How to diagnose and recover from a drop in search traffic

The first rule is to change nothing on the site until you know what you are fixing. Every change made in a panic adds a new variable, and some of them (deleting pages, adding noindex, mass redirects) are hard to undo. Spend the first day proving the drop is real.

There are two very different kinds of drop. In one, fewer people are arriving from Google. In the other, the same number of people are arriving but fewer of them are being counted. The second kind is common, and the fix for it sits in the tag manager or the consent setup.

Compare Search Console with analytics

Search Console and Google Analytics measure different things. Search Console reports what happens in Google Search before the visit: impressions and clicks. Analytics reports what people do once they arrive. The closest pair of numbers is Search Console clicks and Analytics sessions from organic Google search, and they never match exactly, for several reasons: Analytics depends on the tag being installed and firing, users can decline tracking, the two tools use different time zones, Search Console reports clicks against the canonical URL while Analytics records whatever URL carried the tag, and the two count attribution differently.²

You are not trying to make the two numbers agree. You are looking at whether they moved together. Put daily Search Console clicks and daily organic Google sessions side by side for the last 90 days. If both fell on the same day by a similar share, people really did stop arriving. If analytics fell and Search Console clicks did not, the problem is almost certainly measurement.

Step by step: put the two side by side

In Search Console, open Performance, then Search results. Make sure the search type is Web and tick Total clicks and Total impressions above the chart. To compare periods, open the Date filter, select the Compare tab, choose the two periods and click Apply.³ Use the export button at the top of the report to download the result as Google Sheets, Excel or CSV. The export includes both the chart and the table data, with your filters applied, but tables are limited to 1,000 rows of representative examples.⁴ For a daily total that limit does not matter: the chart data gives you one row per day.

In GA4, open Reports, then Acquisition, then Traffic acquisition. The report's default dimension is Session default channel group, and one of its groups is Organic Search.⁵ Click Add filter at the top of the report to narrow it to Organic Search, or change the dimension to Session source / medium and look at google / organic on its own, which separates Google from Bing and other engines.⁵ To compare with last year, open the date picker at the top right, choose your range, switch on Compare, pick Previous year and click Apply. Export the daily sessions and paste them into the same sheet as the Search Console clicks.

Mark what you find as you go. GA4 lets you add an annotation to a report: in Reports, open a report with a line graph, right-click a data point, click Add annotation, give it a title, a date or date range and a colour, and click Create annotation. You need the Analyst role or above on the property, and an annotation shows across all reports with line graphs.⁶ Add one for the start of the drop and one for every change you find in this tutorial.

Check the things that break tracking

Several common changes make analytics fall without any change in real traffic:

• The analytics tag was removed or duplicated in a template change, a theme update or a move to a new tag manager container. Pages built from that template stop recording, or record twice and then get "fixed".
• A consent banner was added, replaced or reconfigured. Google's consent mode lets tags adjust to a visitor's choice. In the basic implementation, Google tags stay blocked until the visitor consents, so nothing is recorded for people who decline. In the advanced implementation, tags load before the banner and send cookieless pings when consent is denied, which Analytics can use for modelling.⁷ Moving from one setup to the other, or from no banner to a banner, can change recorded sessions sharply overnight.
• Filters, internal traffic rules or channel groupings were edited, so traffic is still recorded but now sits in a different channel.
• A redirect or checkout domain change strips the referrer, so organic visits start appearing as direct.
• A new cookie or privacy tool in the browser or on the site blocks the tag for a share of visitors.

Ask whoever manages the tag manager and the consent banner for their change history covering the week before the drop. Most tag managers keep a version history with dates, which is often the fastest way to find the answer.

Check consent mode in GA4 and Tag Assistant

Two checks tell you whether consent is changing what GA4 records. First, in GA4 open Admin and, under Data collection and modification, click Consent settings. The page shows which consent signals GA4 is receiving, split into advertising signals and behaviour analytics signals, and you can select a data stream to see the detail for it. It may also show the share of your traffic that comes from the EEA.⁸ A behaviour analytics signal that was active and is now missing, or the reverse, is worth noting against the drop date.

Second, check the live site with Google Tag Assistant. The steps are to open Tag Assistant, enter the site's URL so it opens in a new tab, interact with the cookie banner, then look at the Consent events in the summary. The earliest consent event should show ad_storage, ad_personalization, ad_user_data and analytics_storage with an on-page default of Denied, and the latest, after you accept, should show them as Granted.⁹ If the default is Denied and nothing ever updates, or analytics_storage is missing, the banner and the tags are not talking to each other, and recorded sessions will fall with no change in real traffic.

Rule out seasonality and demand

Widen the date range in the Performance report to 16 months, so you can see whether the same dip happens every year, and compare periods with the Compare option.¹ A month-on-month comparison is close to useless for a seasonal business. Compare the affected weeks with the same weeks last year, and look at the shape of the year as a whole.

