Your Google traffic dropped, but the aggregate line does not tell you what broke. Search and Discover can move for different reasons, and treating them as one channel can send you toward the wrong fix.
Separate the surfaces first. Then inspect timing, geography, impressions, clicks, queries, and affected page groups. That sequence will tell you whether to investigate distribution, content-market fit, measurement, or a broader site problem.
Start by separating Search from Discover
Google Search begins with an expressed query. Discover recommends content around a user’s inferred interests. A page can therefore lose Discover distribution while retaining Search demand, rankings, and clicks. The reverse can also happen.
The distinction became especially important during Google’s February 2026 Discover core update. Its rollout ran from February 5 through February 27 and applied, at completion, only to Discover for U.S. users viewing English content. It was the first confirmed update announced specifically for Discover. Search fluctuations during the same period were not confirmed as part of that update.
| Surface | What starts the experience | What to inspect first | Common diagnostic mistake |
|---|---|---|---|
| Google Search | A query expressed by the user | Queries, landing pages, countries, devices, impressions, and clicks | Attributing a Search decline to a Discover-only update |
| Google Discover | A personalized recommendation based on interests | Discover pages, countries, devices, impressions, and clicks | Treating a feed-distribution change as a sitewide Search loss |
Use the February scope only when interpreting that rollout window. Google said it planned to expand the update to other countries and languages later, so the original U.S.-English boundary should not be assumed for subsequent periods without verification.
Diagnose the change before editing content
Do not start by rewriting pages. First establish exactly where visibility changed. Otherwise, a Discover decline can trigger unnecessary Search edits, while a measurement fault can be mistaken for an algorithmic loss.
- Verify the measurement. Compare your analytics platform with Search Console. If analytics traffic fell while Search Console impressions and clicks remained consistent, investigate consent, tagging, reporting, and attribution before changing content.
- Split Search and Discover. Review each performance surface independently. Record the start of the change rather than relying on the combined organic traffic line.
- Mark relevant rollout dates. If the movement began around February 5 through February 27, 2026, note that window. Timing creates a hypothesis; it does not prove a cause.
- Segment the exposed audience. Compare the United States with other countries. Because Search Console does not give you a simple content-language diagnosis, also isolate the page groups serving your English-language U.S. audience.
- Separate reach from response. Falling impressions indicate that the content was shown less often. If impressions are relatively stable but clicks fall, investigate placement, presentation, headline fit, and intent before concluding that visibility disappeared.
- Find the affected page cluster. Group pages by subject, format, geography, creator, and publishing pattern. A concentrated decline is more actionable than a sitewide average.
If Search is stable and Discover falls, keep the investigation inside Discover until the evidence points elsewhere. Review which topics and geographic audiences lost impressions. Do not change title tags or Search-focused copy merely because the combined organic total declined.
If Discover falls mainly for U.S.-facing English pages around the rollout window while other markets remain steadier, the update is a plausible contributor. It is still not proof. Check whether the loss is concentrated in sensational headlines, thin coverage, non-local material, or topics where your site has little sustained expertise.
If Search declines but Discover remains stable, investigate Search demand, query visibility, landing pages, indexing, and technical conditions. The February Discover update is not an adequate explanation for that pattern.
If both surfaces decline, widen the scope. Confirm tracking, crawling, indexing, templates, site changes, demand, and the affected directories. A simultaneous decline may be broad, but the shared timing alone does not identify the cause.
Use long Search queries to expose conversational demand
Traditional keyword lists often miss the way people now phrase complex tasks, comparisons, and concerns. Search Console gives you a useful first-party proxy: the longer queries for which your pages already received impressions or clicks.
You can filter for queries containing at least 10 whitespace-separated words with this process:
- Open Search Console and go to Performance > Search queries.
- Select Add filter > Query.
- Choose Custom regex.
- Enter
^(?:S+s+){9,}S+$. - Apply the filter and export the resulting queries with their available performance data.
The expression looks for at least 10 non-whitespace terms separated by whitespace. It is a practical threshold for finding prompt-like language, not a definition of an AI prompt.
That caveat matters. Search Console can contain data connected with AI Mode, and unusually conversational searches may resemble prompts used in an assistant. But a long query does not reveal where or how it originated. The user may have typed it directly into Google. Treat the data as evidence of conversational demand, not proof of ChatGPT, AI Mode, or another platform.
After export, cluster the queries by the behavior they reveal:
- User job: planning, comparing, troubleshooting, learning, checking, or choosing.
- Entity: your brand, a competitor, a product, a location, or a named problem.
- Decision context: constraints, desired outcome, use case, audience, or risk.
- Unresolved concern: reputation, an old incident, compatibility, trust, or a reason not to buy.
- Current destination: the page that received the impression and whether it actually resolves the full request.
