When you see a Google Analytics keyword not provided, GA4 cannot show the exact organic query that brought an individual user because Google does not pass that query in the referrer. That does not leave SEO teams blind. Use Google Search Console for queries and search exposure, GA4 for landing-page behavior and key events, and connect the datasets at the landing-page and period level. This supports decisions, but it is not query attribution for a specific session.
This guide explains how to work with not provided keywords in Google Analytics without promising to “unlock” unknown search phrases. It separates observed Google data from calculations and makes the boundary between fact and estimate visible.
Use three evidence levels before analyzing keywords
| Evidence level | What you know | How to label it |
|---|---|---|
| Observed | Search Console recorded a query, impressions, clicks, CTR, and position for the selected slice; GA4 recorded landing-page behavior separately. | Google data |
| Matched | Query demand and organic outcomes moved together for the same landing page and period. | Aggregate relationship, not session attribution |
| Modeled | An algorithm estimated which query may belong to a session using several signals. | Prediction with a documented confidence level |
This distinction is the article’s operating rule. It prevents a common reporting error: presenting a useful page-level relationship as if Google had revealed the hidden keyword for an individual visit.
What does keyword not provided mean in Google Analytics?
The (not provided) label means Analytics recognized an organic visit but did not receive the search phrase. This is not normally a broken GA4 setting. A not provided keyword is absent from the source event, so changing a filter, channel definition, or report cannot recreate it.
Do not confuse (not provided) with (not set). The second label means Analytics could not populate a particular dimension in that report row. Its cause depends on the dimension, tagging, consent, and report scope. It is a missing classification value, not an encrypted organic query.
The practical consequence is that “which query produced this session?” often has no verifiable answer. Ask two answerable questions instead: “Which queries generated impressions and clicks for this page?” and “What did organic visitors do after entering on this page?”
Which data comes from Search Console and which comes from GA4?
| Question | Source | Valid conclusion | Invalid conclusion |
|---|---|---|---|
| Which queries generated impressions and clicks for a page? | Search Console | Queries, clicks, impressions, CTR, and average position within the selected slice | A selected query caused a specific session |
| Which landing pages received organic sessions and key events? | GA4 | Behavior and outcomes by page, channel, and supported dimensions | The hidden query that caused an individual event |
| Did search demand and page outcomes move together? | GSC + GA4 | An aggregate relationship by date and landing page | Causation or exact keyword attribution |
| What were the omitted queries? | No report | You can quantify part of the difference between totals and visible rows | The text of omitted queries |
The official Search Console link for GA4 adds two reports. Google Organic Search Queries displays queries with Search Console metrics, but you cannot freely combine them with Analytics dimensions. Google Organic Search Traffic connects landing pages with compatible metrics from both systems. That boundary is useful: the interface does not pretend it knows the query behind each session.
Plan for the operational limits before troubleshooting. Search Console data can take 48 hours to appear, linked reports cover at most 16 months, and the Search Console report collection is unpublished in GA4 by default. Creating the link requires an Editor role in GA4 and verified ownership of the Search Console property.
Why do GSC clicks and GA4 sessions differ?
The totals should not be expected to match. Search Console measures an action in Google results, while Analytics measures events after a website loads. A person may click more than once; a page may fail to load; the tag may be blocked; consent settings may change collection; and the systems process time and identity differently.
Search Console also does not list every query row. Google explains that anonymized queries are omitted for privacy, while their clicks can remain in chart totals. Tables also retain the most important rows rather than every possible row. A sum of the Queries table that is lower than the chart total is therefore not proof that an export failed.
Consider an illustrative calculation. The chart shows 1,000 clicks, while visible query rows sum to 860. The difference is 140 clicks, or 14% of the total. This is our calculation on hypothetical data. It quantifies an unitemized share; it does not reveal 140 unknown search phrases.
How can you find not provided keywords in Google Analytics?
- Lock one date range. Use the same dates in both sources and avoid incomplete recent days while Search Console data is still settling.
- Choose one search type. Do not casually combine Web, Image, Video, and News because their page sets and user behavior differ.
- Export GSC at query × page level. Treat this as the source for organic keywords and Google search terms. For larger properties, use the Search Console data export and document API limits.
- Export GA4 landing pages for Organic Search. This is the behavioral side of the analysis, not a hidden keyword report. Keep sessions, engaged sessions, key events, and revenue only when those metrics are correctly configured.
- Normalize URLs. Apply one rule for protocol, host, trailing slash, and parameters. Do not merge distinct canonical URLs merely because their paths look similar.
- Join by landing page and period. Add country or device only when the dimensions are compatible and the resulting sample remains useful.
- Decide at page level first. If query impressions rise while page outcomes do not, investigate intent, the first screen, and the path to action. Do not assign an event to one query without evidence.
GSC Workspace places queries, pages, and GA4 reports in one workspace while keeping source ownership explicit. GSC remains the source for search performance; GA4 remains the source for on-site behavior.
How do you turn Google Analytics keyword analysis into an SEO decision?
Start with a landing-page decision rather than an attempt to allocate each conversion to a phrase. A page has an observed set of queries and an observed organic outcome. That is enough for many high-value actions.
- Many impressions, few clicks. Compare intent with the title and snippet, then inspect the page in CTR Lens.
- Clicks rise, but key events do not. Check whether the query group matches the offer, whether the mobile first screen is clear, and whether event tracking works.
- One page attracts multiple intents. Cluster the queries and decide whether the page needs a new section, a separate page, or an explicit exclusion of an irrelevant topic.
- One query appears with several pages. Check country, device, and date before diagnosing cannibalization.
- Chart clicks materially exceed listed queries. Record the gap as a reporting constraint instead of filling it with invented keywords.
For a large export, use the detailed guide to the Search Console API and pagination. Google recommends daily extraction and warns that grouping by page and query can drop some data. Preserve that limitation in the analysis log.
When do you need the API or BigQuery?
The interface is adequate for a quick check of one page or a small property. The Search Analytics API is appropriate when you need a repeatable extract, more rows, stable filters, and your own validation. Google documents Search Analytics API pagination and data limits. API extraction still does not expose anonymized queries.
Large sites may need bulk export to BigQuery. According to Google, bulk export removes the daily row limit while retaining privacy filtering. BigQuery gives you a more complete observed table, not a way around privacy protection.
A probabilistic model is reasonable only when the business knowingly accepts an estimate. The model needs documented features, a validation sample, uncertainty, and a clear result label. A “predicted query” field must never be mixed with an observed Search Console query.
What does this workflow not do?
This workflow does not decrypt Secure Search, bypass Google controls, or recover personal data. It also does not prove that a query change caused a conversion change. Seasonality, paid campaigns, redesigns, prices, availability, or broken measurement may change the landing-page outcome.
Granularity creates another limit. Every added dimension reduces row volume and may increase the impact of privacy filtering. Begin with page × date, then add only one decision-relevant slice. When the sample is small, combine periods or decide at topic-cluster level instead of pretending a single phrase is reliable.
What should you do next?
Link Search Console to GA4 and inspect the official reports first. Export query × page and Organic Search landing pages for the same period, normalize URLs, and build a decision table. Use GSC Workspace for a repeatable review or begin with the GSC export when you need a file for your own model. Keep observed metrics and estimates in different columns.