Google Trends shows whether the drop reflects a wider change in what people are searching for, or something that is happening only to your site.¹ Enter the two or three queries that brought the most clicks before the drop, set the country and the time range to match, and see whether interest fell at the same time. If it did, your traffic followed demand down, and there is nothing on the site to fix.

One more cross-check takes ten minutes if the site is verified in Bing Webmaster Tools. Its Search Performance report shows clicks, impressions and click-through rate from Bing.¹¹ As a rule of thumb, if Bing traffic fell on the same day by a similar share, the cause is more likely on your own site or in demand. If Bing held steady while Google fell, look harder at what changed on Google's side.

Know the limits of Search Console data

Search Console is the most reliable source you have for search traffic, but it has limits that can mislead you in the first days of a drop:

• It is not real time. Collected data should normally be available in two to three days, and the most recent periods can include preliminary data that may still change.¹² Do not judge a drop on the last two days of data.
• Days follow California time. The Performance report labels each day by local time in California, so a UK day and a Search Console day do not line up exactly.¹²
• Some queries are hidden. To protect user privacy, the report does not show all data. Anonymised queries are still counted in the chart totals, but they disappear when you apply a query filter, so filtered totals will add up to less than the whole.^{12 13}
• Tables stop at 1,000 rows in the interface. Some rows may be omitted, the API gives access to more, and bulk data export gives the fullest list of queries.^{12 13}
• Search Console occasionally logs data wrongly, and a public page of data anomalies lists known logging errors and the dates they affected.¹⁴ Check it whenever a drop starts on a precise date with no other explanation.

The example makes a point worth holding on to: a drop can have more than one cause at once. Do not stop at the first plausible explanation. Check how much of the fall it accounts for.

Once you know the drop is real, the most useful single fact is the day it started. A drop that begins on the morning a release went out is almost always about the release. A drop that begins two days into a confirmed core update and builds over a fortnight is almost always about the update. A drop that slides gradually over three months is usually about competitors, demand or slow content decay.

Find the start date and the shape

Switch the Performance report to daily data for the last three months. Look at clicks and impressions separately. Mark the last normal day and the first abnormal day. Then describe the shape in words: a cliff (a sudden step down that stays down), a slope (a decline over weeks), a step down that partly recovers, or a series of steps. The shape matters because different causes leave different marks. Technical faults and manual actions tend to produce cliffs. Core updates often produce changes that build over the rollout period. Competitor gains and demand shifts tend to produce slopes.

Check Google's updates

Ranking updates, and incidents affecting crawling, indexing and serving, are posted on the Google Search Status Dashboard, whose ranking history lists each named update with the date it started and how long it took to roll out. Recent entries include the March 2026 core update, which started on 27 March 2026 and took about 12 days, the May 2026 core update, which started on 21 May 2026 and took about 12 days, and spam updates in June, August and September 2026.¹⁵ Trade press reports the confirmations as they happen; Search Engine Roundtable, for example, reported Google's confirmation that the March 2026 rollout was complete on 8 April.¹⁶ Check the history for any update that overlaps your drop, and note its start and end dates, not just its name.

Wait at least a full week after a core update finishes before analysing its effect, then compare the week after it ended with the week before it began.¹⁷ If you are looking at a drop during a rollout, record it, but do not draw conclusions until the update has finished and the week has passed.

Not every change is announced, and smaller updates happen beyond the named core updates.¹⁷ An unexplained drop that does not line up with anything on the dashboard can still be algorithmic, but you should treat that as a conclusion of last resort, after the technical and site-change explanations have been ruled out.

Check your own changes

The most common causes of sudden drops are changes the site made to itself. Build a list of everything that changed in the four weeks before the drop and the days after it:

• Code releases and deploys, including small ones to templates, navigation, headers or footers.
• CMS, theme or plugin updates, especially SEO plugins, caching plugins and anything that writes robots meta tags, canonicals or sitemaps.
• Redesigns, new templates, changes to internal links or navigation, and removed sections.
• URL changes, domain changes, HTTPS or subdomain moves, platform migrations and redirect changes.
• Hosting, CDN, firewall or bot protection changes. A new rule that challenges or blocks automated traffic can block Googlebot too; when your CDN blocks AI crawlers covers how that happens.
• Content changes: pages merged, deleted, rewritten or noindexed, or large batches of new pages published.
• Changes off the site: a paused advertising campaign that drove branded searches, a product withdrawn, a press story.

Release notes and deploy logs are often thin. Ask the developers directly, check the repository history if you have access, and look at the CMS revision history for the pages that lost the most clicks.

Get the data out of Search Console

The interface is fine for a first look, but segmenting a drop properly means getting the data into a spreadsheet or a database. There are four routes, and which one you need depends on the size of the site.

In Data Studio (the product many people know as Looker Studio), the setup is: click Create, then Data Source, choose the Search Console connector, pick the site, pick Site Impression or URL Impression, choose the search type and click Connect.¹⁹ Site Impression is aggregated by property and URL Impression by page, so use URL Impression for anything you want to split by template.

