You rewrite a title on Tuesday. On Thursday the keyword has jumped several places and you tell the client the rewrite did it. The following Tuesday it's back where it started, and now you need a second story. Positions move for reasons you didn't cause far more often than for reasons you did, and reading them honestly means knowing which is which before you say a word.
A position is a range, not a number
There is no single place your page ranks. Google's own explanation of how Search works says results depend on the searcher's location, past search history and search settings, and they also differ by device and by the moment you ask, because Google is always testing changes on a slice of its users. Ask from a phone in Leeds and a laptop in London and you can get two different pages in the top spot.
Every measuring instrument picks one vantage point. A rank tracker asks from a fixed location and device, with no search history. Search Console's average position is the opposite: an average over every real impression your page received, from every location and device that saw it. The two rarely agree, and neither is wrong. Pick one instrument per keyword and never compare its numbers with the other's.
Results pages also aren't a plain list of links. An AI Overview, ads, a map pack, shopping results and a People Also Ask box sit in the same column, and where "position three" lands on the screen depends on what's above it. Some trackers count those features as positions and some don't. And the deeper a keyword sits, the fewer impressions it gets and the more its average wobbles.
Noise versus signal
Daily wobble of a position or two means nothing. Google re-ranks continuously, trackers sample, and a keyword that sits a couple of places lower tomorrow than today hasn't "dropped". Read positions the way you'd read your weight: a trend over weeks, not a reading on a single morning.
Three checks separate signal from wobble:
- The move holds across several consecutive checks instead of snapping back.
- Search Console shows impressions for that query moving the same way, with the most recent few days excluded because the data is still arriving.
- Clicks follow. A position change that changes neither impressions nor clicks didn't change anything that matters.
Compare like with like. Weekends, holidays and seasons move demand, not rankings, so falling clicks in August with steady positions is your customers on holiday. Use Search Console's Compare option against the same period last year.
If every tracked keyword falls off a cliff on the same day, the cause is almost always your own site: a deploy that added a noindex, a broken redirect, a robots.txt that now blocks everything. Check the site before you check Google.
Four things that move rankings without you
Before a change gets credit or blame, rule out what had nothing to do with you.
Google changed. The Google Search Status Dashboard lists confirmed ranking updates with start and end dates. Google's core updates documentation says a rollout takes time to finish, often weeks, so positions mid-rollout aren't final and nothing inside that window can be cleanly credited to your work. Check the dashboard first, every time.
The results page changed. An AI Overview or an extra ad appears above you, your position stays the same, and your clicks fall. The signature is steady impressions and steady position with falling CTR. The fix is a better snippet or a different query target, not a ranking fix; see What actually moves rankings.
A competitor changed. Someone published or refreshed a better page and took the slot. Search the query, see who's above you now and when their page last changed. Losing to a better page is information, not injustice.
Your own site changed. A redesign, a template tweak, a speed regression, a batch of internal links removed from the footer. Keep a log of every change with its date and the URLs it touched. A rank change you can't explain usually sits a few days after something in that log.
How to claim credit without lying
Set the rules before the change, not after. Decide which keywords and pages you're judging, which instrument, and how long the window is, in weeks. Then find a control group: pages of the same type you didn't touch. If the pages you changed moved and the untouched ones didn't, you have a case. If both moved together, Google moved, and your change rode along.
When you write it up, say exactly that. "We rewrote the titles on these pages on this date. Over the following weeks their average position improved, impressions and clicks rose, the untouched pages of the same type stayed flat, and the status dashboard shows no confirmed update in the window." That sentence survives scrutiny. "Rankings are up" doesn't.
When the evidence is mixed, say that too. A change that coincides with a core update gets "we can't separate the two yet; we'll know once the rollout ends". Refusing credit you can't prove is what makes your diagnosis credible when something really breaks; Proving ROI honestly takes it further for clients.
What to take away
- A ranking is a range that varies by location, device and the day, so pick one instrument per keyword and read it as a trend over weeks.
- A position change only matters when impressions and clicks move with it and the move holds across several checks.
- Before crediting your work, rule out a Google update, a results-page layout change, a competitor and a change on your own site.
- Decide the evaluation window and a control group before you change anything, and report movement you can't explain as unexplained.
Next
The next chapter puts Claude to work on the data you can now read, starting with Connect Claude to your SEO data.