Keyword Ranking History: How to Read Daily Movement
by Rohail Taha
20 minute read
Keyword ranking history is the chart you open after a keyword falls 12 positions overnight. Much of the movement in that chart is not a ranking change. Some of it never happened at all. You still have to decide this morning whether to rewrite the page, tell the client, or do nothing.
This article does three things. It explains what a ranking history measures. It uses RankCaddy’s own Google Search Console data to show how much of the daily movement is noise. It then gives you a way to tell a real drop from a false alarm.
Why tracking rankings still matters is a separate argument. This article assumes you already track your rankings.
One thing to know before you start. Across seventeen months of RankCaddy’s own ranking history, the two largest movements in the graph were not ranking changes. Both were changes in the way the ranking was measured. Both have a date you can put on the graph.
TL;DR: Collect positions daily and read your keyword ranking history weekly. In RankCaddy’s own Search Console data a page 1 keyword typically moved under 1 position a day. A keyword past position 50 moved about two and a half positions. Of 732 overnight falls of 5 or more positions, about 2 in 5 were back where they started within a week, with nothing done to the page. Wait five days before you diagnose a fall, and mark on the graph the dates when the measurement changed.
What is keyword ranking history?
Keyword ranking history is the record of the positions one keyword has held over time for a given site, on a given device, in a given location, on a given search engine. Every point on the chart is one measurement, taken by one method, on one day. That is why two tools can show the same keyword at two different positions and both be right.
A ranking history does three jobs, and it is not equally good at all three:
- Diagnosing a drop. This is what people open the chart for, and it is the job the chart is worst at. A single day’s reading tells you almost nothing.
- Proving progress across a reporting period. The chart is good at this, as long as both dates you compare were measured the same way.
- Checking whether a competitor’s move is new or long-standing. Useful, but it needs a different source from the first two jobs.
A ranking history is a set of measurements, not a set of facts. Those three jobs need three different sources.
Where ranking history comes from, and what each source measures
Ranking history comes from three sources, and each one measures something different. A report that mixes them will not add up.
What Search Console can and cannot show
Google Search Console gives you your own site’s positions for the last 16 rolling months. As a new month arrives, Search Console deletes the oldest one. Its average position is the average position of the topmost result from your site, averaged across every impression in the period.
Search Console also runs 2 to 3 days behind while the data finalises. To protect user privacy, anonymised queries are omitted from the query table, and reporting limits can omit additional rows, so the query list will not always add up to the property total. . Search Console is free, and provides first-party performance data, but its average position is not a live rank.
What a rank tracker measures instead
A rank tracker checks the position itself, on a schedule, from the day you added the keyword. A rank tracker can check the same keyword repeatedly using a fixed device, location and language, which makes one reading more comparable with the next.
Using the same method every time is the whole point. It makes Tuesday’s number comparable to Monday’s. Search Console’s average cannot do that, because the set of searches behind the average changes from day to day.
A rank tracker cannot go backwards, though. There is no history for the period before you started tracking.
Where competitor history comes from
For a domain you do not own, competitor history usually comes from a third-party SEO tool’s own SERP database or rank-tracking system. How often that history is recorded depends on the tool and the product: some datasets are sampled periodically, while tracked competitors can be updated daily. .
An archive like that answers a different question from Search Console and a rank tracker, and it answers it well. The one thing you must not do is present a monthly snapshot as if it were a daily series.
Which source to use for which job
| Where the history comes from | Whose positions it holds | How far back it goes | What it is good for |
|---|---|---|---|
| Google Search Console | Your own site only | 16 months, rolling, then deleted | The free baseline, and the only source with impressions attached |
| A rank tracker | Your site, plus competitor domains you add | From the day you started tracking | Day-to-day comparisons on a fixed method |
| A tool’s own crawl archive | Anyone’s, as the tool saw it | As far back as that tool’s own crawls | Checking whether a competitor’s position is new or long-standing |
If you were not tracking a keyword at the time, you may still find historical ranking data in a third-party database, but it may not have the same daily methodology as your own tracked history.
