How do I know if my SEO agency is working?
Ask for the list of pages they deliberately did not touch, and the comparison against them. If there is no such list, their report cannot tell you, however good the work was.
An agency report shows rankings, traffic and a line going up. All of it can be true while the work did nothing, because none of it is compared with what would have happened anyway. Seasons change, competitors come and go, Google changes its mind several times a year. A rise of 30% in a good quarter is a description of the quarter.
There is one test that settles it, and it is a hundred years old: leave part of the site alone and compare. If the pages the agency worked on pulled ahead of the matched pages nobody touched, the work did it. If they did not, it did not. Farmers have run this exact test on one field since 1843 for the same reason: without the strip left alone, no one could say what the fertiliser was worth.
Two questions to put to your agency, in writing. First: which pages did you not touch, and how did those do over the same period? Second: what is the range around your number, that is, how different could the true result be? An agency that measures will have both answers in a minute. An agency that does not will explain why the question is unfair.
Be wary of the opposite promise too. Guarantees of a ranking are treated as a warning sign by the field's own reference publishers, because nobody controls Google's results and a guaranteed rank can be for a phrase nobody searches. The only guarantee worth anything is one tied to a comparison you can check.
The longest-running field experiment in the world has kept one strip unfertilised since 1843, so the fertilised strips have something honest to be compared against. It is still running.
Broadbalk Wheat Experiment, sown autumn 1843
The field's own reference publishers treat ranking guarantees as a warning sign: nobody controls Google's results, and a guaranteed rank can be for a phrase nobody searches.
Ahrefs and Semrush guidance
Repeated runs of the same prompt on commercial AI engines agree on only 34% to 42% of sources within 24 hours. A move from 8% to 11% visibility cannot be told from random variation.
Survey of GEO measurement, bootstrap intervals 3 to 7 points wide
Google announces core updates several times a year and says they can change how pages rank without anything on the page having changed.
Google's own documentation
The free check reads your own site and tells you how many of your pages have not changed in a year, and whether the site is big enough for a real comparison. No account, no estimate, about a minute.