A spreadsheet works for a small export. A language model can accelerate a larger clustering task, but preserve every original query so you can audit its grouping. A useful instruction is: Group these queries by user job, entity, decision context, and concern. Preserve each original query, name the likely content gap, and do not infer which platform generated the query.
Treat query exports as potentially sensitive. Conversational strings can contain personal information. Remove or mask identifiable details before uploading the file to an external analysis tool, and follow your organization’s data-handling rules.
The result should not be an enormous list of literal sentences to monitor. Build a smaller prompt-tracking set around recurring themes. Prioritize a theme when it repeats, has a meaningful commercial or reputational consequence, intersects with a page already receiving visibility, and can be answered with credible content.
For example, several differently worded queries may all ask whether your company is a safe alternative to a better-known competitor. Track representative comparison and risk-objection prompts, then create or improve the page that should answer them. The theme is durable even when the exact wording changes.
Build the topical signals Discover is trying to reward
The February 2026 update was designed to surface more locally relevant material, less sensational content, and more original, timely, in-depth work from sites with subject-specific expertise. Those are editorial directions, not a checklist that guarantees feed placement.
Build a recognizable topical footprint
Discover’s expertise assessment can operate topic by topic. A broad publisher can establish a strong specialist section, while a site with one unrelated page offers much weaker evidence of sustained knowledge. You do not need to turn the whole domain into a single-topic publication, but the section you want recognized must be coherent.
Audit that footprint directly:
- Name the subject for which you want the site or section to be recognized.
- Label existing URLs as core coverage, genuinely supporting coverage, or unrelated material.
- Connect related pages through clear navigation and internal links so the section is understandable as a body of work.
- Use long-query clusters to find missing questions that belong naturally inside the subject.
- Resist publishing a one-off page merely because a neighboring topic is popular.
The aim is not volume. It is continuity. Each new page should deepen the same audience’s understanding or help that audience complete the next related task.
Make originality, depth, and timeliness visible
Calling content original is not enough. The distinct contribution should be easy to identify. Before publishing, ask what the page adds that a competent reader could not get from a generic summary.
- Originality: include your own reasoning, evidence, process, examples, or decision criteria rather than merely restating familiar advice.
- Depth: answer the follow-up questions, constraints, tradeoffs, and failure cases implied by the main query.
- Timeliness: explain what changed and why the change affects the reader. Do not refresh a date when the substance is unchanged.
- Actionability: give the reader a next step, setting, filter, check, or decision they can actually use.
The conversational-query export can guide this work. If users repeatedly add the same constraint to a broad query, that constraint belongs in the content. If they keep asking about an old reputational issue, silence does not make the concern disappear; a current, factual answer may be necessary.
Treat local relevance as audience fit, not decoration
The update placed more weight on locally relevant content from domestic websites. A non-U.S. publisher serving a U.S. audience could therefore have experienced reduced Discover traffic during the initial U.S. rollout.
Segment that audience before reacting. If the decline is limited to U.S.-facing pages, examine whether the material genuinely reflects the market’s places, rules, products, terminology, and context. Do not disguise the site’s origin or add superficial location phrases. If your strongest expertise belongs to another market, preserve it and make the geographic scope explicit.
Remove the gap between the headline and the page
Discover’s move away from sensational content makes the headline-content relationship a practical audit point. The title should communicate the real value of the page without withholding the central fact or overstating the evidence.
- Put the actual subject and consequence in the headline.
- Remove unsupported superlatives, manufactured urgency, and curiosity gaps.
- Deliver the promised answer near the beginning, then add context and depth.
- Check that the headline still makes sense when separated from the image and surrounding feed.
- If a restrained headline makes the content seem uninteresting, improve the substance instead of restoring the hype.
Google also said its systems would continue personalizing Discover around favored creators and sources. You cannot force that preference, but consistent subject expertise and dependable promises give readers a coherent reason to recognize and return to your work.
Key takeaways and your next move
- Diagnose Search and Discover separately; a change in one surface does not establish a change in the other.
- The February 2026 Discover core update ran from February 5 through February 27 and initially covered U.S. users viewing English content.
- Use the 10-word Search Console regex to find conversational demand, but do not label every long query as an AI prompt.
- Track recurring prompt themes rather than every literal query variation.
- For Discover, strengthen sustained topic expertise, original depth, genuine timeliness, honest local relevance, and headline-content alignment.
- Make changes only after you have identified the affected surface, audience, metric, and page cluster.
Your next visibility review should end with one explicit hypothesis. Write down the surface, change window, country, affected pages, impression pattern, click pattern, proposed change, and metric that would support or weaken the hypothesis.
Then change the smallest relevant layer. Fix measurement when the data disagrees, improve a page when conversational demand exposes an answer gap, or strengthen a coherent topic section when the Discover loss is concentrated there. Evaluate the result on the same surface and segment that led you to act.