Bulk export is worth switching on today, even in the middle of a drop, because it only collects data from the day it starts. The setup steps are: in the Google Cloud project, enable the BigQuery API and the BigQuery Storage API; in IAM, grant the principal search-console-data-export@system.gserviceaccount.com the BigQuery Job User and BigQuery Data Editor roles; then in Search Console open Settings, then Bulk data export, and enter the project ID (not the project number), a dataset name and a location. The first export arrives up to 48 hours later.²¹ Data accumulates indefinitely unless you set a partition expiration, which should be 14 days or longer.²¹ For the months before you switched it on, use the API.

Lay the data out so the drop shows itself

However you export, build the same sheet. One row per page per period, with these columns: page URL, template (product, category, guide and so on, assigned by a lookup on the URL path), clicks before, clicks after, change in clicks, impressions before, impressions after, change in impressions, position before, position after, CTR before, CTR after. Make a second tab with the same layout for queries, plus a column marking each query as branded or non-branded. Then add a pivot table that sums the change in clicks by template. This pivot is often where the investigation narrows, because one or two rows turn out to hold most of the loss.

Keep the before and after periods the same length and, where you can, the same days of the week. For a seasonal site, add a third period: the same weeks last year. Sort each tab by change in clicks, smallest first, so the biggest losers sit at the top.

Start a drop log

Keep every event and finding in one dated log from now on. It stops the investigation going in circles, it shows stakeholders the reasoning, and when traffic moves again in six months it tells the next person what has already been tried. Use the template below and fill in one row per event, whether it is a Google update, a release or a finding from the data.

A site-wide total hides almost everything. A fall of 15% overall might be a 60% fall in one template and no change anywhere else, and those two situations have completely different causes. Filtering the Performance report by search type, pages, countries, devices and search appearance shows where the change happened.¹

Segment by page type and template

Most sites are built from a handful of templates: home page, category or service pages, product pages, articles or guides, location pages. Use page filters (URL contains /products/, for example, or a custom regular expression) to split the drop by template. If the URLs do not reveal the template, export the pages table with the comparison and tag each URL by type in a spreadsheet. The question is simple: is the drop spread evenly, or concentrated?

In Search Console, click + Add filter in the filters row, choose Pages, then either URLs containing (for a plain string such as /products/) or Custom (regex) for a pattern, then click Apply. The plain match is not case-sensitive but otherwise exact; regex filters use RE2 syntax, match anywhere in the URL unless you anchor them with ^ or $, and are not case-sensitive unless you start the expression with (?-i). To exclude a pattern, choose Doesn't match regex.³

Once a template filter is on, the chart and every tab below it (Queries, Pages, Countries, Devices, Search appearance, Dates) show only that template. Add the date comparison and the table shows each row for both periods, ready to export and sort to find the biggest losers inside the template.

Segment by query type: branded and non-branded

Branded searches (people looking for you by name) and non-branded searches (people looking for what you sell) fall for different reasons. Search Console now has a branded queries filter in the Performance report that splits data into branded and non-branded, across web, image, video and news search. The classification is made by an AI-assisted system that includes the brand name in all languages, typos, and queries that refer to a unique product or service of the site without naming it, and that some queries may be misidentified.²² The filter is not available for sites with a low number of impressions and that its data history starts from when it was introduced in March 2025.¹³

If the filter is not available for your property, build the split yourself with a query filter using a regular expression of your brand name and its common misspellings. Remember that query filters exclude anonymised queries from the totals, so the branded and non-branded figures will not sum to the site total.¹³

Say the garden furniture retailer trades as Oakbank Garden Co. People search for it as "oakbank", "oak bank", "oakbank garden", and misspell it as "okbank" and "oakbnak". Its own product line, the Hartwell bench, is searched by name too. In Search Console, click + Add filter, choose Queries, choose Custom (regex), and enter:

Leave it on Matches regex for branded queries and click Apply. For non-branded, edit the same filter and switch it to Doesn't match regex.³ Because matching is partial and not case-sensitive by default, "Oakbank Garden Co reviews" and "OAKBANK delivery" both match without extra rules. Check the result by scanning the first few hundred branded queries for anything that is not about you (a place called Oakbank, for instance) and the first few hundred non-branded ones for misspellings you missed, and refine the expression until both lists are clean.

A fall in branded clicks usually means fewer people are looking for you by name, which is a marketing and reputation question more than an SEO one, unless the brand queries are now showing someone else or your home page has gone missing from the results. A fall in non-branded clicks with branded holding steady points at rankings, indexing or the results page.

Segment by country, device and search appearance

Check countries if the site serves more than one market. A drop in one country only often points at hreflang, geotargeting or a local competitor (auditing hreflang covers the first). Check devices: a drop on mobile only can point at a mobile rendering fault, an interstitial or a mobile template change. Check search appearance: if a rich result type (review snippets, product results, videos) disappears from the list, the structured data on those pages may have broken.

Read the four metrics together

For each segment that fell, look at clicks, impressions, average position and click-through rate together. The combination tells you more than any one of them. Average position in Search Console is the average of the topmost position your site held for each impression, so it is an average across very different queries. Read it as a direction of travel.

