Lost keyword rankings are the page-level version of a traffic drop: one URL slides on the queries it was built for while the site around it holds. There is a site-wide version too, every folder down from one date, and that one is a different investigation with a different first question; it’s in the website rankings loss audit, and the second step below is where the two part ways. This article is about the page. Which queries it lost, why, what to do, and the cases, more of them than people expect, where the answer is to leave it alone.
The guides ranking for this query mostly don’t make the distinction. They open with the page and within three paragraphs are talking about penalties and Penguin, which were site-wide and in Penguin’s case have been part of the core algorithm since 2016. A page that slipped from 4 to 15 on its own query didn’t get a penalty. It got outranked, or the query changed its mind about what it wants, or the page changed under it, and the useful list of reasons is different for a page that sells something than for a page that explains something. That split is the fourth step, and the three before it exist to make sure you’re looking at the right page and the right queries.
What “lost” means: the rank tracker’s definition and Search Console’s
In a rank tracker, “lost” is a status with a threshold. Ahrefs’ Movements report and Semrush’s Position Changes report both mark a keyword Lost when the domain no longer appears in the top 100 for it, and a keyword can carry that label while the site still ranks: the ranking URL switched to another page of yours, the tracker follows www and Google served non-www, the result blinked out on the day of the check, or it ranks in a country the project isn’t set to. A Lost column full of those is a tracker artifact, and the first thing to do with it is open the Organic Keywords view and see whether the domain is really gone.
Search Console has no “lost” status at all. It has a compare you run yourself, and that compare is where lost keyword rankings become measurable in the plain sense: a query the page used to be shown for, with fewer impressions now and a worse position. The tracker tells you a keyword fell out of a list; Search Console tells you what Google served instead, to whom, and from which URL. The steps below run on the second.
Step 1. Find the lost keyword rankings in Search Console
Performance report, date range set to the last three months against the previous period, filter to the page. The Queries tab now shows each query with position and impressions for both periods, and the column to sort by is impressions in the earlier period, because that puts the queries the page used to be shown for at the top, whatever they do now. Clicks can wait. Impressions say whether Google still shows the page for the query and position says where, and a query whose impressions halved while its position went from 4 to 15 is a lost ranking in the plain sense. A query whose impressions held and whose clicks fell is a different thing, a snippet problem, and it gets its own question later. What the tools sell as content decay is usually several of these on one page, spread over months, and this compare is the manual version of their report.
The average is where the report misleads, and it misleads in two ways that pull in opposite directions. Position is averaged over every impression, and the page-level figure over every query, so a page that starts appearing for a new tail of queries at position 40 sees its average sink with no ranking lost. That one is widely known. The other is in Google’s own definition: when a site holds more than one result for a query, only the topmost one counts toward position. A second URL of yours sitting at 30 for the same query doesn’t drag the average down, which sounds like good news until you notice it also means cannibalization is invisible in this column. It shows up in impressions splitting between two pages, not in position. Which is why the queries get read one by one and the average last; twenty of them carry most pages, and the rest moves the average and nothing else.
One more thing the report won’t give you. A share of the impressions sits under queries Google anonymizes, the rare ones, and on a long-tail page that share can be a third of the total. I have no way to see which of those a page lost. If the visible queries held and the total impressions still fell, that’s where the loss is, and the only move is to treat the page’s topic as the unit instead of the query.
Rank trackers disagree with all of this by design. A tracker reports one location, one device, one time of day; Search Console averages across every impression Google served, and it counts SERP features as positions, so a video carousel at the top pushes the first organic result to 2 in Search Console while the tracker may still call it 1. For finding what was lost, the Search Console number is the one built from what was actually shown. The tracker comes back later, when the page is being watched for recovery and a daily number for one city is exactly what’s wanted.
Step 2. One page, one folder, or the whole site
Before diagnosing lost keyword rankings on a page, check that it’s the page. Filter off, same compare for the whole site; then for the folder the page lives in, /blog/ or /services/ or /catalog/; then brand queries against non-brand, which in Search Console is a regex filter on the query column with the brand name and its misspellings. If the site held and the folder held, it’s the page, and everything from here applies. If the folder fell and the rest held, the cause is shared across a template: a change shipped to that template, a canonical rule, a block removed from the layout, or an intent shift across the folder’s queries, and the method is to diagnose one page and apply the fix to the template, because fixing pages one at a time inside a template problem is a quarter of work for nothing.
If everything fell from one date, stop. That isn’t a page, it’s a deploy, a manual action, a security issue or a core update, and the audit linked above walks those in order with the Search Console reports for each. The fact to carry back from it concerns timing. A site reassessed in a core update tends not to recover until the next one, whatever gets fixed in between; Google’s own guidance says as much, in softer words. The last two core rollouts ran twelve and eighteen days. A page-level refresh shipped inside that window is work that can’t be measured for months, and I’d rather it waited.
