Rank tracking: your tool measures one country, not your traffic

Your tracking tool says position 4. Search Console says 11. The question that follows is always the same: which one is lying?

Neither, and that is testable. We took 30 non-brand queries on vydera.com, read their Search Console average position over 364 days, then measured the actual rank of vydera.com in Google's SERP on the morning of 28 August 2026. 44 SERP pulls, $0.1495 of budget, one single morning.

The result does not go where we expected. We set out to put rank trackers on trial. It is the SERP pull that turned out to be the narrower reading of the two.

What we measured, and with what

The sample is 30 queries in two strata. Stratum A: the 20 vydera.com queries with the most impressions over twelve months and an average position between 1 and 30. Stratum B: 10 more queries, ranked this time on impressions recorded in France, French position between 1 and 30, excluding stratum A. That is 15,502 impressions and 6 clicks over the window from 27 August 2025 to 25 August 2026.

On the SERP side: the DataForSEO serp/google/organic/live/regular endpoint, depth 20, desktop, filtered on the domain. 32 pulls on the French SERP in French, 9 on the US SERP in English, 1 positive control and 2 stability repeats. All of it lands between 07:50 and 07:59 UTC.

Three precautions are worth copying as they are:

• A positive control before the first cent is spent. A brand query has to return the domain at rank 1. Without it, a badly written filter publishes "absent everywhere" when nothing was measured at all. That control also returned a lookalike domain at rank 6, later excluded by filtering on the exact domain and its subdomains.
• Two queries pulled twice, six minutes apart, in the same configuration. Same ranks: 7 and 7, 5 and 5. A comparison only means something if the pull reproduces.
• A server error is not an absence. 2 calls out of 44 came back with an empty result. They were replayed, never counted as absences. A "no results" code carrying a page count, on the other hand, is a genuine absence.

One warning before the numbers: 20 of the 30 queries carry the name of a competing tool and come from the fan-out of a single one of our articles. This sample describes the real traffic of vydera.com; it does not represent an ordinary keyword portfolio.

First result: a 3.82 rank gap, where the comparison exists at all

Across the 12 queries where vydera.com appears in the top 20 of the French SERP, the median absolute gap between the Search Console average position and the observed rank is 3.82 ranks. Mean 4.82, maximum 13.77. Five queries exceed a 5 rank gap, three stay under 1 rank.

The two strata do not behave the same way, and the difference is instructive. The 20 stratum A queries, the ones carrying the most impressions, show a median gap of 3.82 ranks across 8 queries found. The 10 stratum B queries, selected on French impressions, climb to 5.16 across 4 queries found. Tiny counts on both sides: we flag it, we conclude nothing from it.

The gap does not run in one direction either, which is enough to rule out a systematic offset:

• On a French link-building query, 98% French by impressions, Search Console says 9.26 and the SERP puts the page at rank 7. 2.26 ranks better than the average.
• On a conversational French question about GEO tools, Search Console says 5.00 and the SERP returns rank 5. Exactly zero gap.

One thing this comparison mixes and does not separate, and it belongs here rather than in a closing footnote: the difference in method and the time elapsed. 18 of the 30 queries concentrate at least 90% of their impressions in a single month, sitting somewhere between May and August 2026, and the SERP was pulled on 28 August. Part of the gap is ageing, not method, and the measurement cannot tell you which part.

The number you must never publish on its own

3.82 ranks is a survivor's number. It only covers the 12 queries where both measurements exist. On the other 18 there is no gap to compute: vydera.com is nowhere in the top 20 of the French SERP.

Here too, sort before you dramatise. Three of those 18 queries already have an average position beyond 20 in Search Console: their absence from the top 20 is consistent with it, not divergent.

That leaves 15 queries whose gap is not measured but censored: all we know is that it exceeds a bound. That bound is at least 11.41 ranks, 16.44 at the median, up to 18.11. Two of those queries are credited with an average position of 2.08 and 2.73 and appear nowhere in the first twenty French results the same morning. The worst treated is a long Spanish question, credited with an average position of 1.89 and carried 87% by Colombia: its implied gap exceeds 18 ranks, and it says nothing other than that we were querying the wrong country.

This is the hard part of the article. On those 15 queries, no tracking tool querying the French organic SERP will ever find anything to track. These are not two measurements diverging, it is one of the two that does not exist. Publishing "3.82 ranks of gap" without saying what it covers would be survivorship bias, plain and simple.

Why: the average follows the country that carries the impressions

The Search Console calculation is not a hypothesis, it can be checked. Reweighting the per-country positions by their impressions lands back on the published average position: zero difference to the sixth decimal on the first five queries of the sample, and the per-country impressions sum exactly to the published impressions.

Now, on this sample, 48% of impressions come from the United States and 9% from France, ahead of Colombia at 7% and Brazil at 5%. The "average position" read from a French interface is therefore largely an American position.

