radius do most of the damage. Four conditions remain, and then the procedure for settling which reading is wrong.
Depth: every ruler stops somewhere
Both the single check and the grid read twenty results deep. Past twenty you are not ranked 21; you are not measured. The check says so in words — you're not in the top N here — and names who holds #1, so a null result still carries information.
A tool reading a hundred deep reports #34 where this one reports nothing. Neither is wrong; they are different rulers, and deeper is not automatically better, because nothing at #34 is visible to a human on any surface.
Depth buys early warning — #61 to #38 is progress you cannot otherwise see — and costs you an invitation to average numbers that describe invisibility. Two tools censoring at different depths therefore cannot produce comparable "average position" (reading a geo-grid without fooling yourself).
Language is not cosmetic
You can request results in a specific language or leave it blank and let the provider pick. Blank is not neutral; it is a default you did not choose and cannot see.
Language is part of a tracked keyword's identity here — the same phrase in English and in Finnish is two independent rows with independent histories, because they are two different measurements.
What language certainly changes is the strings that come back. Whether it also shifts which businesses come back, and in what order, is a market-by-market question rather than a rule (inference — run the two rows side by side; in a monolingual market you will usually see no difference).
Helsinki, Montreal, Barcelona and Brussels are where two trackers disagree hardest, and language is usually why.
The moment
Rank is a sort order over near-tied scores, and a sort is discontinuous: a hair of movement underneath swaps two businesses and moves a reported position by a whole integer. "The number changed" and "the world changed" are different claims.
A grid scan is not an instant. Points are measured in sequence with a deliberate pause between them, so the far corner is not read at the same moment as the centre.
Compare stamps before numbers. Every card showing fetched data is stamped for this reason. A day between two readings is a day in which the disagreement may be entirely real.
Search volume stales differently, and is not local. The monthly volume beside a keyword comes from Google's Keyword Planner data, which Google documents as "the approximate number of monthly searches on a query, averaged for the past 12 months" — approximate is Google's word.
Practitioners have long observed the returned values landing on a fixed set of rounded buckets rather than exact counts (observed and widely reported; Google does not document the buckets). The request carries a country-level geographic target and a language, and the answer is cached here for about a month.
Two tools reporting different volumes for one phrase can both be faithfully quoting Google, for different targets — and neither figure tells you how many people search it in your town (what people actually search).
Identity: are you sure that result is you?
A rank check must decide whether a result is the business. The honest way is to match the stable place identifier — the machine key from the business entity — which is what happens here, on the single check and on every grid point. The convenient way is to match the name as a string.
Name matching fails both ways:
- False positives — a franchise sibling three towns over is counted as you.
- False negatives — the name on Google no longer matches the name in the tool, because somebody edited the profile.
A tool that quietly matched a different branch will show a beautiful, entirely fictional improvement. Duplicate listings are the same problem wearing a hat — one tool tracks the record that ranks, another the record that does not, and both are "correct" (spam and fake listings).
So which one is wrong?
Run the seven conditions. Fill in what you can and mark the rest unknown.
A disagreement is a defect only when two tools agree on all seven conditions and still differ. Until then they measured different things, and arguing about which number is right is arguing about the wrong question.
Most disputes die at row two or three, and the honest verdict is not comparable.
When two readings genuinely share all seven and still disagree, there is one tie-break: reproduce a single point by hand and see which is nearer. That hand reading is not an oracle — it has its own seven conditions, one of which is your own browsing history — but it is a third opinion from a different mechanism, which is what you need.
What you tell the client is a sentence, not a defence:
"Those two numbers were measured from different points, at different depths, on different days. Here is what ours measures and why we chose it. To reproduce the other one I need the conditions it was taken under, and the report does not state them."
That ends the meeting. "Their tool is inaccurate" starts an argument you cannot win.
Without SEOG · Every condition above can be set by hand — that is Lab 22.3 — and the record-keeping decays after about a dozen readings. Doing all of this without SEOG.