Google's 2026 Core Updates: What Changed and How to Recover

If your organic traffic fell off a cliff this spring, you were not imagining it and you were not singled out. Google shipped two core updates inside five months, and both produced ranking movement large enough that agencies across the UK spent April and June fielding the same phone call. The important thing is that core update damage is diagnosable and, in most cases, reversible. What kills recovery is guessing.

Why this matters now

A core update is not a penalty. Nobody at Google reviewed your website and decided to punish it. A core update is a change to how Google weighs the signals it already collects, which means pages can lose visibility without doing anything wrong in the ordinary sense. They simply stopped comparing favourably against the pages Google now prefers.

That distinction matters because it changes the fix. A penalty is something you remove. A core update loss is something you out-compete. The sites that recovered fastest in 2026 did not find a single broken thing and repair it. They raised the quality floor across their whole domain until they compared well again.

The pattern in both 2026 updates was consistent. Original, first-hand content with named authors and demonstrable expertise gained ground. Thin, templated and mass-produced pages lost it. Sites that had leaned on unedited AI output over the previous year saw the clearest declines, which is the single most useful signal in the whole cycle because it tells you exactly where to look first.

The scale is worth being honest about. A professional services firm across the East Midlands lost 22% of its organic traffic almost overnight after the March update. That is not a rounding error, and for a business generating leads through search it is a genuine revenue event. The same firm was above its pre-update traffic within ten weeks, having audited page speed, tightened thin content and rebuilt its weakest pages. Recovery took under a quarter and required no new spend on links or advertising.

There is a second-order effect too. Because core updates redistribute rather than destroy visibility, somebody in your market gained what you lost. If you do nothing, that position hardens. Competitors who improve during the recovery window tend to hold the ground they took when the next update arrives.

Common mistakes to avoid

The mistakes that follow a core update are rarely lazy ones. They are the well-intentioned reactions of people trying hard to fix something quickly, which is precisely what makes them dangerous.

• Rewriting everything at once. If you change 200 pages in a fortnight you destroy your ability to learn anything. When traffic moves you will not know which change caused it. Work in batches of ten to twenty pages and leave a measurement gap between them.
• Assuming it was technical. Core updates are content and trust evaluations. Your server did not cause this. Technical work is worth doing, but if you spend six weeks on render-blocking scripts while your thin pages sit untouched, you will not recover.
• Deleting pages in a panic. Removing a page destroys any link equity pointing at it and any residual long-tail traffic. Merging two weak pages into one strong one is almost always better than deleting both.
• Adding more content on top of weak content. Publishing twenty new articles while the existing hundred remain thin makes the average worse, not better. Quality is assessed across the domain, so volume without a floor is counterproductive.
• Waiting for the next update to reverse it. Recoveries do sometimes land on subsequent updates, but only for sites that changed in the interim. Waiting is not a strategy, it is just a delay with the same work at the end of it.

There is a specific version of this that caught a lot of businesses in 2026. Many had spent 2025 scaling content production with AI, publishing forty or fifty articles where they previously published four. The output read plausibly and cost almost nothing, so the incentive was obvious. What the updates exposed is that plausible is now the baseline, not the bar. Content that reads like it could have been written by anyone about any business is exactly what the 2026 updates demoted.

If that describes your last year of publishing, the fix is not to delete it all. It is to go through it with a red pen and ask of each page: does this contain anything a competitor could not have written? If the answer is no, either add first-hand specifics or fold the page into something stronger.

Step-by-step recovery plan

This runs over roughly 30 days and assumes one person with a few hours a week, not an agency retainer. Work in order. The sequence matters because each step narrows what the next one has to consider.

