We measured the local heat maps against their own grid data

What we pulled from every account and how much of it we verified

We exported the full grid summary for the account, which returns one row per grid pin rather than one row per keyword. That gave us 35,492 rows covering 874 keyword grids across 24 business records. Every average in this document was recomputed from those rows.

We then pulled Google Search Console and Google Analytics directly, using our own OAuth credentials rather than anything reported back through the platform. That gives us a second opinion that does not originate from Search Atlas.

One limit up front. We could not independently confirm any individual rank claim. We tried, and the section near the end lists exactly what we tried and how each attempt failed. Everything in this document is therefore a check of Search Atlas against Search Atlas, or against Google data that measures a different surface. Where we are guessing, we say so.

Where the local heat maps hold up against the underlying data

We want to lead with this because it is the larger part of the finding. The grid mechanics look real and internally consistent.

The problems we found are not with the crawl. They are with how the collected data is summarised into the single number that a client actually reads.

How the average rank on your dashboard is actually calculated

We reverse engineered the headline average by testing candidate formulas against the exported pins. One fits, and it matched the displayed figure for 17 of the 20 business records that show one. The three misses were all within 0.6 of a rounding boundary.

Step 1  For each grid, average the 39 pins. A pin where the business does not appear counts as 21, one worse than the bottom of the visible scale.
Step 2  Discard every grid where the business appeared on no pin at all.
Step 3  Average the grids that are left. That is the number on the dashboard.

Step 1 is defensible and we have no argument with it. Step 2 is the one we would like explained. A keyword where the business is invisible across the entire grid is the strongest possible signal of a visibility problem, and it is the signal that gets removed before the average is taken.

The grids that disappear from the average before you see it

Across the account, 124 of 874 grids (14.2 percent) return no ranking on any pin, and every one of them is excluded from the average its client sees. The effect is not evenly spread. For some businesses it changes the headline substantially.

Procare Consulting is the widest gap in the account: 26 of 50 grids dropped, moving the average from 11 to 15.93.

A business that ranks nowhere is recorded as position zero

When a business appears on none of the 39 pins, the detail endpoint reports avg_rank: 0 and best_position: 0, alongside a position distribution showing all 39 pins in the worst band. The list endpoint reports the same grid as null.

So the same condition is expressed three different ways across the product: null, 0, and 21 inside the per grid maths. Zero is the one that causes damage, because on any scale where one is the best rank, zero reads as better than first.

We checked what this does downstream in our own reporting, and we found the fault on our side too. Our report builder passes avg_rank straight through and sorts the map pack table ascending, so a grid reporting zero sorts to the top and is presented to the client as their strongest keyword. That is our bug to fix and we are fixing it. We raise it here because any customer building on this field will hit the same thing.

One keyword where the chart reports the opposite of what really happened

This is the finding that prompted the review. Texas Strong Plumbing tracks the keyword katy plumbing. Its two snapshots read like a collapse.

Worth adding as context rather than criticism: the top competitor on that grid is The Katy Plumbing Company Heating & Air, holding 38 of 39 pins with 3,639 reviews. The keyword is close to that competitor's brand name, so the tracked term may simply be a poor choice on our side.

Two endpoints in the platform return two different average positions

Calling the list endpoint and the detail endpoint for the same grid, within the same minute, returns two different averages. The detail endpoint agrees with the pins. The list endpoint is consistently higher by exactly half a position.

The first row is the cleanest test. All 39 pins report rank one, so the average can only be one. The detail endpoint says one. The list endpoint says 1.5. We would like to know which figure the dashboard and the reporting exports use, because a portfolio view built on the list endpoint would show every client half a position worse than their own detail page.

Ranks that sit above the top of the twenty position scale

No pin in the export carries a rank above 20, yet 15 grids report an average above 20. This follows directly from counting absent pins as 21, and it is arithmetically correct. It is also unreadable: a client asking what position 20.9 means has no good answer, because the scale stops at 20.

In every one of these cases the business ranks on one or two pins out of 39. A label such as "visible on 2 of 39 points" would carry the same information without implying a position that cannot exist.

Grid settings drift quietly between clients and inside accounts

Our own standard is a three mile circle. The export shows that is not what every client has, and in one case the difference is large enough to make the client's number incomparable to everyone else's.

Spacing. Mayday runs at null miles against a portfolio standard of three, producing a measured grid radius of 25.05 miles. Half that grid sits in towns a local pest control business has no reason to serve, which is most of why it reports 46 percent of pins absent and an average of 13. The number is not wrong. It is measuring a different question.

Mixed spacing inside one account. 0 records carry more than one spacing value across their grids, so a single business average blends grids covering different geographies.