There is a difference between a small drop in position, such as moving from position 2 to 4 for a query, and a large drop, such as falling from the top 10 to position 29 across a wide range of terms.¹ Small drops need no drastic action, and large drops are the ones that justify a deeper self-assessment of the content.¹⁷ Note which kind you are looking at for each segment.

AI Overviews and AI Mode in the data

Traffic from AI Overviews and AI Mode is included in the Web search type in the Performance report.²³ Search Console also has a separate Generative AI performance report for Search, which shows impressions only, not clicks, for AI Overviews and AI Mode, broken down by page, country, device and date. It rolled out to all websites worldwide by 31 August 2026, though it only appears for sites with enough impressions in those features.²⁴ If your clicks fell while impressions held, this report shows whether your pages are appearing in AI features at all, which helps you judge whether the AI feature is citing you or replacing the click. For more on measuring this side, see how to measure visibility in AI search.

Independent studies suggest the effect on clicks can be large for informational queries, which helps you judge whether a fall of this kind is unusual. Pew Research Center tracked the browsing of 900 US adults in March 2025: on Google searches that showed an AI summary, people clicked a traditional result link in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary in 1% of visits.²⁵ Seer Interactive measured 3,119 informational queries across 42 of its client organisations from June 2024 to September 2025. Organic CTR on queries with an AI Overview fell from 1.76% to 0.61%, but it also fell on queries without one, from 2.74% to 1.62%, so not all of the decline can be put down to the Overview. Seer also found brands cited in the Overview had a higher organic CTR than those not cited, and says plainly it cannot prove the citation causes it.²⁶ Ahrefs, comparing aggregated Search Console data for 300,000 informational keywords in December 2023 and December 2025, found an AI Overview correlated with a 58% lower average CTR for the top-ranking page.²⁷ The samples and methods differ and none of them tells you what happened to your site, but together they make a CTR fall on informational queries with an AI Overview look common, and not by itself a sign of a fault.

The safest way to see a change in the results page is still to look at it. For the 10 queries that lost the most clicks, search them in a private window from the right country and record what appears above your listing now. Screenshots dated in the drop log are evidence you will want later.

Technical faults are the causes you most want to find, because they have a clear fix and traffic usually returns once Google recrawls the corrected pages. They range from site-wide problems such as the site being down to page-level problems such as a misplaced noindex tag.¹ Work through the checks below for the sections that fell, starting with the reports that cover the whole site.

Manual actions and security issues

Check these two reports first, because they take a minute and change everything if they show a problem. The Manual actions report shows whether a human reviewer at Google has found that pages break Google's spam policies, and most manual actions result in pages or sites ranking lower or being left out of results.²⁸ The Security issues report shows whether Google believes the site has been hacked or behaves in a way that could harm visitors, such as malware, phishing or unwanted software, and affected pages can carry a warning in search results or an interstitial warning in the browser.²⁹ An empty report in both is good news. Note it in the log and move on.

Page indexing report

The Page indexing report shows which pages Google can find and index and why others are not indexed. The reasons it lists include excluded by noindex tag, blocked by robots.txt, server error (5xx), redirect error, not found (404), soft 404, duplicate without user-selected canonical, Google chose a different canonical than the user, crawled but currently not indexed, and discovered but currently not indexed.³⁰ Not every page needs to be indexed.³⁰ What you are looking for is a change: a reason whose count jumped around the drop date, and whether the affected URLs belong to the section that lost traffic. Filter the report to a sitemap that contains only that section, if you have one, to make this easier.

URL Inspection on sample pages

Pick five to ten URLs from the affected section that lost the most clicks and inspect each one. The URL Inspection tool shows whether the page is indexed, the canonical you declared, the canonical Google selected, whether crawling and indexing were allowed, and when the page was last crawled. The live test checks the current version and shows the rendered page.³¹ The tool does not check for manual actions, security issues or quality, so a clean result means only that it found no technical barrier.³¹

Crawl Stats

The Crawl Stats report shows Google's crawling over the last 90 days: total requests, download size, average response time, and host status for robots.txt fetching, DNS resolution and server connectivity, with breakdowns by response code, file type, purpose and Googlebot type.³² Look for a fall in crawl requests, a rise in average response time, a rise in 5xx or 404 responses, or a host status warning around the drop date. The report is aimed at advanced users and is only available for root-level properties.³²

robots.txt

Fetch your robots.txt directly and read it. A single Disallow line added to a staging environment and then deployed to production can block a whole section. Also check how it responds, not just what it says. The standard, RFC 9309, lets a crawler treat a robots.txt that returns a 4xx error as if no file existed, and requires it to assume complete disallow when a server or network error makes the file unreachable.³³ Google follows that closely: it treats a 4xx (other than 429) as no file, and on a 5xx it stops crawling the site for the first 12 hours while retrying, then uses the last good version for up to 30 days.³⁴ A robots.txt that intermittently errors behind a CDN or firewall can quietly slow crawling. Robots.txt for AI crawlers covers the newer user agents if you have recently edited the file to deal with them.