1. Confirm the cause before touching anything. Open Search Console, set the date range to cover March through June, and compare week on week. If your decline starts inside 27 March to 8 April or 21 May to 2 June, it is a core update. If it starts somewhere else, you are chasing the wrong problem and everything below is wasted effort.
2. Establish what you actually lost. Export your top 200 pages by clicks for the eight weeks before the drop and the eight weeks after. The pages with the largest absolute loss are your priority list. Ignore percentage drops on pages that were getting nine clicks a month.
3. Categorise the losers. Every declining page falls into one of three buckets: thin (under 300 words or no substantive information), duplicative (covers the same ground as another page), or dated (accurate once, not now). The bucket determines the fix, so do not skip this.
4. Fix the thin pages first. Expand with genuine first-hand detail: your actual process, real numbers, named examples, specific outcomes. If you cannot add anything real, the page should not exist. This is the step that does the most work in most recoveries.
5. Merge the duplicates. Pick the stronger URL, fold the useful content from the weaker one into it, and 301 redirect. Two pages competing for the same query is a self-inflicted wound that core updates punish harder than they used to.
6. Add verifiable authorship. Put a real named person on your substantive pages, with a role and relevant credentials. Not a generic company byline. The 2026 updates rewarded demonstrable expertise, and an anonymous page cannot demonstrate any.
7. Audit page speed on the pages you just fixed. This is where technical work belongs in the sequence, after the content is right. Slow pages compound a content problem rather than causing it. Our Core Web Vitals guide for non-developers covers what to check without needing a developer.
8. Strengthen internal linking into the fixed pages. A page Google cannot easily reach from your important pages reads as unimportant. Link to your rebuilt pages from your homepage and top service pages, using descriptive anchor text.
9. Leave it alone for three weeks. This is the step everyone skips and it is the one that determines whether you learn anything. Google needs to recrawl and reassess. Changing more during this window contaminates your read on what worked.
10. Review, then run the next batch. Compare the fixed pages against the untouched ones. Whatever moved, do more of. Whatever did not, stop doing. Then take the next twenty pages.

The discipline in that sequence is the batching and the waiting. Most failed recoveries are not failed because the work was wrong. They failed because everything changed at once, nothing could be attributed, and after two months of effort nobody could say what helped.

If your site is small, under thirty pages or so, you can compress this to a fortnight. The order stays the same. What you cannot compress is the reassessment window, because that is Google's clock, not yours.

Recovery checklist

Run this against every page you rebuild. If you cannot tick an item, the page is not ready to ship.

• The page contains at least one fact, number or example that a competitor could not have written.
• A named author with a real role and relevant credentials appears on the page.
• No other page on the site targets the same query.
• Every claim has a reason, a number or a named example behind it.
• Anything dated has been checked and updated, or removed.
• The page loads acceptably on a mid-range phone on mobile data, not just on your desktop.
• At least two internal links point into the page from relevant pages.
• The page answers the question in its title within the first two paragraphs.
• Nothing on the page exists purely to reach a word count.
• You could send the page to a client without embarrassment.

That last one is not a joke. It is the fastest usable proxy for the quality bar the 2026 updates were reaching for, and it takes two seconds to apply.

How to measure recovery

Recovery is slower to read than the decline was, because the drop happens in days and the return happens over weeks. You need the right instruments or you will conclude too early that nothing is working.

Watch impressions before clicks. Impressions move first, because Google starts showing your page for more queries before users start choosing it. If impressions on your fixed pages are rising while clicks are flat, the recovery has started and you should hold your nerve.

Compare fixed against unfixed. This is the single most valuable measurement you can take, and it costs nothing. Track your rebuilt batch against a comparable set you have not touched. If both rise, the market moved. If only the fixed set rises, your work is responsible.

Track queries, not just pages. A recovering page often returns on different queries to the ones it lost. Search Console's query view tells you whether Google has re-understood what the page is about, which is frequently what actually changed.

Give it ten weeks before judging. The East Midlands recovery referenced earlier took ten weeks to pass its previous level, and that was with focused work. Assessing at three weeks will tell you almost nothing and will tempt you into abandoning something that was working.

Set a single monthly review. What improved, what stalled, what did we learn. Feed the answers into the next batch. That loop is worth more than any tool you could buy.

Key terms in plain English

Core update: A broad change to how Google weighs ranking signals, rolled out over one to three weeks and affecting results across every industry at once.

Thin content: A page with little substantive information, either because it is very short or because it says nothing specific. Length alone is not the test.

Domain quality: Google's assessment of your site as a whole. Weak pages affect the evaluation of strong ones, which is why your worst pages matter more than they seem to.

Cannibalisation: Two or more of your own pages competing for the same query, splitting signals so neither ranks as well as one strong page would.

Impressions: How often your page appeared in results, whether or not anyone clicked. The leading indicator of recovery.

Reassessment window: The period after you change a page during which Google recrawls and re-evaluates it. Typically two to four weeks.

First-hand content: Information that comes from actually doing the thing, rather than from summarising what other pages say about it.

Conclusion and next move

Core updates feel arbitrary from the receiving end, but the 2026 pair were unusually legible. Google rewarded pages that could only have been written by someone who knows the subject, and demoted pages that could have been written by anyone about anything. That is a bar you can actually aim at.

Start this week by confirming the date of your decline and listing your twenty worst-performing pages. Do not write anything yet. The diagnosis takes an afternoon and it determines whether the following month is productive or wasted. Then fix in batches, wait, and measure the fixed against the untouched. If you are also rethinking how search visibility works now that AI answers sit above the results, our AI search visibility playbook covers the other half of the picture.