The call usually starts the same way: “We got hit by the update.” A drop in Search Console, a date that lines up with something on the Search Status Dashboard, and a conclusion. Then the decomposition shows brand queries flat, non-brand down across one folder, and the folder’s canonical tags pointing at a staging domain since a deploy two days before the update began. The update was a date. The canonical was the diagnosis.
A website rankings loss audit is the procedure that keeps those two apart: it establishes what dropped, where, when and by how much before it names a cause. Most guides for this term list the same eight causes and check them in parallel. The steps below run them in the order of what it costs to rule each one out, because the cheap checks catch most of what comes in, and the expensive one, “it was the algorithm,” is also the one nobody can prove.
What a website rankings loss audit is
A website rankings loss audit is a diagnostic read of Search Console, analytics and crawl data that locates a loss in organic visibility by date, segment and metric, rules out causes from cheapest to most expensive, and ends in a plan with a cause, the evidence for it, an owner for the fix and a date to measure again.
A traffic loss audit is wider, since it covers every channel; this one stays on organic. A full SEO audit reads the whole site; this one reads only what changed, and only where the line bent.
Step 1. Which metric dropped: positions, impressions or clicks
Rankings, impressions and clicks are reported together and lost separately. Positions can fall while clicks hold, when the queries that slipped were the ones nobody clicked. Clicks can fall while positions hold, when the result page grew a feature above you, an AI Overview, a video carousel, a forum block, and the click went there. And impressions can rise while clicks fall, for a reason that has been in the documentation since June 2025 and in very few reports since.
HOW SEARCH CONSOLE COUNTS AI MODE
AI Mode has counted inside the main Performance report since June 2025, and the documentation added with it says a user who asks a follow-up question in AI Mode is “essentially performing a new query”: its own impressions, a position that starts again at 1, and usually no click. A page cited in a long conversation logs many impressions and reads as growth, impressions up on queries you never targeted, clicks flat. Since 3 June 2026 a separate Generative AI performance report splits AI Overviews and AI Mode out of that total; it carries impressions and clicks and no average position.
Search Console shows which of the three moved in one view: Performance → Compare, previous period, with impressions, clicks and average position on the same chart. The gap between the three lines is the first piece of evidence, and the rest of the audit reads differently depending on which line bent first.
Step 2. Verify the drop is real
The cheapest check, and the one most often skipped. A rank tracker that swapped its location setting, a Search Console property that lost a subdomain, a GA4 tag that fell off the template in the redesign: each produces a cliff that looks like a ranking loss and isn’t. Two comparisons settle most of it. Analytics against Search Console: if GSC clicks are flat and sessions fell, the loss is measurement. Google against Bing: if Bing Webmaster Tools shows the same drop on the same date, the cause is on your site, because Bing didn’t run Google’s update; if only Google moved, the cause is on Google’s side or in how Google reads you.
Then seasonality, year over year rather than month over month, because “down 20% from July” is the shape of most B2B traffic in August. Then the country filter, since a US property reading a UK-only drop looks like a mild site-wide one.
Search Console’s interface holds 16 months, enough for one year-over-year read and not two. The Bulk data export to BigQuery has no cap, and a site that switched it on in 2023 has three years of daily query and page data. It should have been running before the drop. If it wasn’t, the baseline is whatever the rank tracker kept, and the audit says so in its first paragraph.
Step 3. Locate it: date, segment, brand against non-brand
Once the drop is real, it gets located before it gets explained. The question has five parts: when it started, where on the site, which queries, how much, and in which metric. The pattern the five answers form points at the cause more reliably than any single check.
