Position three used to be a respectable outcome. For a year we treated it that way — a page sitting under two ads and a featured snippet, still collecting a predictable share of the click. That share is no longer predictable, and the ranking reports have not caught up.
We lined twelve months of Search Console clicks against the same queries as they appeared in a weekly SERP sample. The ranking column barely moved. The click column did. The gap opened first on informational queries, then on the comparison pages that used to convert from position two.
The ranking did not fall. The click did. That is a different problem.
What the traffic line actually shows
On the thirty queries we could hold still — same intent, same page, no major content rewrite — clicks fell 18% while average position improved by two tenths. The pages that held up were the ones already answering the question in the first 120 words, not the ones optimized for a ten-blue-link world.
That is the part most dashboards will miss. They will credit the next release, or the next core update, because those are the events with names. The quieter change is a SERP that no longer needs your click.
What to watch instead
If you only watch rank, you will keep shipping the wrong work. Watch clicks per impression on the queries that still look healthy. Watch which pages lose sessions in the same week the overview appears. Then put those dates on the same calendar as the changes you actually made.
| Signal | Still useful? | Why |
|---|---|---|
| Average position | Sometimes | Can improve while clicks fall |
| Clicks / impression | Yes | Catches overviews and snippets |
| Sessions on the URL | Yes | Confirms the click actually arrived |
| Named core updates | Rarely | Too loud, too late |
The work is not to panic about AI. The work is to stop treating a stable ranking as a stable business.