noindex

Check the affected templates for a robots meta tag and for an X-Robots-Tag HTTP header. Both work, and the header is easy to miss because it never appears in the page source.³⁵ It is a de facto standard across crawlers and the only way to set indexing rules on non-HTML files such as PDFs.³⁶ Check the raw HTML the server sends, not only the page in a browser. On JavaScript sites, Google may skip rendering and JavaScript execution when it finds a noindex tag, so a noindex in the initial HTML that JavaScript later removes can still keep the page out.³⁷ Also remember the reverse: a page blocked by robots.txt cannot have its noindex seen at all.³⁵

Canonicals

Check the canonical tag on each affected template. A template change that points every product to the category page, or every page to the home page, is a common and damaging fault. For Google, redirects and rel=canonical are strong canonicalisation signals and sitemap inclusion a weak one, and conflicting signals are best avoided.³⁸ If URL Inspection shows Google choosing a different canonical from the one you declared on many pages, find out why: duplicates, conflicting signals, or a canonical pointing at a page that redirects or returns an error.

Redirects and status codes

Crawl the affected section with a crawler of your choice and check the status code of every URL that lost clicks. In HTTP terms, a 301 says the resource has a new permanent address and a 302 that it is temporarily somewhere else.³⁹ Google acts on those codes: URLs that return a 4xx code (other than 429) are not indexed, and URLs already in the index that start returning one are removed. Server errors (5xx) and 429 make Google slow its crawling, and URLs that persistently return a server error are removed from the index. A 301 is treated as a strong signal that the redirect target should be canonical, a 302 as a weak one.⁴⁰ Look for redirect chains, redirects to the home page, redirects to pages that return 404, and loops. Building a redirect map covers how they should be set up.

Rendering

Google processes JavaScript pages in three phases: crawling, rendering and indexing.³⁷ If the affected template relies on JavaScript for its main content, links or metadata, use the live test in URL Inspection and look at the rendered HTML and screenshot. Is the content there? Are the internal links there? Did a new script, consent wall or bot protection layer stop the content loading? Checking what crawlers see on a JavaScript site goes through this in detail.

Crawl the affected section yourself

Search Console shows what Google saw, with a delay. A desktop crawler shows what the site serves now, URL by URL, and can be pointed at exactly the pages that lost traffic. Two widely used examples are Screaming Frog SEO Spider and Sitebulb. Neither is required, and other crawlers do the same job; these two are described because their documentation covers the settings this stage needs.

Screaming Frog's free version crawls up to 500 URLs; a paid licence removes that limit and adds JavaScript rendering, the Google Analytics and Search Console API integrations, crawl comparison and custom extraction. Check the current price on its site.⁴¹ Sitebulb offers two crawlers: the HTML Crawler, its default and quickest option, and the Chrome Crawler, which renders pages with headless Chrome and is the one to choose if the site depends on JavaScript.⁴²

Settings that matter for a drop investigation in Screaming Frog:

• Use List mode (Mode, then List) and paste in the URLs that lost the most clicks from your Search Console export, so the crawl checks exactly those pages.
• Under Configuration, then Spider, then Rendering, choose JavaScript if the affected template relies on it, so the crawl sees the rendered page.⁴³
• In the user-agent configuration, pick the Googlebot preset so you see what the server sends to a client that identifies as Googlebot.⁴³ Firewalls that verify Googlebot by IP may still treat your crawler differently from the real thing, so a clean crawl does not prove Google was served the same page.
• Under Configuration, then Robots.txt, choose Ignore robots.txt but report status, so blocked URLs are still crawled and flagged as blocked.⁴³
• Use Configuration, then Custom, then Extraction (XPath, CSS Path or Regex) to pull a value from every page, such as a product's price or the template version printed in a comment, which helps you tie pages to a release.⁴³

In Sitebulb, choose the Chrome Crawler at audit setup for JavaScript sites. It then produces a Response vs Render report showing whether meta robots, canonical, title, meta description and links are unchanged, created, modified or deleted by JavaScript.⁴⁴ A meta robots tag that JavaScript modifies, or a canonical it creates, is exactly the kind of fault that a template release introduces.

Compare a crawl from before the drop with one from now

If anyone crawled the site before the drop (an agency, a scheduled audit, a previous investigation), comparing that crawl with a fresh one is the fastest way to find what changed. In Screaming Frog this needs a licence and Database Storage mode (File, then Settings, then Storage Mode). Switch to compare mode with Mode, then Compare, and pick the two crawls. URLs are sorted into Added, New, Removed and Missing for each filter, and Config, then Compare, lets you switch on change detection for elements such as page titles, meta descriptions, word count, crawl depth, internal links, structured data and content.⁴⁵ In Sitebulb, tick Compare next to two audits in a project and click Compare Audits; the comparison shows the change in each report and hint as a percentage and can be exported to CSV.⁴⁶

If you have no earlier crawl, compare what you can: a staging crawl against live, or the crawl export against the URLs in the sitemap and the URLs in your Search Console export. Pages that earned clicks before the drop and now return anything other than a 200, or now carry a noindex or a canonical to another URL, go straight into the drop log.