When is the first date the line bends, which is rarely the date somebody noticed. Daily, read against two calendars: the site’s deploy log and the Search Status Dashboard. Where is a regex on the folder or template in the Performance report’s Page filter, one segment at a time. Which queries starts with brand against non-brand, because those two move for different reasons: in the Query filter, Custom (regex) with the brand name and its common misspellings gives brand, and the same regex under Doesn't match regex gives everything else. How much is position loss against click loss per segment; a page that fell from 3 to 5 and lost 60% of its clicks tells a different story from one that fell from 3 to 5 and lost 10%.
| Pattern in the decomposition | What it usually means | Where to look next |
|---|---|---|
| Everything down, brand included, overnight | Technical, reporting or security: Google stopped reaching or trusting the site | Steps 2, 4, 5 |
| Brand flat, non-brand down site-wide | A search-side reassessment: update, site-wide quality signal, or a link set that stopped counting | Steps 6, 7 |
| One folder or template down, the rest flat | A rule shipped in that template: canonical, noindex, rendering, internal links removed | Step 4, deploy log; on large sites the enterprise SEO audit reads it per template |
| One country or language down | hreflang or canonical conflict, geo-redirect, a locale folded into another | Indexation per locale |
| Mobile down, desktop flat | Rendering or mobile parity: what the mobile version lacks, the index lacks | Raw vs rendered crawl on a mobile UA |
| Positions flat, clicks down on a query set | The result page changed above you: AI Overview, video, forums, a new SERP feature | Search appearance filter, then Step 8 |
| Impressions up on new queries, clicks flat or down | Cited inside AI answers without a click; AI Mode follow-ups counted as new queries | Search appearance; query list sorted by impressions with zero clicks |
A large tail moving from Indexed to Crawled – currently not indexed, no template change | A site-wide quality reassessment: the tail reclassified as adding nothing | Step 7 |
| A cluster of related queries down, one page holding them | Displacement by a competitor page, or by your own newer page | Step 8 |
| A slow slope over months, no event on either calendar | Decay: content aged, internal links thinned, the SERP moved without you | Step 9 |
“Page authority drop” gets its reading here too. When a page loses positions after a redesign and its external links are unchanged, the authority it lost is internal: the links from the navigation, the category page and the three related articles that used to point at it went out with the old template, and the page now sits four clicks deep with two inlinks. That shows in a crawl before and after, and it is a template finding, which is where Step 4 starts.
Step 4. Rule out access and site changes against the deploy date
Can Google still fetch, render and index the pages that dropped? This is the technical audit, read here for one thing only: what changed, and on what date. The Pages report against the deploy log: a jump in Excluded by 'noindex' tag, Blocked by robots.txt, Redirect error, Soft 404, Alternate page with proper canonical tag or Duplicate, Google chose different canonical than user dated to a release is the finding. Each status names its mechanism. A staging robots.txt shipped to production. A canonical rule in one template pointing every page at its parent, which is how this article opens. A JavaScript framework upgrade after which the raw HTML ships an empty shell, so URL Inspection’s HTML tab shows nothing where the screenshot shows a page. A deleted section redirected wholesale to the homepage, which Google reads as soft 404s. None of these needs an update to explain it, and each coincides with an update often enough that the update gets the blame.
The same rung has a slower version: Google still reaches the site but visits it less. Crawl stats (under Settings, 90 days of history, so pull it before the window closes) shows requests falling from a date, the logs show verified Googlebot dropping off one template, and the Pages report shows the tail sliding into Discovered – currently not indexed. Positions fall afterwards because the index holds a version of the pages that is weeks old. The cause is usually upstream of the pages themselves, a response-time regression or a parameter explosion that shipped with the same release, and that layer belongs to the technical SEO audit.
What the site itself changed is the rest of this rung: the migration nobody mentioned, the rewrite that removed the FAQ block holding the featured snippet, the merge of two pages into one that answered neither. The Wayback Machine, page by page for the pages that dropped, against the current version. Migrations produce the largest drops here and carry a signature of their own: the old URLs gone from the Pages report, the new ones in Discovered – currently not indexed, and a redirect map that covers the top 200 URLs and none of the long tail that carried the traffic.