Step 3. Reading the compare: four patterns and what each one means
With the page confirmed, each declining keyword falls into one of four patterns in the compare, and the pattern decides which question gets asked next. The order matters, because the first pattern makes the other three moot.
| What the compare shows | What happened | Where to look |
|---|---|---|
| The query’s impressions fell for the whole site, not only this page | The query lost demand, not the page: seasonality, a category going out of fashion, a question an AI Overview now answers in full. A refresh fixes nothing here | Same compare with the page filter off, Queries tab, that one query; then the date picker set year over year |
| Impressions held, position held, clicks fell | Something between the user and the page changed: a SERP feature above it, a rewritten title, structured data that stopped validating and took the rich result with it. The snippet needs fixing, not the page | Search appearance filter in the same report, rich-result impressions on their own; the live SERP for the query |
| Impressions and position fell; the pages above yours are a different type | The intent moved. Service pages a year ago, comparison articles now; guides then, product listings now. Refreshing a page in the wrong format is the most common wasted refresh there is | The top ten listed by page type; the query’s SERP a year ago in a rank tracker’s history, if the project goes back that far |
| Impressions and position fell; the pages above yours are the same type | Another URL took the query. Yours first: two URLs on the site trading the query week to week is cannibalization, and the position column won’t show it. Then a competitor’s: which URL sits where yours was and what it carries that yours doesn’t | Filter by the query, Pages tab, for your own URLs; the table in the next step for the competitor’s |
The first row has a wrinkle that makes it look like the second. Links inside an AI Overview are counted as impressions at the overview’s position, usually 1, so a page quoted in one can gain impressions and position and lose clicks at the same time, and the compare reads like a success. When impressions rose, position improved and clicks fell on a question-shaped query, that is what happened, and no row of the table applies: the page did its job and the overview kept the click.
Step 4. Why keyword rankings drop: the cause depends on the page type
Most cause lists merge commercial and informational pages into one, and a service page ends up being told to add an author bio while an article gets told to check its pricing block. Split by what the page is for, the causes look like this:
| Commercial page (service, category, product) | Article or guide |
|---|---|
| Intent shift. The SERP moved from service pages to aggregators, comparisons or marketplaces, and a single-brand page can’t match a directory | Intent shift. The query moved from informational to commercial, or from a long guide to a short answer the SERP now shows in a feature |
| Blocks missing against the top ten. Pricing, delivery terms, a comparison table, reviews, a calculator, a location grid that the pages above now carry | Subtopics missing against the top ten. Questions the ranking set answers in H2s that the page doesn’t, entities that entered the topic over the year |
| The listing emptied. Products went out of stock, pagination broke, filters got noindexed, and the category page reads as thin | The page restates the top ten. Nothing in it that the pages above don’t say, so when a competitor adds a number or a screenshot, it takes the slot |
| Cannibalization by a new landing page, subcategory or campaign URL built for the same query | Cannibalization by a second article on an overlapping intent, or a service page that now answers the question |
| Lost external links. The directory, partner or supplier page that linked to this one went away or got redirected. Most lost links change nothing; the one that mattered has a date that lines up with the curve | Lost external links. The referring article was deleted, merged or moved behind a redirect that no longer resolves. Same read: referring domains to the URL over twelve months, by date, against the compare |
| Technical. A canonical pointing elsewhere, Product or Offer markup gone, a redirect chain after a URL change | Technical. The same, plus Article markup and a dateModified that moved without the text changing |
| — | Age. The facts are dated, the sources are gone, the author line is empty, and the pages above are from this year |
The technical row is on both sides and goes first, since it’s the one row that can be cleared in an hour without opening a competitor’s page: URL Inspection, the canonical Google chose against the one declared, the status code, the rendered HTML with the content actually in it. The technical SEO audit guide has the status-by-status read. Everything else in the table needs the SERP open next to the page, and for the commercial side the read is specific: the top ten as a set of blocks, which ones every page carries and which none does. That is what the commercial audit does, and it answers the second row in one pass. The listing row is the one no guide writes about because it isn’t an SEO problem until it is: a category that had 80 products and now has 11 in stock reads as thin to the same classifier that reads a 300-word article as thin, and the fix belongs to whoever manages inventory.
Step 5. The fix, matched to the cause
One cause, one fix, and resisting the urge to do all of them, because a page that lost keyword rankings and then gets every treatment at once gets rewritten into a different page, and a different page loses the queries the old one still had.
- Snippet lost clicks. Rewrite the title and description against the queries that still bring impressions; fix the markup so the rich result returns. An AI Overview above the page is not something the page controls.
- Intent moved. Build the page type the SERP shows now, on a new URL if the old page still ranks for other queries, and let the old one keep those. A service page converted into a guide to chase an informational SERP loses the commercial queries and rarely wins the informational ones; the guide goes to the blog and links to the service page.
- Blocks or subtopics missing. Add the ones the top ten carry, in the format the SERP shows for each (a table for pricing, a list for a checklist, a definition block for “what is”), and one thing none of them carry. A refresh that only adds what the competitors already have brings the page to parity, and parity is where it was before it slipped.