You now have the start date, the shape, the segments that moved, the pattern of metrics, the events that line up and the technical results. Most drops can be explained from this evidence alone. Work through each segment that fell and give it a cause, a confidence level and the evidence behind it. Some segments will have more than one cause; some causes will be ruled out completely.

Technical fault

Signs: a cliff, often starting within days of a release or infrastructure change, concentrated in one template or section, with impressions falling and the Page indexing or Crawl Stats reports showing a matching change. You can point to the exact fault on a URL. This is the most common cause of sudden, concentrated drops and the most fixable.

Migration that went wrong

Any significant change to a site can bring ranking fluctuations while Google recrawls and reindexes it, and that for a medium-sized site it can take a few weeks or more for new URLs to replace old ones in results, longer for large sites.⁴⁷ A dip spread evenly across the site that recovers over those weeks is normal. A drop concentrated on old URLs that do not redirect, on one template, or on new URLs that are not being indexed is a fault. Common migration mistakes include leaving noindex or robots.txt blocks from development in place and redirects that point to the wrong place.⁴⁷ What to check in the first month after a site migration covers the checks in order.

Manual action or security issue

Signs: a message in Search Console and an entry in the Manual actions or Security issues report. The drop can be site-wide or partial, depending on what the action or issue covers. If neither report shows anything, this cause is ruled out.

Spam update

A spam update is a notable improvement to how Google's automated systems detect spam, and sites that break its spam policies may rank lower or not appear at all.⁴⁸ Signs: a drop that lines up with a spam update on the status dashboard, no manual action, and a site with practices covered by the spam policies, such as large amounts of scaled or thin content, bought links, doorway pages or third-party content published to borrow the site's reputation.⁴⁹ For link spam updates, Google's position is that when its systems discount spammy links, any ranking benefit those links gave is lost and cannot be regained.⁴⁸

Core update reassessment

Core updates are significant, broad changes to Google's search algorithms and systems, made several times a year.¹⁷ Signs: a drop that lines up with a confirmed core update rollout, builds over the rollout period, affects many queries (often across many sections), shows a falling average position with impressions falling alongside it, and survives every technical check. The core updates documentation says a drop does not necessarily mean anything is wrong with the site.¹⁷ It often means Google now judges other pages to be more useful for those queries. Practitioner data shows how slowly this can move in both directions. Lily Ray's analysis at Amsive of the June 2025 core update, using Sistrix visibility data for US mobile results, found that many sites hit by earlier updates, including the September 2023 helpful content update, saw substantial recoveries, while noting that for some the recovery was small and came after two years of suppressed visibility.⁵⁰

AI features and other results page changes

Signs: impressions and position steady, clicks and CTR falling, often concentrated on informational queries, with an AI Overview or another feature now sitting above your result. Your page has not lost its place, but fewer people need to click it. This is a change in how the results page works, and there may be nothing wrong with the page at all. Pages need to be indexed and eligible to be shown with a snippet to appear as supporting links in AI features, and, by Google's account, no special optimisation is required beyond normal SEO practice.²³

Competitor or market change

Signs: a slope more than a cliff, position falling on specific queries, and the same competitor pages appearing above you each time you check. Sometimes it is a new entrant, sometimes a competitor that rebuilt its pages, sometimes a marketplace or publisher that started covering your topic. Compare the dated screenshots from stage three with a fresh set: whose pages moved up, and what do they offer that yours do not?

Lost links or lost demand for the brand

If a handful of strong links pointed to a section that fell, and those links were removed or the pages they sat on disappeared, the section can lose ground gradually. Any link tool will show lost links by date, although none sees the whole web. Separately, if branded queries fell while non-branded held, the most likely explanation is that fewer people are searching for you by name. Look at what changed in advertising, PR, offline marketing and reviews.

Use the tool below to work through these causes one question at a time. Answer from the evidence in your drop log, not from what you suspect. If you get to an outcome that does not fit the evidence, go back and check the answer that led there.

Each cause has a different recovery. Some have a defined fix and a route back. Some have no fix in the usual sense, only work that may help over months. Order the work by what is most certain to return traffic: technical faults, migration faults, manual actions and security issues first, then results page changes, competitors and core updates.

Technical faults

Fix the fault at its source, usually a template, a server rule or a configuration, not page by page. Then confirm the fix on live URLs with URL Inspection, use Validate fix in the Page indexing report for the relevant issue, and resubmit the sitemap for the affected section. Indexing after a request typically takes a day or so but can take much longer in some cases.³¹ For a large section, recovery depends on how quickly Google recrawls it, which you can watch in Crawl Stats. Finally, add a check to the release process that would have caught the fault: a test that fails if a production template contains noindex, for example.