Step 5. Manual actions and security: the two causes that announce themselves
Two reports, thirty seconds. Security issues catches the hacked-page and injected-redirect cases. Manual actions is empty for almost every site that comes in convinced it was penalized; when it isn’t, the report names the policy, the fix is the reconsideration path, and the rest of this audit waits. A manual action is the rarest cause on this list and the one people check for first.
Step 6. Was it the update? The rollout window and the competitor-set test
Only now does the calendar matter. If the first bend in the line sits inside a confirmed rollout window on the Search Status Dashboard, the update is a candidate. It isn’t a conclusion, because coincidence is cheap: rollouts run one to three weeks (the March 2026 core update ran 12 days and 4 hours from 27 March; the December 2025 core update, 18 days and 2 hours from 11 December), sites deploy every day, and a drop that starts inside a window has a deploy inside the same window more often than not, and the deploy is checkable while the update isn’t.
Treat “the update” as a hypothesis and ask what would confirm it. An update that reassessed you almost certainly reassessed sites like you, and that gives the hypothesis a test:
Google’s own guidance on core updates says sites that lose ground typically don’t recover until a subsequent core update, and sometimes recover then without changing anything. On the timescale a business needs an answer, that makes the explanation unfalsifiable. During a rollout, change nothing on the pages that dropped, because whatever you change, you won’t know which of the two moved the line. Afterwards, keep reading down the list for what the reassessment landed on, and if the read comes back empty, the pages as good as their replacements by every measure available, the report says so and names the date to measure again. That is a result, and the one clients like least, which is why the slot so often gets filled with a cause.
Step 7. Reassessments that send no letter: site-wide quality, scaled content, links
Between the update rung and displacement sits a class of cause the top-ranking guides fold into “the algorithm”: Google’s spam and quality systems mostly reclassify and discount rather than penalize, without a notification, and on the line the effect is indistinguishable from an update until you decompose it. Two shapes account for most of what I’ve read this year, and lost links and neutralized links, which look alike on the line, sit at the end of the second.
Site-wide quality. The helpful-content signal has been site-wide since 2022 and part of core ranking since March 2024; scaled content abuse has been a named spam policy since the same month, with the August 2025 spam update and this year’s three aimed at that shape. From inside: brand flat, non-brand down across the site rather than one folder, and a wide tail of templated or generated pages sliding from Indexed to Crawled – currently not indexed with no template change on the deploy log. The pages that carried answers usually hold; the tail that restated the SERP gets discounted and drags the templates that sell with it.
A blog that grew from under 200 ranking URLs in 2023 to about 600 by spring 2026, on articles commissioned to “close topical authority” around one service. Traffic peaked near 32K a month in August 2025, the month of that year’s spam update, and slid through the autumn while the page count kept rising. That divergence is the signature: production continued, and each new page was scored against a SERP the site had already restated. The 24 March spam update (19 hours and 30 minutes) and the core update that started on the 27th (12 days and 4 hours) took most of what remained, and the page count itself turns down at the right edge of the chart, which is the tail losing its rankings. The test for this shape runs per template: score a sample of each template’s pages for information gain against the SERPs they targeted, and the templates that come back as echo are the ones the classifier found.
Links that are still there. A drop that the backlink tools can’t explain, because the referring domains haven’t moved, is often a link set that Google stopped counting. Since December 2022 Google has said its systems neutralize links they classify as bought or placed to pass weight; neutralized links stay in the profile and stop counting, and no report says which.