- Restates the top ten. The refresh is three things that weren’t on the page or in the SERP a year ago: a number from your own data, a case with a date, a screenshot of the thing the article describes. If there aren’t three, the page is a merge candidate, not a refresh candidate.
- Listing emptied. Restore the inventory or the pagination. If the category is permanently smaller, merge it upward with a 301.
- Cannibalization. Merge into the stronger URL (more impressions on the shared queries over the last twelve months, then more referring domains) with a 301, or split the intents and retitle both. Noindex on the weaker page takes its links out with it over time; a canonical keeps them.
- Lost external links. If the referring page moved, ask for the link on its new URL; if it died, the fix is one new link from a page in the same cluster, internal or earned, and not a campaign. One lost link rarely moves a ranking on its own; a lost link plus a thinner page than the competitor’s does.
- Age. Update the facts that changed, replace the dead sources, name the author, and move
dateModifiedonly if the text moved. Google’s documentation on dates asks for exactly that, and a new date on an unchanged page is the one refresh it can check.
Internal links go with every one of these. One contextual link from the strongest page in the same cluster, anchored on the query, pointing at the refreshed page. Pages that lose a ranking have often lost a link first, when a navigation block or a related-posts widget was rebuilt and nobody diffed the before and after.
Step 6. When not to touch it
Some lost keyword rankings are real and the fix is still to wait. A page younger than six weeks: new URLs get tried on a range of queries and settle, a move from 8 to 20 in week three is Google re-sorting rather than a loss, and a rewrite in that window resets the clock. A core update mid-rollout, for the reason in step 2; diagnose from the data after the end date on the Search Status Dashboard. A query set that changed shape, so the average fell while the original queries held. And seasonality, which the compare catches only if it’s run year over year, an option the date picker offers and almost no one uses; a January drop on a page about a summer product stops being a drop the moment the comparison period is last January.
Step 7. Lost keyword rankings recovery: what it looks like and how long it takes
Request Indexing after the change gets the page recrawled within days, within the daily quota, which Google doesn’t publish and which runs out after about a dozen URLs. Then the signals come back in an order. Impressions on the lost queries first, often inside two weeks, because Google is trying the page again. Clicks next, if the snippet is right. Position last, and the average looks worse before it looks better while the page is tested on more queries than it had. Judge the refresh at four to eight weeks on the same compare and the same queries, impressions before position, and this is where the rank tracker earns its place: one location, one device, daily, because the Search Console average hides the day the page moved.
| Cause fixed | First signal | Judge it at |
|---|---|---|
| Technical (canonical, markup, redirect chain) | URL Inspection shows the new canonical or the valid markup within days of the recrawl | One to three weeks |
| Snippet (title, description, rich result) | Clicks on the held queries, within the first full week after recrawl | Two to four weeks |
| Content (blocks, subtopics, the three new things, age) | Impressions on the lost queries, often inside two weeks; position last | Four to eight weeks |
| Cannibalization merged with a 301 | The surviving URL takes the merged queries’ impressions as the old one drops out of the Pages tab | Four to eight weeks after the redirects are right |
| A site reassessed in a core update | Usually nothing until the next rollout, whatever was fixed | The end date of the next core update on the Search Status Dashboard |
Those are patterns from pages I’ve watched, not a mechanism; no one outside Google has that. A page that gets its impressions back and not its position has reached parity. It re-entered the set and sits where pages like it sit. The next move is the one thing the top ten doesn’t carry, from the brief, not another round of adding what they already have; a second round of parity work is where refresh budgets tend to go, and it doesn’t move anything.
Lost keyword rankings FAQ
Do lost keyword rankings come back on their own?
In three situations: a rollout that hasn’t finished, a young page still settling, and a query whose demand is seasonal. Outside those, no. Whatever took the page’s place is still there, and it isn’t leaving on its own.
Is a drop from 3 to 8 a lost ranking?
Not in the tracker’s sense, where lost means out of the top 100, and not in Search Console’s, where the page is still shown for the query. It is a lost click: on most queries the move from 3 to 8 cuts clicks by more than half while impressions barely change, so it shows in the compare as the second pattern in step 3, and the questions are the same ones: what sits at 3 now, and what does it carry that the page doesn’t.
Should I rewrite the page or delete it?
Rewrite if the query still wants this page type and there are three things to add that the top ten doesn’t have. Merge with a 301 if another page on the site answers the intent better, or if the page restates the top ten and there’s nothing to add. Delete only when the query is gone and no page should own it, and even then a page with referring domains is worth redirecting rather than returning a 404 for.
How long does recovery take?
Days for the recrawl, two to eight weeks for impressions and position after a page-level fix, and until the next core rollout for a site that was reassessed in one. The honest version of the last one is that nobody outside Google knows the mechanism, only the pattern.
This is the page-level half of the work. The site-level half is the audit, and both run inside our SEO service, where a lost ranking is a line in the quarterly cluster review rather than an emergency.