Historical keyword rankings versus website ranking history
A website ranking history is one average across a set of keywords. Historical keyword rankings are the individual lines underneath that average. They answer different questions, and they can point in different directions. .
A site average can improve while your most valuable keyword slides. Forty long-tail keywords can each drift two positions and move the average, even though no keyword you care about has changed.
Read the average to judge a quarter. Read one keyword’s own line to judge a page. Never let one stand in for the other, because a report that shows only the average can hide a real problem for a month.
How much of the movement in a keyword ranking history graph is real?
Less of the daily movement is real than the graph suggests. How much less depends on where the keyword already ranks.
Where these numbers come from
Every figure in this section comes from RankCaddy’s own Google Search Console property. That data covers 30 keywords, from 1 April 2025 to 5 September 2026, and 10,921 daily position readings. Three things are worth knowing about that dataset.
- These numbers come from one B2B software site and 30 keywords. They describe our own data, not an industry average.
- The positions are Search Console’s averages, not live rank checks. We chose that source because most readers already have Search Console open in another tab.
- Search Console reports only the best-ranking page from your site for each search. It also runs a few days behind.
The higher you rank, the calmer the graph gets
| Where the keyword ranked | Typical day-to-day movement | How often it moved 5 or more positions in a day |
|---|---|---|
| Positions 1 to 10 (page 1) | Under 1 position | 1 day in 22 |
| Positions 11 to 20 (page 2) | About 1 position | 1 day in 14 |
| Positions 21 to 30 (page 3) | About 1.5 positions | 1 day in 8 |
| Positions 31 to 50 | About 2 positions | 1 day in 4 |
| Positions 51 and worse | About 2.5 positions | 1 day in 4 |
Based on 8,399 pairs of consecutive days. Exact medians: 0.75, 0.96, 1.44, 2.19 and 2.64 positions. Exact shares for a move of 5 or more: 4.5%, 7.4%, 12%, 26% and 26%.
How much a movement matters depends on where the keyword already ranks.
A fall of 5 positions at position 6 is rare enough to be worth an hour of your attention. The same fall at position 46 happens every fourth day and means nothing at all. The graph shows the same drop in both cases. What you do about it is opposite.
Low-traffic keywords exaggerate everything
| How many searches showed the keyword that day | Typical day-to-day movement |
|---|---|
| Fewer than 5 | About 4.5 positions |
| 5 to 19 | About 2.5 positions |
| 20 to 49 | About 1.5 positions |
| 50 to 99 | Just over 1 position |
| 100 or more | Under 1 position |
Exact medians: 4.62, 2.52, 1.38, 1.22 and 0.93 positions.
A keyword with a handful of searches a day swings roughly five times as much as a keyword with a hundred. None of that extra movement is a real ranking change.
The reason is arithmetic. If a keyword appeared in two searches that day, one odd result moves its average a long way. If it appeared in two hundred, one odd result barely shifts the average at all.
Google’s own advice runs the same way. John Mueller made the point in September 2024, with a line borrowed from mathematics: “there aren’t enough small numbers to meet the many demands made of them”. His practical instruction was to look for long-term trends instead.
Put the two tables together and you get one rule. A movement only means something once you know where the keyword ranks and how much traffic it gets. On its own, the number of positions tells you nothing.
What a real gain looks like
Twenty-eight of these keywords carry at least six months of data. For a typical one, seven months separated its worst month from its best month.
A real gain is a slope you can see across quarters, not a jump you notice on a Tuesday. Some of the movement in this window came from the measurement changes described below, not from the SEO work.
Keyword ranking fluctuations: normal, or a real drop?
On day one you cannot tell the difference between normal movement and a real drop. That is not a reason to give up. The numbers below show that the answer arrives about four days later.
How often an overnight fall reverses
The dataset holds 732 days on which a keyword fell 5 or more positions overnight. A week after each of those falls, the keyword was in one of three states.
| A week after an overnight fall of 5 or more positions | How often that happened |
|---|---|
| Back where it started, or better | About 2 in 5 |
| Partly recovered | About 1 in 8 |
| Still down 5 or more positions | About half |
From 732 overnight falls. Exact shares: 39%, 12% and 49%.