This one came in as “the December core update.” The profile behind it was bought links at volume, exact-match commercial anchors on most of them, from domains with nothing else on them, and none of it had moved in the backlink tools. The dates are the finding. The bend is in late October, six weeks before the window opens on 11 December; the second leg runs through the rollout and past it into February, where the line flattens at roughly a third of the October peak. An update that starts six weeks into a decline didn’t start it. What fits the shape is a link set discounted in stages, and the update was the label the owner had for it.
| Lost links | Neutralized links | |
|---|---|---|
| In the backlink tools | Referring domains fall, with a date: a site closed, a partner removed a page, a redirect chain broke | Nothing moves. Referring domains flat, DR flat |
| In Search Console | Positions fall on the pages the lost domains pointed at, from the date they left | Positions fall in stages, on the pages whose rankings sat above what their content holds alone; brand flat |
| What the donor looks like | A page that had search traffic of its own and linked from its body | A “partner content” folder on a news site, a blog that exists to sell placements, a page with no traffic and a commercial anchor off its own topic |
The read is per referring page, not per domain. For the pages that dropped, pull their linking pages and check three things on each: whether the donor page has search traffic of its own, whether the anchor fits what the donor page is about, and where the link sits, in the body of a page the site links to or in a sponsored folder nobody navigates to. A news site’s partner section and a blog built to sell placements fail all three; those are the links the December 2022 systems were built to classify, and the backlink tools still count them. The 2024 leak has fields named anchorMismatchDemotion and siteAuthority; what they do is unconfirmed, and nothing here rests on them. A disavow adds nothing to a neutralized set, since Google already stopped counting it; the recovery is the content-versus-links read, page by page, and the mechanics of what passes and what gets zeroed are in the 301 redirect piece.
Neither shape is a penalty, and neither appears in Manual actions. Both are recoverable, and waiting for the next core update recovers neither.
Step 8. Who took your place: SERP features, a competitor URL, your own page
This is the section most guides reduce to “check competitors,” and it is where the audit either finds something or admits it hasn’t. For each query cluster that dropped: who holds the positions now? Three kinds of answer.
A different kind of result. The query used to return ten articles and now returns an AI Overview, three videos and a Reddit block above the first article. Positions may be unchanged; the click budget above the fold went elsewhere. Search Console’s Search appearance filter shows which features your pages appeared in; the SERP itself, fetched from the right country, shows what sits above you now. Whether the intent is still yours to win in that format is a decision, and it isn’t one made on the page.
A competitor’s page. Not “competitors improved” but a URL, for each query, that now ranks where yours did. Pull it and read it against yours the way a retrieval model would: what it covers that yours doesn’t, which entities and passages it has that yours lacks, what it adds against the rest of the top ten that yours restates. I run that as an information-gain score, your page and the replacement against the same SERP corpus, and the two numbers usually end the argument: a page that was original against last year’s top ten and is an echo of this year’s has been displaced by the SERP moving. The method is on the information gain page; here it is one page against one replacer.
Your own page. The most common displacement on sites with a blog, and no rank tracker reports it as a loss, because the domain still ranks. The page that dropped was replaced by a newer page on the same site answering the same intent, and Google picked the newer one. In Search Console it shows as the query’s top page switching URL on the date of the drop. On the topical map it is a pair of pages above the cannibalization threshold (0.995 cosine for a within-cluster pair, on our last run), both holding impressions on the same queries. A merge with a 301 to the survivor fixes it, or a split of intents. A rewrite of the older page keeps both in the index and changes nothing about which one Google picks.
Step 9. No date on the drop: decay and intent drift
A page that held position 4 for two years and now holds 9, with no bend in the line and no event on either calendar, has decayed, and decay has three mechanisms that get confused for one. The content aged: the facts are from 2023, the tools it names have been renamed, three newer pages in the top ten carry this year’s numbers. The internal links thinned: every redesign and every new article since took one or two links away, and the page’s internal PageRank halved without anyone deciding it should. Or the SERP moved: the intent behind the query shifted from “what is” to “which tool,” and the page still answers the old question well.
Decay is measured against the page’s own peak, because a cluster average hides it: three pages growing and two decaying net out to flat. A content audit that grades each page on its impressions-to-peak ratio surfaces the decaying ones as a list, and which of the three mechanisms a page shows decides between refresh, merge and leave. Before refreshing, apply the date test to the page: three things in it that weren’t true a year ago, or the refresh is cosmetic and the slope continues.