Across the 5 queries pulled in both countries and found in both, the gap against the Search Console average position drops from 4 ranks to 1 rank depending on whether you query the French or the US SERP. The same morning, the same queries rank vydera.com 5 ranks lower in France than in the United States.

The working title of this article accused tracking tools of measuring badly. The measurement says something else: a tracker measures one location at one instant perfectly well, and the traffic does not necessarily live there. Five pairs is an observation, not an estimate; no statistical test is possible at that count. And the 48% American share belongs to this sample, not to the web at large: what generalises is the weighting mechanism, not the percentage.

The three axes an average position flattens

Country is not the only aggregate hiding inside a Search Console row. Within a single row, the position range is:

• Country: 3 ranks at the median, up to 11.05, across all 30 queries.
• Device: 0.75 ranks at the median across the 12 queries served on two devices, but up to 19.65 on one of them, a single row melting a mobile and a desktop position that are almost opposites. The fact exists, it is not general, and it is read in Search Console: no mobile SERP pull was purchased here.
• Month: 1.64 ranks at the median, up to 8.46, across the 19 queries observed over at least two months.

A fourth trap hides on the time axis, and it is the nastiest. 18 of the 30 queries concentrate at least 90% of their impressions in a single month, with the busiest month holding a median 94% share. The most recent month observed runs from May to August 2026 depending on the query. An average advertised as "over 12 months" therefore covers, in practice, a few weeks. Comparing it to a SERP from 28 August sometimes means comparing August to May.

Three details the pull handed us along the way

An AI Overview was present in all 39 usable pulls, France and United States alike, on every single query. The endpoint we used exposes neither AI Overviews nor People Also Ask as positions: it says an AI Overview was there, never where. What follows from that weighs on how the whole article should be read, so let us write it plainly: nothing here proves that Search Console and an organic pull measure the same surface. Search Console does not say whether an impression came from the classic SERP, from an AI Overview or from AI mode.

Of the 12 queries found, 6 rank an English page on google.fr and 6 a French one. None of the 44 pulls returned two vydera.com URLs at once, so on those 6 queries the French page was not in the French top 20 at all. A tracker pointed at a target URL rather than at the domain would have logged six absences there, while the site was in fact ranking. That is the kind of false negative a correct hreflang setup at least lets you diagnose.

Finally, the result count Google announces is not data. On the same query pulled two and a half minutes apart it goes from 35,300 to 7,120 while the rank does not move a notch. That field proves nothing beyond the fact that the SERP was actually crawled.

What this measurement does not say

• No commercial rank tracker was queried. The comparison puts Search Console against a SERP pulled by API, on 30 queries and one single morning. This is not a tracker benchmark.
• The pull depth is 20. A missing query is missing from the top 20, not missing from Google. No position beyond 20 is observed.
• Google does not publish its position counting rule. Neither the organic rank nor the absolute rank reproduces it. The measurement establishes a gap; it does not prove either number wrong.
• One site, one pull per configuration. This is a documented case, not a general law.
• No personalisation, no browsing history, and no multi-day tracking: stability beyond two repeats six minutes apart is not measured.
• The query shape classifier we use elsewhere finds only 7 synthetic queries out of 30, with a median of 4 words. Reading them by eye, we would have claimed many more. What gets published is the classifier's verdict, not the impression from reading.

What to do, in order

1. Open Search Console on the country dimension before configuring anything. The country carrying your impressions is not necessarily yours. That is the one that should set up your tracker.
2. Never compare an average to a snapshot without segmenting first. Filter Search Console on the country, device and month of the pull before calling a gap a gap.
3. Check the time concentration. If 94% of a query's impressions land in a single month, the "12 month average" describes that month and nothing else.
4. Check which URL actually ranks. On half of our found queries, it was not the page the locale expected.
5. Treat the two tools as complementary, not competing. Search Console describes the real population of your impressions, all surfaces mixed together. A tracker gives you a reproducible control point precisely because it freezes country, language and device. The first tells you what happened, the second whether your work moves anything.
6. Do not steer on a position when the surface has changed. An AI Overview was present in every one of our pulls. A rank 3 sitting under a generative block is not a rank 3 from 2019, and that is a KPI selection problem before it is a tooling problem.

Running the measurement yourself

You need a Search Console export carrying the country dimension, and a SERP endpoint. Ours bills $0.0035 per pull at depth 20 and $0.002 at depth 10. Our 44 calls cost $0.1495, and the balance delta before and after matches the sum of the per-task costs exactly.

The sequence is five moves: positive control first, filter on the exact domain and its subdomains, one query pulled twice to prove reproducibility, any server error replayed and never counted as an absence, and the pull country chosen from the impression split rather than from your own address.

For the definitions: SERP and query fan-out. If your pages show up nowhere because they are not in the index, the problem sits elsewhere and we measured that too in our article on non-indexed pages. And if you want us to look at your numbers with you, get in touch.