The short answer
Repair an observed access or indexing fault. Refresh an existing page when it already owns the reader's decision but needs a better answer or current evidence. Create a new page when the reader needs a distinct decision and you can supply a substantive contribution. Keep investigating incomplete traffic evidence while preparing useful work in parallel; a weekly decline does not identify its cause or prove that more pages will reverse it.
Write down what decreased before naming a cause
A fall in search impressions, a fall in clicks, and a fall in measured visitors describe different observations. Record the property, report, filters, timezone, exact dates, absolute counts, and processing state. Treat consented analytics as its own coverage. A pageview count is not a count of inquiries or attributed answers.
Google's troubleshooting guidance separates technical problems, changes in rankings, changes in demand, and reporting issues. It recommends examining affected pages and queries. A site average also changes when the set of observed queries changes. Compare recurring queries before describing the entire movement as a ranking loss; retain the unreported-query gap.
This guide contributes an editorial allocation method. It is not a diagnosis of a particular site or a prediction about a search engine or assistant. The seven downloadable situations are fictional review examples.
Source: Google Search Central: Debugging drops in Google Search traffic
Choose an action from the evidence and reader task
Use the table to choose the primary action for one resource. Several resources can have different actions at the same time. A measurement uncertainty can remain open while a separate, well-supported research question moves into preparation.
The decision labels are proposed editorial judgments. They are not a score supplied by Google, an automatic content-quality classifier, or a mandatory waiting period.
| Observed situation | Primary action | Evidence needed to finish |
|---|---|---|
| The intended canonical page returns an error or carries an unintended noindex | Repair the demonstrated fault | Correct response and directives; Google indexing state checked separately |
| The latest report is still processing or the periods have different boundaries | Investigate the measurement | Comparable settled periods or an explicitly provisional comparison |
| An existing answer owns the same reader decision but omits a necessary example | Refresh that answer | A new useful example with checked inputs and sources |
| A destination's documented contract changed | Verify and refresh the affected guidance | Current primary source; every affected format and artifact reviewed |
| A new task requires captured observations rather than definitions | Prepare a distinct study | Declared sample, actual evidence, all gaps and reproducible analysis |
| A new task needs an executable integration rather than another explanation | Prepare the workflow | Working example, observed failures and bounded acceptance cases |
| The proposed page restates the same answer using a query variation | Consolidate the answer and navigation | Intent comparison and a useful addition to the existing resource |
A refresh changes the answer, not only the date
Name the new contribution before changing the existing page: a corrected claim, an observed case, a missing decision step, a verified artifact, or a clearer explanation that resolves an actual ambiguity. Keep the stable URL when the question is the same. Record the affected sections and review the HTML, downloadable material, and alternate editions together.
Google's content self-assessment asks whether a resource adds original information or analysis and helps its intended audience complete a task. It explicitly rejects changing dates without substantive changes or adding volume simply to make a site appear fresh. Those are documented recommendations; they do not establish the selection mechanism for any particular provider.
An added related link can improve the reading path without becoming a substantive article revision. Retain the publication history and distinguish navigation changes from changed evidence.
Source: Google Search Central: Creating helpful, reliable, people-first content
A new decision earns a page; a new format earns access
Write the proposed reader question beside the closest existing answer. Compare the action, inputs, output, and evidence required. If all four substantially overlap, strengthen the existing resource. If the new question requires a distinct observation set or executable task, explain that difference in the editorial record.
For example, an identifier definition and an audit of which exact variant a URL selects have different jobs. A margin formula and a working request/response acceptance workflow also have different jobs. Separate pages can connect those jobs without claiming that one caused demand for the other.
HTML, Markdown, JSON, and a repository copy of one contribution are access routes. Count them as one substantive resource unless a separately declared format experiment actually studies format selection. If overlapping URLs already exist, review the canonical and linking plan; canonicalization addresses duplicate representations, not the usefulness of the answer.
Source: Google Search Central: How to specify a canonical URL
Allocate a small batch across different kinds of work
A fictional editor has six person-hours available. The example allocates one hour to an access investigation, one and a half hours to an existing answer, two and a half hours to a distinct study, and one hour to shared review. The sum is six person-hours. These are planning assumptions, not observed completion times or promised productivity.
Reserve review time explicitly. If the study cannot produce defensible observations within its allocation, retain the gaps and adjust scope. The other tasks can continue. Do not turn insufficient evidence into a confident conclusion to keep a planned publication date.
| Work | Planned person-hours | Reviewable output |
|---|---|---|
| Investigate a demonstrated access concern | 1.0 | Response/directive evidence and unresolved indexing state |
| Refresh one existing answer | 1.5 | One checked addition at its existing URL |
| Prepare a distinct observed study | 2.5 | Declared sample, results and explicit gaps |
| Review the batch | 1.0 | Sources, arithmetic, exports and reading paths checked |
| Total capacity | 6.0 | Preparation only; release decision recorded separately |
Keep the decision and its unknowns in one record
Use the download to record the reader question, closest resource, observed trigger, chosen action, contribution, sources, effort allocation, access state, and next observation. A missing value stays null. The seven filled situations illustrate the method; the blank record is for your own evidence.
These files are manual worksheets. They contain no automatic classification, platform submission, analytics collection, or publication step. Keep private analytics and user-level exports in your own authorized records rather than copying them into a public example.
Download seven fictional decisions and a blank record (JSON)Declare what would count as progress
An access repair succeeds when the demonstrated fault is corrected; index inclusion remains a separate observation. A completed guide succeeds as a preparation task when its claims, sources, examples, exports, and presentation pass review. Audience outcomes require subsequent measurement.
For discovery work, record exposure, attributed answers, referrals, and independent use separately. An internal probe, a fixture pass, a download, or a successful deployment does not establish an audience result. Preserve the baseline prompts and comparable reporting periods. Multiple simultaneous improvements can be useful while leaving causal attribution unknown.
At the next review, state which evidence changed and which decision follows. A quiet result does not require discarding a useful resource; an appealing hypothesis does not justify describing unobserved gains.
Download the blank editorial decision worksheet (Markdown)Sources & scope
Primary references checked for this edition. The notes below distinguish source-backed facts from the frameworks and examples proposed in this guide.
- Google Search Central: Debugging drops in Google Search traffic ↗
Supports checking technical and reporting issues, affected pages and queries, comparable periods, and ranking/demand changes. It does not diagnose this site or promise recovery.
Checked October 17, 2026 - Google Search Central: Creating helpful, reliable, people-first content ↗
Supports assessing originality and reader usefulness and avoiding date-only changes or content volume as a freshness tactic. The allocation method and fictional examples are authored proposals.
Checked October 17, 2026 - Google Search Central: How to specify a canonical URL ↗
Supports handling duplicate or similar URL representations. It does not establish whether a proposed editorial contribution deserves a separate page.
Checked October 17, 2026
AI-assisted research and drafting. Provider-specific claims link to primary sources. Frameworks are editorial proposals; worked examples are illustrative and are not employer performance results.
Editorial policy & corrections ↗Cite this guide
George Kelly. Traffic fell: repair, refresh, or publish a new page? iamgeorgekelly. Updated October 17, 2026. https://www.iamgeorgekelly.com/field-guide/content-refresh-or-new-page
Keep the source notes and example labels with an excerpt. For a provider requirement, follow the original documentation in Sources & scope.