Migration faults

Fix the redirects first. Every old URL that had traffic or links should return one 301 to the most relevant new URL. Remove staging leftovers such as noindex and robots.txt blocks, update sitemaps to list new URLs only, and update internal links to point straight at the new URLs. Keep redirects in place for as long as possible, generally at least a year.⁴⁷ The website migrations guide covers the full process.

Manual actions and reconsideration requests

A manual action has the clearest route back of any cause. The process is to read the action, fix every instance of the problem across the site (not only the example URLs), make sure Google can reach the fixed pages without a login, robots.txt block or noindex, and then request a review. A good reconsideration request explains the exact quality issue on the site, describes the steps taken to fix it, and documents the outcome of those efforts.²⁸ Reviews can take several days or weeks, and longer for some link-related requests, and you should not resubmit while a request is being reviewed.²⁸

• Write the request as a plain account: what was wrong, how it happened, what you changed, and how you checked.
• For unnatural links, document the links you had removed, the sites you contacted and the dates, and any links you disavowed and why.
• For thin or scaled content, say how many pages were removed, merged or rewritten, and how the site will avoid the problem from now on.
• Do not submit until the fix is complete. A partial fix is likely to be rejected.

Keep the evidence in a shared folder as you work: a spreadsheet of every URL changed and what was done to it, before and after screenshots of representative pages, and for link actions, the outreach log with dates. Link to it from the request if your organisation is comfortable sharing it, and summarise it in the request either way. The request is read by a person; a short, specific, honest account is easier to approve than a long one.

Security issues

Clean the site, close the vulnerability that let the attacker in (an outdated plugin, a weak password, an exposed admin panel), and check that no hacked pages, injected links or redirects remain. Fixing only some pages will not earn a partial return to search results. Once everything is fixed, request a review from the Security issues report; most reviews take several days or weeks.²⁹

Spam updates

There is no review request for an algorithmic spam demotion. Changes may help a site improve if Google's automated systems learn over a period of months that the site complies with the spam policies.⁴⁸ Go through the spam policies and remove or fix everything on the site that falls under one: scaled content made mainly to rank, doorway pages, hidden text, link schemes, third-party content published to exploit the site's reputation.⁴⁹ If the update targeted links, any value those links passed is gone and will not return.⁴⁸

Disavowing links: when it applies

The disavow tool is often reached for first and needed least. It is meant only for sites with a considerable number of spammy, artificial or low-quality links pointing to the site and those links have caused, or are likely to cause, a manual action. It is an advanced feature, and incorrect use can harm the site's performance in Google Search.⁵¹ A drop after a core update, or the ordinary spammy links every site collects, does not meet that test.

When it does apply, the file is a plain .txt in UTF-8 or 7-bit ASCII, with one URL or one domain per line, domain:example.com to disavow a whole domain, and lines starting with # as comments. It can be up to 2MB and 100,000 lines. The tool does not support Domain properties, so you upload it to a URL-prefix property, and it can take a few weeks for the list to be taken into account.⁵¹ Try to get links removed at the source first, and keep a record of who you contacted and when, which also feeds a reconsideration request.

Core updates: what to expect from recovery

This is the cause people most want a quick answer to, and there is none. Google's core updates documentation says that changes can take effect in a few days, but "it could take several months" for its systems to learn and confirm that a site has improved, and that "There's no guarantee that changes you make to your website will result in noticeable impact in search results."¹⁷ For small drops there is no need for drastic action, and changes should make sense for users and be sustainable over the long term.¹⁷ Be honest with stakeholders about this. A site can do everything right and still not recover its old position, because other pages may now be better answers. Where recovery does come, it can arrive with a later core update: the Amsive analysis above found sites hit in 2023 recovering in June 2025.⁵⁰

Google points sites with large drops to its guidance on creating helpful, reliable, people-first content, which opens by saying its automated ranking systems are designed to prioritise "helpful, reliable information that's created to benefit people" and not content created to manipulate rankings.⁵² The guidance includes self-assessment questions. Use them on the pages that lost the most, ideally with someone who did not write those pages. Among them: whether the content provides original information, reporting, research or analysis, and whether it is the sort of page you would want to bookmark, share with a friend or recommend.⁵² It also asks whether there is an existing or intended audience that would find the content useful if they came to you directly, and warns against content made mainly to attract visits from search engines or produced on many topics with extensive automation.⁵²

The same guidance asks site owners to think about who created the content, how it was created and why. It should be self-evident who wrote the content, any use of automation should be explained, and the why should be creating content primarily to help people.⁵² In practice, for a page that lost ground, ask:

• Does the page answer the query better than the pages now above it, judged by someone who needs the answer?
• Does it show first-hand experience: real products tested, real work done, real photographs, real prices?
• Is it clear who wrote it, and why they are qualified to?
• Was it written for a reader, or produced in bulk to cover keywords?
• Would the site lose anything useful if this page were removed or merged with another?

