What we measured
Two automated captures of the same search, taken five days apart, were rendered offline at four different screen widths. Both painted only part of the right hand column: the About card and a short list of profiles.
What was in the page against what was painted:
- The subject's name: 86 times in the page source, drawn only in the ordinary results on the left.
- The panel subtitle: 6 times in the source, never drawn.
- The claim invitation: twice in the source, never drawn.
- The internal marker for the whole panel: 13 times, never drawn.
The data was all present. The code that draws the panel header, the subtitle and the photograph mosaic runs in the viewer's browser and does not run in an offline capture. So the capture is not stale and it is not broken. It is structurally incapable of answering the question.
Why that matters commercially
We refund against whether a panel appears. If the instrument that decides cannot see a claimed panel, then the instrument is deciding money on a measurement it cannot make.
A scraper sees less of a claimed panel than its verified owner does. We wrote that down, then ignored it, then measured it.
It is also why a capture that does show a panel still cannot tell you whose it is. Six of twenty one clients had a panel for their name and none of them owned it.
The mistake we made with it
In September we reported our own founder's panel as unclaimed. The evidence was the presence of a claim invitation in the page source and the absence of a verified marker, both read off exactly the kind of capture described above.
Three things were wrong with that at once. The instrument cannot see claim status. Not observed was reported as absent, which breaks the rule we apply to everything else. And a signal from the maps and reviews system was carried across to the knowledge panel, which is a different system. That confusion is common enough to have its own page.
What does work
A person, signed in, looking. That is how every panel image on this site was taken, including the one on our homepage. For a developer, Google's own data interface answers it properly and returns the identifier, the description and the official URL.
And whatever the method, three states rather than two: present, absent, and not observed with the reason given. The third is the honest answer far more often than anybody selling a checking tool would like, and it is the one that keeps a monitoring report trustworthy.
Two automated captures of the same search five days apart, rendered at four viewport widths. The subject's name appeared 86 times in the page source and was drawn only in the ordinary results column.