Act on day one and you are acting on a coin flip. If the keyword then recovers on its own, you will still take the credit or the blame for it.
The same thing happens to overnight rises. Of 749 overnight rises of 5 or more positions, nearly 1 in 3 had disappeared within the week. Nobody warns you about a rise, because a jump upwards is not the kind of thing anyone investigates.
Larger falls also sometimes reversed. Of 241 falls of 10 or more positions, more than one-third were back to their earlier level within seven days.
Two keywords, one week, opposite answers
Here are two keywords from the same site, in the same summer. Both fell hard. One of them was noise.
Keyword A is the query agency rank tracker. Its fall reversed.
| Date | Days from the fall | Position |
|---|---|---|
| 29 Jun 2025 | Two days before the fall | 20 |
| 30 Jun 2025 | The day before | 30 |
| 1 Jul 2025 | The fall | 44 |
| 3 Jul 2025 | Two days after | 45 |
| 5 Jul 2025 | Four days after | 42 |
| 7 Jul 2025 | Six days after | 24 |
| 9 Jul 2025 | Eight days after | 24 |
Keyword A was back in the twenties six days after the fall, with nothing done to the page.
Keyword B is a second keyword on the same site. Its fall was real.
| Date | Days from the fall | Position |
|---|---|---|
| 24 Aug 2025 | Three days before the fall | 23 |
| 26 Aug 2025 | The day before | 32 |
| 27 Aug 2025 | The fall | 56 |
| 28 Aug 2025 | One day after | 64 |
| 31 Aug 2025 | Four days after | 66 |
| 4 Sep 2025 | Eight days after | 67 |
| 12 Sep 2025 | Sixteen days after | 73 |
Keyword B kept sliding. Sixteen days after the fall it was worse than on the day of the fall itself.
Keyword B never came back. It sat 30 or more positions down for the next fortnight and was still in the 80s by December.
On day one the two keywords look identical. Both are hard falls, on the same site, in the same summer. By day five they look nothing alike, and that difference is where the five-day rule comes from.
Keyword A also shows how a false lesson gets learned. Anyone who rewrote its page on 1 July would have watched the keyword recover on 7 July and concluded that the rewrite worked. Nobody had touched the page.
React on the same day often enough and you collect evidence for fixes that did nothing. That is the strongest argument for waiting five days.
What Google says about small and large drops
Google’s own drop documentation draws the same line, in positions rather than days. It splits diagnosis into a small drop in position, using position 2 to 4 as its example, and a large drop, using position 4 to 29.
For the small case Google says fluctuations can happen at any time, and that a page can move back up without you doing anything. It recommends “avoiding making radical changes” to a page that is already performing well.
A 12-position fall from position 6 is more likely to deserve investigation. The same-sized fall from position 46 may be less significant, especially for a volatile or low-impression keyword.
What is SERP volatility?
SERP volatility is how much search results across the web are being reshuffled at a given moment, rather than on your site alone. An industry-wide volatility tracker samples a large fixed set of keywords every day. It measures how far those results moved against the day before, then publishes the answer as a single score, usually broken out by industry.
Widespread Google ranking fluctuations do not change your position. They change what your position means.
Say your keyword fell 8 positions on a day when the whole category was moving. You were part of a reshuffle, and the honest next step is to wait. If it fell 8 positions on a quiet day while the surrounding results stayed relatively stable, that is stronger evidence that the cause may be specific to your page or site. It is still a signal to investigate, not proof of the cause.
You can answer the same question from your own data. That is what SERP analysis means in practice: reading the whole results page around your listing instead of reading only your own number.
A SERP ranking history for the top ten will tell you whether everyone moved or only you did. So will a competitor’s serp position history plotted next to yours.
The Google dance, and whether it still applies
New or recently changed pages can fluctuate while Google recrawls, reprocesses and reassesses them. Early movement is therefore less reliable than a sustained trend over a longer period.