Then decide, page by page or section by section, whether to improve, merge or leave alone. In my view, the most common mistake after a core update is acting on the whole site at once: deleting large numbers of pages, rewriting everything, or changing templates, all in the same month. That makes it impossible to tell what helped, and some of those changes can do harm. Make deliberate improvements to the pages that matter, record them in the drop log with dates, and watch.

Run the page review as a worksheet

A review is easier to do honestly, and to explain later, if it is written down. Take the 20 to 50 URLs that lost the most clicks in the affected section and make one row per URL with these columns: URL, main query, clicks before and after, the page now ranking where you used to, what that page offers that yours does not, your answers to the helpful content questions above, the decision (improve, merge, remove, leave), the reason, the owner and the date done. Do the comparison with the competing page open in another tab. If the honest answer is that their page is better for the searcher, the worksheet says what to change; if yours is already the better page, it says to leave it alone and wait.

If you use an AI coding agent on the site, the content audit skill in the same skill pack inventories pages and recommends one action per page (keep, improve, merge, refresh, prune or create). Treat its recommendations as a first draft of the worksheet, and make the final call on each page yourself.

Recovery tactics that make things worse

Several tactics are passed around after every core update. Some are harmless but useless; some are warning signs in Google's own guidance; some break its spam policies. Google's helpful content guidance lists, among signs of search engine-first content, changing the date of pages to make them seem fresh when the content has not substantially changed, and adding or removing a lot of content primarily because you believe it will make the site seem fresh to search engines.⁵² Its spam policies define link spam as creating links primarily to manipulate rankings, including buying or selling links, and scaled content abuse as generating many pages primarily to manipulate rankings and not to help users.⁴⁹ Paid links are not a violation when they carry rel="sponsored" or rel="nofollow".⁴⁹

Results page changes and AI Overviews

If an AI Overview or another feature now answers the query before anyone reaches your listing, rewriting the page will not bring back the old click-through rate. What you can do is check that the page remains eligible to be cited (pages must be indexed and eligible to show with a snippet²³), see from the Generative AI performance report whether it is being shown in those features,²⁴ and then decide where to put effort. Queries where people still need to visit (to compare, to buy, to book, to use a tool, to get detail that a summary cannot hold) are worth more of your time than queries a two-line answer satisfies. Report this to stakeholders as a change in the results page, with the screenshots from stage three, so it is not mistaken for a ranking failure.

Competitor gains, lost links and brand demand

For competitor gains, compare your page and theirs on what the searcher needs, and improve where yours genuinely falls short. Sometimes the honest conclusion is that a larger or more specialist site now owns the query and your effort is better spent elsewhere. For lost links, find out why they went (a partner site closed, a page was moved without a redirect) and recover the ones you can. For lost brand demand, take the evidence to whoever owns marketing; search reflects interest in the brand and cannot create it.

Recovery has to be measured. Set up the monitoring before the fixes go live, so you have a clean before and after for each one.

What to watch, and how often

Annotate every fix and every Google update on your reporting charts. When a section starts to recover, you want to know whether it was your fix, a later update or the season. Without annotations, people will claim credit or assign blame on instinct.

Set up early warnings

Most of the damage from technical faults is done in the days before anyone notices. A few cheap checks shorten that:

• Turn on Search Console email notifications for every owner and make sure at least one person reads them.
• Run a scheduled crawl of a sample of key URLs from each template after every release, checking status code, robots meta, X-Robots-Tag and canonical.
• Monitor robots.txt for changes and alert on any change.
• Set an alert when weekly Search Console clicks for a key section fall more than an agreed share against the same week last year.
• Keep the drop log going as a change log: every release, every Google update, every significant content change.

Two of these are easy to set up with tools already covered. A paid Screaming Frog licence includes crawl scheduling,⁴¹ so the list of key URLs per template can be crawled automatically after each release window and the export compared with the last one. And a Data Studio report built on the Search Console connector, with one chart per template using URL Impression data,¹⁹ gives stakeholders a page to check without asking you for an export.

The stakeholder write-up

Write one document that a director can read in five minutes and an engineer can act on. It should not be a list of charts. Use this structure:

Be precise about what you know and what you suspect. A sentence such as "the product drop was caused by a noindex in release 4.12, confirmed in Search Console and fixed on 5 June" is worth more than three paragraphs of hedged possibilities. Where the cause is a core update or a results page change, say so directly and set expectations accordingly; the worst outcome is a stakeholder expecting traffic back next month because nobody told them otherwise.

For the running example, the opening of the write-up might read: "Organic search clicks fell by about 15% in the week of 2 June, measured in Search Console. Recorded sessions in GA4 fell further because a new consent banner went live the same week; that extra fall is a change in measurement and does not reflect lost visitors. Two causes account for the real fall. A noindex shipped with the new product template in release 4.12 removed product pages from Google; it was fixed on 5 June and pages are returning. Separately, AI Overviews now answer most of the questions our guides used to get clicks for; rankings there are unchanged, and we do not expect those clicks to return to their old level." Every figure in it traces back to a row in the drop log.