The report, and how long recovery takes
One line per finding, five fields. The cause, in a sentence. The evidence: a status count dated to the deploy, a before/after crawl diff, the replacer’s score, the cannibalization pair. The segment the cause explains and how much of the loss it carries. The fix and its owner, since a canonical rule is engineering’s, a merge is content’s and a redirect map is the platform team’s. And the date to read the same segment again. A rankings loss audit that names a cause without a re-measure date is a hypothesis dressed as a report.
| Cause | Recovery runs on | Where it shows first |
|---|---|---|
| Technical (Step 4) | Google’s next crawl of the affected pages: days to weeks | The same Pages report status, falling |
| Displacement, decay (Steps 8, 9) | A content decision and a build; then a recrawl | Query-level position on the rebuilt page |
| Update, reassessment (Steps 6, 7) | A later core update, or not at all; no date anyone can give | Non-brand impressions on the discounted templates |
Website rankings loss audit checklist
| Check | Where to read it | What counts as a finding |
|---|---|---|
| Which metric moved | Performance → Compare, three lines | The lines diverge: positions without clicks, clicks without positions, impressions without either |
| Reporting | GSC clicks vs GA4 sessions; Bing Webmaster Tools | Sessions fell, clicks didn’t; Bing fell on the same date |
| Seasonality, country | Compare, same period last year; Country filter | The drop vanishes year over year, or sits in one country |
| Date of the first bend | Daily chart against the deploy log and the Search Status Dashboard | A release within days of the bend |
| Brand vs non-brand | Query filter, Custom (regex) / Doesn’t match regex | Brand flat while non-brand falls |
| Folder, template, device | Page filter regex; Device filter | One segment carries most of the loss |
| Access and indexing | Pages report by reason; URL Inspection HTML vs screenshot; Crawl stats | A status jumping on the release date |
| Site changes | Wayback Machine, page by page; redirect map against the old crawl | Content, blocks or URLs removed before the bend |
| Manual actions, security | The two reports | Anything at all |
| Update | Rollout window; five competitors’ visibility over the same fortnight | The bend inside the window and the competitors flat |
| Quality, links | Crawled – currently not indexed over time; referring pages per dropped URL | Tail sliding without a template change; donors with no traffic, off-topic anchors |
| Displacement | Search appearance; the SERP from the right country; top page per query over time | A feature above you, a replacer URL, or your own URL switching |
Frequently asked questions
Rankings dropped after a migration. What first?
The Pages report for the old URLs and the new. Old ones should read Page with redirect, new ones Indexed. Anything else, old URLs still indexed, new ones in Discovered – currently not indexed, Redirect error, is the migration, and the redirect map is where it broke.
Should we build links to recover?
Only if the audit found a dated link loss, a referring-domain set that disappeared on a date. Even then, the links that left were probably the reason the page ranked above its content. Links are the last rung, and most drops never reach it.
Can you guarantee recovery?
No. The audit can guarantee a cause with evidence behind it, or a statement that the evidence doesn’t reach one. Recovery is what the fix does, and some causes, a SERP that changed format, have no fix on your side.
What I don’t have
A way to prove that an update caused a drop. There is no control group for a single site; the competitor-set test is the nearest thing, and it says whether the category moved, not whether Google reassessed you specifically or why. When the ladder reaches the update rung and the displacement read comes back even, the report says “coincides with the update; no site-side cause found; re-measure on this date,” and I’d rather write that sentence than a paragraph of E-E-A-T advice that fills the space and explains nothing.
The other number I’d like is how often “it was the update” turns out, on decomposition, to have been something on the site. In the cases I’ve read it is most of them, and I can’t give a share, because I only see the ones that come in with the update already blamed; the sites where the update really was the cause and nothing else moved don’t call. If a drop of yours is still explained by a date, the decomposition above, on your Search Console and your crawl, is what our SEO audit starts with.