Why did my website traffic suddenly drop?

Before opening a single tool, look at the shape of the line. It narrows the diagnosis more than anything else you can do in the first five minutes.

Read the shape first

Also check one thing before anything else: is the drop in Search Console as well as analytics? If Search Console shows healthy clicks and analytics shows a collapse, your tracking broke. This accounts for a meaningful share of reported emergencies and takes thirty seconds to rule out.

The cliff: something broke

A near-vertical drop over one or two days is almost never an algorithm. Algorithms roll out gradually. Cliffs mean something changed on your side or your server’s.

Work these in order.

1. Check the Pages report in Search Console. A sharp fall in indexed pages is the diagnosis. Read the excluded reasons.
2. Check for an accidental noindex. In the HTML and in the X-Robots-Tag HTTP header, which is easy to miss. This is the most damaging and most common cliff cause.
3. Read robots.txt. A disallow shipped with a deploy will do this instantly.
4. Check server response codes across a sample of important URLs, and check crawl stats for 5xx spikes.
5. Check whether anything was deployed or migrated in the 72 hours before the drop. Ask, rather than assuming you would have been told.
6. Check the Manual Actions report. Rare, always disclosed, worth ruling out in ten seconds.

If a migration preceded it, four failures cause most of the damage: redirects not mapped one to one, staging directives carried over, canonicals pointing at old URLs, and internal links still pointing at redirected addresses.

The slope: update or decay

A decline spread over one to three weeks, affecting many pages across whole topic areas, is the signature of an algorithm update.

Confirm by matching the start date against confirmed core and spam update timelines, and by checking breadth: updates hit topics, not individual URLs. If only four pages fell, it is not an update however tempting that explanation is.

Algorithmic recovery is slow and conditional. Positions typically do not return until a subsequent update, and only if the underlying quality issue has genuinely been addressed. The productive response is to improve the affected content substantively rather than to make cosmetic changes and wait.

A slower decline over several months, page by page, is decay rather than an update — competitors published something better, or your information aged. That is easier to fix and easier to prevent, and our guide to page triage covers finding decayed pages before they cost you much.

The partial drop: one section

When the decline is confined to one directory, template or topic, the cause is almost always local to that thing.

Check what is shared across the affected pages and not across the healthy ones. A template change that introduced a canonical error, a section-level robots directive, a redesign that moved content behind JavaScript, a bulk edit, a category page that stopped linking to its children.

Segment Search Console by page path to confirm the boundary precisely. Knowing exactly which URLs are affected and which are not usually identifies the shared cause immediately, and it is far faster than auditing the whole site.

The drop that is not real

Four scenarios produce an alarming chart and no actual loss.

Broken tracking. A tag removed in a deploy, a consent banner change, a container misconfiguration. Search Console will show normal clicks. Check this first, always.

Seasonality. Compare year over year rather than month over month. A great many autumn crises are August returning to normal.

A reporting change. Analytics platform migrations, attribution model changes, filter adjustments and bot-filtering updates all produce step changes that look like traffic events.

Losing traffic you never wanted. If a page ranking for an irrelevant high-volume term stops ranking, total traffic falls and nothing of value is lost. Check whether conversions moved with the traffic; if they did not, the loss was noise.

The cause that is new

One pattern has become common enough to check explicitly: positions flat, impressions flat or up, clicks down.

That is an AI Overview absorbing the click. Nothing about your ranking changed; the page changed around you. Pew Research, tracking roughly 68,000 real queries, found users clicked a result in 8% of visits where an AI summary appeared against 15% where one did not. Ahrefs measured a 58% click-through decline for top-ranking pages in February 2026, up from 34.5% in April 2025.

To confirm: check AI Overview presence on your highest-impression queries, and check whether the CTR decline is concentrated at positions one to three, where displacement bites hardest. Informational queries falling while transactional ones hold is a strong signature.

The response is different from every other cause on this page. There is no fault to fix. The work is to be cited inside the answer rather than listed beneath it, and to start measuring a surface where no impression is ever recorded. Our AI Search 2026 report covers the data and the response in full.

The first hour

When a drop is reported, the temptation is to open everything at once. A fixed order gets to the cause faster and prevents the panic changes that make later diagnosis impossible.

1. Confirm it is real. Does Search Console agree with analytics? If not, it is tracking.
2. Establish the shape and the exact start date. Everything downstream depends on these two facts.
3. Check indexed page count. A fall here is the diagnosis and shortcuts everything else.
4. Check whether anything shipped in the preceding 72 hours. Ask people directly; deploy logs miss CMS and DNS changes.
5. Segment by page path and by query type. Whole site, one section, or one intent group — each points somewhere different.
6. Only then start forming a hypothesis.

Write down what you find at each step. Incidents get re-diagnosed from scratch by the next person far too often, and a fifteen-line record turns a two-day investigation into a ten-minute read.

Communicating during an incident

Say three things and no more: what the data shows, what has been ruled out, and when the next update comes. Resist offering a cause before you have one — an early guess becomes the accepted narrative and is remarkably hard to correct once someone senior has repeated it.

It is also worth stating the realistic recovery window as soon as the cause is known, because the gap between “fixed” and “recovered” is where most of the organisational anxiety lives. A technical fix that will take three weeks to show is a very different conversation when that is said upfront rather than explained in week two.

What not to do in the first week

Do not make sweeping changes. Rewriting content, changing site architecture or rolling back unrelated work while the cause is unknown adds variables to an already unclear picture and destroys attribution.

Do not disavow links. It is almost never the cause, it is easy to do damage with, and the tool exists for a narrow situation that does not describe most drops.

Do not assume an update. It is the most comfortable explanation because it is nobody’s fault, which is precisely why it gets reached for before the evidence supports it. Check breadth and timing first.

What recovery looks like

Set expectations by cause, because they differ enormously and mismatched expectations cause more panic than the original drop.

• Technical fault, fixed: days to three weeks, as pages are recrawled.
• Migration error, fixed: two to eight weeks, longer for a large site.
• Algorithm update: often not until a subsequent update, and only with genuine improvement. Months.
• Decay or competition: as fast as you improve the pages, typically four to twelve weeks.
• AI Overview absorption: the clicks do not come back. The response is to change what you are optimising for.

One discipline throughout: change one thing at a time while the cause is uncertain, and write down what you changed and when. Incident response is where teams make five simultaneous changes, recover, and learn nothing transferable — which guarantees the same fortnight repeats the next time it happens.

Sources

• Pew Research Center — approximately 68,000 real search queries; clicks fell from 15% to 8% of visits when an AI summary was present.
• Ahrefs — CTR impact of AI Overviews; 34.5% decline for position one (April 2025), approximately 58% (February 2026).
• Google Search Central — documentation on core updates, manual actions and the Pages index coverage report.

Frequently asked questions