The name is out of date, though. The original Google dance was a periodic index refresh you could put in a calendar. Google updates continuously now, so there is no scheduled refresh to wait for. The bouncing is still real. The scheduled refresh the name described is gone.
One myth is worth retiring here. This dataset shows no weekend effect. Every day of the week averaged between position 42 and 44, and weekends drew as many impressions as weekdays. If someone is blaming Saturday for a drop, the data does not support it.
When the graph moves but your rankings did not
On a keyword ranking history graph, a change in how the position is measured looks exactly like a change in ranking.
Twice in seventeen months, RankCaddy’s own ranking history moved further than any of our SEO work moved it. Neither of those two movements was a ranking change.
September 2025: tools stopped being able to load 100 results at once
For years, rank trackers loaded search results 100 to a page. Every one of those results counted as your site appearing in search.
Between 10 and 14 September 2025 Google stopped honouring the parameter that made those long result pages possible, and getting the top 100 became roughly ten times as expensive in page requests. Those deep appearances, the ones only a tool ever loaded, stopped being counted. What was left was closer to what people actually see, which is the results near the top.
| RankCaddy’s whole keyword set | Average reported position |
|---|---|
| July to early September 2025, before | 58 |
| October to December 2025, after | 34 |
The whole keyword set improved by twenty-four positions in a few days, and nobody had done anything to earn it.
Individual keywords moved even further.
| Keyword | Before | After |
|---|---|---|
| accurate rank tracker | 81 | 13 |
| rank tracker free trial | 82 | 27 |
| rank tracking data | 43 | 9 |
Not one of those pages was edited.
We wrote about the reporting fallout at the time, including how to re-baseline a client report around the date. Three numbers moved on the day the change landed. Impressions fell, average position improved as a number, and click-through rate rose.
Click-through rate is clicks divided by impressions, so a fall in impressions lifts it even when nobody clicks more. None of those three changes had been earned.
May 2025 to April 2026: a Search Console logging fault that ran for about a year
Google records known data anomalies in Search Console, and this one is on its list. A logging error stopped Search Console reporting impressions accurately from 13 May 2025 until 27 April 2026, about 50 weeks. It affected impressions and the two metrics calculated from them, click-through rate and average position. Clicks were not affected.
Google’s entry says the issue is resolved, and that a decrease in impressions is what you may notice as a result. Google says the issue was resolved, but the affected historical period should still be treated cautiously when making impression-based comparisons. So the correction arrives on your graph as a fresh drop, rather than as a quiet edit to the past.
In RankCaddy’s own numbers, the fix shows up as a 12-position collapse in the month it landed. No ranking actually collapsed.
Average reported position across the keyword set was 19 before the fix and 31 after.
What to do about a measurement break
Four habits are worth building, and each one takes less time than the argument it prevents:
- Mark the dates of known measurement changes on the graph, before you need them.
- Avoid treating periods on opposite sides of a known measurement break as directly comparable unless you can isolate an unaffected metric.
- Check Google’s data anomalies page before you diagnose a change that looks too large to be your own work.
- Know which source each of your positions came from, keyword by keyword.
What the graph can mark for you
This is where your tool either helps or gets in the way. A rank history graph in RankCaddy marks Google core updates and the days keywords were added, and it takes your own dated events alongside them. It does not label Google’s data anomalies, so the September 2025 marker is one you add yourself, once.
Per-keyword history then shows where each point came from: scraped results, Search Console, or a value carried forward. You choose which source to show. Page 1 stays exact and daily. Beyond page 1 the value comes from Search Console, matched to country and device and marked with a tilde: what we changed in how we track beyond page 1.
None of that makes a number truer. It makes the number readable, and readable is all explaining a ranking dip to a client requires.
How to track keyword rankings over time without overreacting
How often you collect positions and how often you read them are two different decisions. Only the reading needs slowing down. You track keyword ranking fluctuations daily, because daily collection is what makes a weekly view possible. What changes is how often you look at the graph. Seven steps follow, in the order they pay off.
Step 1: Collect daily, read weekly
Weekly rank tracking is not a blurrier version of daily tracking. It is the same data with the daily noise averaged out. A page 1 keyword that moves under 1 position a day has nothing to tell you on a Tuesday.
