Short answer: date the drop precisely, then rule out causes in order — a technical eligibility failure, a Google update, a headline or content-pattern change on your side, or simple topic decay. Most publishers guess at the fourth and never check the first. Recovery from a genuine core update is not a switch: Google says changes can take days but may take several months, and you may have to wait for the next update before anything moves.
A Discover collapse feels different from a search decline because it happens overnight and there is no query report to interrogate. That absence pushes people straight to speculation. Resist it — the diagnosis is a sequence, and running it in order saves weeks.
Step one: date the drop precisely
Before anything else, establish the exact day traffic changed, and whether it fell off a cliff or eroded.
A single-day cliff points to an update, a technical break, or a manual action. A gradual slope over weeks points to topic decay, competitive displacement, or a content-quality trend.
Get the date to the day, not the week. Then set it against the record of known Google updates. A drop that begins on the exact day an update starts rolling out is a very different investigation from one that began eleven days earlier and nobody noticed.
Also check whether the drop is site-wide or confined to a section, a topic, or a single author. Segmented drops usually have segmented causes, and that alone eliminates most hypotheses.
The February 2026 Discover core update
Google shipped its first core update explicitly labelled as a Discover core update, rolling out from 5 February 2026 and completing on 27 February — roughly 22 days, longer than the two weeks Google initially estimated (Google Search Central; completion reported by Search Engine Land).
What it targeted matters for diagnosis. The update aimed to surface more locally relevant content from sites based in the user's own country, to reduce sensational content and clickbait, and to favour in-depth, original, timely reporting from sites with demonstrated topic expertise.
One scope detail that is easy to miss and changes the analysis: the update initially affected only English-language users in the United States, with expansion to other countries and languages to follow. If your audience is primarily outside the US and your traffic moved in that window, the update is a weaker explanation than it looks and you should keep investigating rather than settle on it.
The four causes, in order of what to check
| Cause | Signature | How to confirm |
|---|---|---|
| Technical eligibility | Sharp, often total, may affect a section | Check indexing of recent URLs, image markup, preview restrictions |
| Google update | Sharp, dated to a rollout window, partial | Match the date to the update record; check whether peers moved too |
| Content or headline pattern | Gradual, correlates with an editorial change | Compare headline style before and after; check the sensationalism guidance |
| Topic decay | Gradual, tracks a news cycle ending | Compare against interest in the underlying subject |
Work top to bottom. The first is cheapest to check and the most often skipped.
Running the diagnosis
Rule out eligibility before blaming the algorithm
More Discover collapses are technical than editorial, and the technical ones are the only kind you can fix quickly.
Confirm recent articles are actually indexed, and how fast. Confirm your image markup resolves to a genuinely large image on a live article — the tag, not the hero you can see. Confirm nothing restricts image previews site-wide. Confirm no template change altered your markup around the date of the drop.
A deploy that shipped the week traffic fell is the single most common cause we find, and it is almost never the first thing anyone checks.
Separate your problem from your beat's
Discover traffic depends on interest in the underlying subject. If the story that carried three months of traffic concluded, your traffic will fall and nothing is wrong with your site.
The test: did comparable publishers on the same beat move at the same time? If everyone covering the topic declined together, that is the news cycle. If you fell and they did not, it is you. Without this comparison you cannot tell the difference, and people routinely rebuild a functioning site in response to a story simply ending.
Then look at headlines, honestly
If the drop coincides with a core update and eligibility is clean, headline and framing patterns are the likeliest remaining cause — particularly given what the February 2026 update explicitly targeted.
Pull thirty headlines from before the decline and thirty from after. Look for withheld information, exaggeration, and curiosity-gap framing. Google's Discover guidance warns directly against "clickbait and similar tactics to artificially inflate engagement by using misleading or exaggerated details in preview content", and against "sensationalism tactics that manipulate appeal by catering to morbid curiosity, titillation, or outrage".
This is uncomfortable because the pattern being penalised is usually the one that was previously rewarded, and somebody in the newsroom introduced it deliberately and successfully.
What Google says about recovery
Google's guidance on core updates is worth quoting because it sets realistic expectations, and because it does not promise what most recovery advice implies.
On what to do: "Consider how you can improve your content in meaningful ways. For example, it could be that rewriting or restructuring your content makes it easier for your audience to read and navigate the page."
On what not to do: "Avoid doing 'quick fix' changes (like removing some page element because you heard it was bad for SEO)."
On timing: "Some changes can take effect in a few days, but it could take several months for our systems to learn and confirm that the site as a whole is now producing helpful, reliable, people-first content in the long term." And critically: "If it's been a few months and you still haven't seen any effect, that could mean waiting until the next core update."
Two honest caveats. That documentation addresses Search core updates and does not specifically cover Discover core updates, so applying it here is reasonable inference rather than stated policy. And "wait for the next update" is a genuinely hard thing to tell a publisher whose revenue is per pageview — but it is more useful than a promise of a six-week turnaround that nobody can keep.
Running the fix without making it worse
Once you have a cause you believe in, the sequencing matters as much as the fix.
Change one class of thing at a time, and log the date. If you correct image markup in week one and revise headline standards in week three, you can attribute whatever moves. If you do both on a Tuesday you have learned nothing and cannot undo half of it.
Fix technical faults immediately and everything else slowly. Eligibility problems have no downside to correcting and often produce movement within days. Editorial changes need weeks to accumulate enough new articles to register at all, so shipping them in a panic buys nothing.
Keep publishing through it. Discover is weighted toward recent content, so a newsroom that slows down while it investigates removes the very material that would demonstrate improvement. This is counter-intuitive under revenue pressure and it is the mistake most likely to deepen a decline.
Set a review date and hold it. Pick a point four to six weeks out, agree in advance what evidence would count as improvement, and do not re-diagnose weekly in between. Discover's day-to-day variance is wide enough that constant re-reading of the graph will generate a new theory every week and a new intervention with it.
Write down what you changed. The single most valuable asset in the next incident is an accurate record of what you did in this one, and it is almost never kept.
What not to do
Do not make ten changes at once. You will never know which one mattered, and if traffic falls further you cannot roll back cleanly.
Do not delete archives reflexively. Pruning is sometimes right and is frequently done in panic, on volume rather than quality, and destroys pages that were performing.
Do not chase an unrelated traffic spike to fill the hole. Publishing off your beat muddies the topic signal that the February 2026 update specifically rewards.
Do not treat one bad week as a trend. Discover is volatile by design, and a single poor week sits comfortably inside normal variance for most publishers. Judge fortnights, and preferably compare them against the same period a year earlier.
For what the feed rewards when you are not in recovery, see Google Discover for news publishers. For the parallel erosion happening in Search, see what AI Overviews did to publisher traffic.
We do this diagnosis with news websites and digital publishers, usually alongside search and answer-engine work and the growth programs it supports. If your Discover traffic moved and nobody can say why, get in touch.