Grid placement. Right On Roof On has its grid centred 21.5 miles from downtown Austin and 2.1 miles from Hutto, while tracking keywords written for Austin. It has no address on record in the platform. All 25 of its grids return nothing, and misplacement is the most likely reason.

Freshness. Texas Strong Plumbing is set to a crawl frequency of zero, meaning one time only. Its last scan was 2026-06-01, 56 days ago, and it will not refresh again. This client received a report during that window.

Records in the account that should not be billed at all

Some of this is ours to clean up rather than yours to fix, but it affects what we are paying for so we are including it.

A home address tracked as a business. The record named 20547 Amberfield Dr is a residential street address carrying 3 grids and 117 pins, tracking the terms house, condo and apartment for rent. Every pin returns nothing, and has since it was created.

Both records above export a single pin and have never been scanned. Their coordinates are also transposed: the Nashville record exports latitude -86.8576 and longitude 36.1306, which are that address's real values in the opposite fields. A latitude of -86.86 is in the Southern Ocean.

Duplicate grids. 6 grids export twice the expected rows, at 78 rows across only 39 distinct pin coordinates, which suggests the same keyword is tracked twice for the same business.

Business record drift. Review counts held against the heat map record differ from the counts held against the same business in the Google Business Profile section of the same platform.

What Google's own search data says about these keywords

We pulled Search Console for every property we control and matched each tracked heat map keyword against real recorded queries over the ninety day window. 537 of 832 tracked keywords (64.5 percent) appear in Search Console at all.

We want to be careful about what that does and does not prove. Search Console measures the website in web results. The heat map measures the business listing in the map pack. They are different surfaces, and a keyword can carry real map pack demand while producing no web impressions. This table cannot falsify a single rank claim and we are not presenting it as though it does. What it does show is which tracked terms have any demonstrable demand behind them.

The two ends of that table are both informative. Right On Roof On sits at the bottom, and the heat map already says the same thing, so the two sources agree that this business is invisible. At the other end several records show most tracked terms carrying real impressions, which is the pattern you would expect from a well chosen keyword set.

Analytics over the same window, for context on whether visibility is reaching these sites at all:

The checks we ran that Google would not let us complete

We set out to verify a specific claim: that at a given coordinate, for a given keyword, a business holds a given position. We could not complete that check, and we would rather state that than imply a verification we did not perform.

4 of the 6 routes we tried are closed to us and 2 are open. The two Google Business Profile interfaces return a quota ceiling of zero on our project, which means the direct profile integration is not available to us at all. That is a gap on our side, not a fault in your product, and we are pursuing access separately.

Questions we would like the Search Atlas team to answer

1. 1Why are fully invisible grids excluded from the average?Dropping the 124 grids where a business appears nowhere removes the strongest negative signal in the dataset before the headline number is produced. Is that deliberate, and is there a setting to include them?
2. 2Can the zero value be separated from a real rank?avg_rank: 0 currently means "appears nowhere" but sorts and plots as better than first place. A null, or an explicit flag, would remove a whole class of downstream error for anyone building on the API.
3. 3Which average is authoritative?The list endpoint and the detail endpoint disagree by exactly half a position on every grid we checked, with the detail endpoint matching the pins. Which one feeds the dashboard, and which feeds report exports?
4. 4How should an average above twenty be read?15 grids report a value above the top of the scale. Is a count of ranking pins available as an alternative summary?
5. 5Can a customer verify a rank at a coordinate?We could not reproduce a single position independently. A documented method, or a raw response sample for one pin, would let us stand behind these numbers with clients.
6. 6What happens to a grid set to a crawl frequency of zero?One of our clients has 50 grids frozen at 2026-06-01. Should a one time grid still present as current data in the interface, and should it be flagged as stale after a period?
7. 7Why do the heat map and profile records hold different review counts?11 businesses differ between the two sections of the platform, by as much as 10 reviews. Which record is the live one?
8. 8Are duplicate grids billed twice?6 grids export duplicate rows for the same keyword and business. We would like to know whether these are two tracked grids or one grid exported twice.

How to reproduce every single number in this document

Nothing here was typed by hand. Every figure is computed from the exported data by a script, so any number in this document can be traced back to a row in the source files.

1  Export the grid summary for all businesses. One row per pin, 35,492 rows.
2  parse.cjs reads the export and groups pins into 874 keyword grids.
3  analyze.cjs recomputes every average, counts the sentinel values,
    matches keywords against Search Console and writes findings.json.
4  build.cjs renders this page from findings.json alone.

Readings taken directly from the API are recorded verbatim in
api-readings.json so the displayed values can be diffed against the export.

The comparisons in the endpoint section were taken by calling both endpoints for the same grid identifier in the same session. Those identifiers are printed in the tables above so they can be checked directly.