The rank history graph in rank tracking has a weekly setting for exactly this. It gives one value per seven days, over whatever date range you pick. Change indicators on individual keywords can sit on a weekly or monthly baseline instead of a daily one.
Step 2: Set your alert thresholds by position band
One rule for every keyword, “tell me about any fall of 5 or more positions”, will bury you. That fall happens on 1 day in 22 inside the top 10, and on 1 day in 4 past position 30. Almost every alert you get will come from the keywords you care about least.
Alert rules in RankCaddy take a condition on the keyword’s previous position, and they can be limited to one tag. So one rule can watch page 1 closely while another watches page 4 loosely, against a 1-day, 1-week or 1-month baseline.
Step 3: Check the impression count before you react
A dramatic fall on a keyword with three impressions is a rounding error, not a ranking change. The check takes one glance, and it rules out more false alarms than anything else on this list.
Step 4: Wait five days before you diagnose
Two keywords that look identical on day one look nothing alike by day five. The wait costs you almost nothing, because about 2 in 5 falls repair themselves inside a week.
Investigations are built for that gap. Flag the keyword with a note and a status, assign it to whoever owns the page, and act when the evidence arrives instead of when the alert does.
A fall that overlaps a Google update follows Google’s own rule instead of the five-day rule. Confirm on the Search Status Dashboard that the rollout has actually finished. Then wait “at least a full week after a core update completes” before you compare the week after the rollout with the week before it began.
Step 5: Annotate the graph
Core updates, your own publishes and migrations, and every measurement change all belong on the timeline. A 17-month graph with no markers on it cannot tell you why anything moved. Here is what sat inside this dataset’s own seventeen months.
| Google update | When it rolled out |
|---|---|
| December 2025 core update | 11 to 29 December 2025 |
| February 2026 Discover core update | 5 to 27 February 2026 |
| March 2026 spam update | 24 to 25 March 2026 |
| March 2026 core update | 27 March to 8 April 2026 |
| May 2026 core update | Began 21 May 2026 |
Five confirmed update windows sit inside these seventeen months, and the longest of them ran 22 days.
Dates and durations for each of these rollouts are on the record. Google’s advice is to wait for the rollout to finish, and then a further week. So a fair assessment takes weeks rather than days. Any graph you keep for a year holds several of these windows, whether you marked them or not.
Step 6: Group the keywords and judge the group
Tag your keywords by client, by page type or by intent. Then filter the graph to one tag, so you are reading a single line for a set of pages instead of forty separate lines. Noise lives in the individual lines. When twenty keywords in a tag move together, one bad reading cannot explain it.
Step 7: Decide in advance what would make you act
Write down the threshold, the wait, and who decides, while nothing is going wrong. Otherwise the decision gets made in the moment by whoever is most alarmed.
What to do when a drop turns out to be real
Five days have passed, the fall has held, and the impression count is high enough to trust. Now the SEO ranking drop is worth an afternoon.
Google’s drop documentation lists the causes, and “an update happened” is not one of them. Work down the list in this order and stop when you find your cause. How to recover lost keyword rankings depends entirely on which cause it turns out to be.
| What to check | What “checked” means |
|---|---|
| Whether Google can still crawl, index and serve the page | You have looked at the page indexing report and the URL inspection tool for that exact URL, not at the site total |
| Whether the content still matches what the query now wants | You have read the current top three results and named what they answer that you do not |
| Whether one of your own pages is competing for the keyword | You know which URL Google is choosing this week, and whether it changed |
| Whether the results page itself changed above the organic results | You know which features sit above your listing today and which of them you appear in |
| Whether demand moved rather than your position | You have checked the query in Google Trends and know whether the fall is yours or everyone’s |
| Whether anything moved or was redirected | You can name every URL change in the last quarter and confirm the redirects resolve |
Most of these checks are cheap to rule out. Do them in order instead of starting with a rewrite.
Two of those rows deserve a sentence more. The quietest cause on the list is cannibalisation, which is two of your own pages competing for the same keyword. In a ranking history, cannibalisation shows up as the keyword flipping between two URLs rather than as a fall. The average across the two URLs looks like a mild decline.
If the cause turns out to be the content, the fixes are the ordinary SEO quick wins rather than a rebuild. Give the page a title that matches the query. Add the section that answers whatever the other results on that page are answering. Add an internal link from a page that already ranks.
If the fall is large and hits the whole site rather than one page, stop reading this section. A site-wide drop is a different job, and Google’s drop documentation is the place to start it.
Frequently asked questions
What counts as keyword ranking history?
The record of the positions a keyword has held over time for a given site, device, location and search engine. It is used for three things: diagnosing a fall, proving progress across a reporting period, and checking whether a competitor’s position is new or long-standing. Each of those three jobs is best served by a different source.
Where can I check my keyword ranking history?
Three places, and each has a trade-off. Google Search Console is free and covers your own site, with averaged positions for 16 rolling months. A rank tracker gives consistent daily positions on a fixed device and location, from the day you started. A third-party crawl archive covers competitors, sampled rather than daily. None of them holds history from before you started measuring.
How do I see keyword ranking in Google?
Open the Search Console Performance report, switch on average position, and filter to the query. Searching for the keyword yourself does not count. The results page you see is shaped by your location, your account and your recent searches, so the position you see is true for you and nobody else.
How far back does keyword ranking history go?
Search Console holds 16 months on a rolling basis, then deletes the oldest data permanently, with no archive behind it. A rank tracker starts on the day you added the keyword and never fills in the past. So start tracking before you need the data, and export Search Console before its oldest month falls off.
What is a good keyword ranking?
It depends on the results page, but here is a usable answer. Page 1 is where the traffic concentrates. Positions 4 to 30 are where realistic near-term gains sit, because searchers are already looking for the keyword and your page is already a candidate. The more useful question is which way the keyword has moved over a quarter, not the number it holds today.
Why did my keyword rankings drop?
Often for no reason at all. In our own data, about 2 in 5 overnight falls of 5 or more positions were back where they started within a week, with nothing done to the page. When a fall is real, Google’s causes are technical issues, content and intent, security or spam problems, seasonality, site moves and algorithmic updates. Wait five days before you choose one.
Is it normal for rankings to fluctuate daily?
Yes. Daily SEO ranking fluctuations are normal, and the size of them is predictable from where the keyword ranks. In RankCaddy’s own Search Console data a page 1 keyword typically moved under 1 position a day. A keyword past position 50 moved about two and a half positions. Keywords with few impressions move more again, and none of that movement is a ranking change.
What is a Google dance?
A Google dance is temporary movement while Google re-evaluates new or recently changed content, before its position settles. Publish a page and expect it to bounce for a week or two. The original meaning was a periodic index refresh on a schedule. That meaning is out of date, because Google updates continuously and there is no refresh date to wait for.
Does Google still use RankBrain?
RankBrain was introduced in 2015 as a machine-learning part of ranking. Google has since described a range of systems working together rather than one named mechanism. The practical answer is that no single named system explains a ranking change, so RankBrain is not a useful way to read your own ranking history.
Can I see a competitor’s keyword ranking history?
Yes, but the data comes from a third party’s crawl archive rather than from Google, so you get sampled snapshots rather than a daily series. For your own daily record of a competitor’s positions, a rank tracker can add their domain to your project and chart it beside yours. That record starts on the day you add the domain, and there is nothing for the period before it.
Read the slope, not the day
Keyword ranking history is the most useful chart in SEO and the easiest one to misread. Much of what a daily graph shows is not a ranking change.
Some of it is a keyword with four impressions, where one odd result moves the average. Some of it is a reshuffle that will reverse itself by Friday. And a little of it never happened at all, because what changed was the measurement.
The skill is not reading the number. It is reading the slope, and knowing which stretches of the graph can be compared with each other.
Reading the slope is easier when the graph is set to weekly and carries the dates you marked on it. Compare the plans, which are sized by how many keywords you track each day, and start the 14-day trial.