Turn a research question into a defensible keyword shortlist

Move from a research question to an approved handoff

Know what one run records

A Keyword Research run freezes its mode, normalized inputs, country, language, competitor settings, quote, request owner, and start time. Its candidates, progress, terminal status, settled charge, and import count accumulate against that contract. This makes the run reopenable evidence, not a live view that silently changes whenever the setup form changes.

Some intelligence shown beside a candidate can be newer than the paid row. Current Keyword Database enrichment, a cached SERP, or a recent domain snapshot can be joined when results are read. Keep those sources and dates separate from the run's own metrics before comparing or exporting evidence.

Access and write permissions are separate gates

The workspace route is available only when the Keyword Discovery feature is enabled and the organization has the Keyword Database entitlement. Viewing the workbench requires keyword view access. Starting a paid run additionally requires keyword enrichment access, and importing candidates requires keyword add access.

Choosing an existing or new manual list also requires permission to manage keyword views. A teammate may therefore be able to inspect a saved run without being allowed to spend balance, import rows, or change list membership. Treat a missing action as an access boundary before assuming that the run or data is broken.

Choose among the three current modes

Older saved runs may carry a legacy mode name, including Google Suggestions. The current setup offers Quick, Deep, and Competitors. Treat an old label as historical configuration rather than proof that the old recipe is still available for a new run.

Inputs and market define the evidence boundary

Seed-led research accepts typed phrases or a one-time prefill from an existing manual or saved list. Selecting a keyword from workspace suggestions only copies it into the research input. Competitor research uses normalized domains instead. Neither action changes the Keyword Database before an explicit import.

Country and language travel with the provider requests. They affect suggestions, rankings, volume, CPC, and difficulty context. Research in Germany and German is therefore a different claim from the same seed in the United States and English. The current selector offers a curated set of country-language pairs rather than arbitrary combinations.

A quote and a worker completion prove different things

Crawl Foundry refreshes the estimate after setup changes and disables the start action until the input is valid, the quote is current, and spendable organization balance is sufficient. Starting atomically creates a pending run and reserves the current server-priced quote. Spendable balance already excludes other active reservations.

A pending row proves admission, not provider execution. Production dispatch passes through an outbox, and the worker must claim the run before the claim deadline. Progress, lease heartbeats, terminal status, candidate count, error code, and final charge provide the later evidence. Dispatch failure, claim timeout, an expired running lease, cancellation, or provider failure can all end a run without a usable result set.

Research does not silently expand the Keyword Database

Candidates remain attached to their run until a person selects and imports them. Filters, term groups, sorting, selection, CSV export, reopening history, and rerun preparation do not add keywords, lists, tags, tracking schedules, pages, briefs, or articles.

That separation protects the shared inventory from exploratory noise. It also means that a promising candidate is not operational merely because it appears in a paid result. Someone still has to judge relevance, intent, business fit, feasibility, and the appropriate next owner.

Read candidates as bounded evidence

A result row can contain the keyword, discovery source, origin seed, volume, CPC, competition, difficulty, intent, monthly history, link-derived fields, and competitor ranking context. Missing values remain unknown and render as a dash. They do not become zero. A row can also be marked already present or not importable, but the import path rechecks live database state and permits a deliberate no-data import.

The workbench adds useful views over a bounded scan. Page-level topic grouping, term counts, matching selection, and CSV export do not necessarily cover every stored row. Their limits stay part of the interpretation, especially on a large Deep or Competitors run.

Import is the approval boundary

The import target can be the database alone, an existing active manual list, or a newly created manual list. Smart and hybrid lists are not valid direct targets. Existing active keywords are not duplicated, but they can still be added to the chosen manual list. Soft-deleted rows are restored, and genuinely new rows carry the available discovery fields into the database.

Import does not purchase a new enrichment, fetch a SERP, begin Position Tracking, create a page, draft content, or publish anything. Those are separate decisions with their own scopes, costs, and evidence requirements.

Reuse history without mistaking it for current truth

Run history is paginated newest first and keeps paid result rows reopenable under the current implementation. A saved run preserves what was asked and returned at that time. It does not refresh its provider metrics when the market changes, although current workspace intelligence may appear beside the stored rows.

Rerun opens a new setup with the old mode, inputs, market, and competitor options prefilled. It does not refresh the old run in place and it does not start automatically. Once confirmed, the rerun is a distinct paid claim with a new quote, status trail, candidates, and settlement.

Review the decision, not only the candidate count

• The mode, market, inputs, and limits match a written research question.
• The terminal run state and any partial or missing evidence are understood.
• The shortlist is smaller and more purposeful than the raw candidate set.
• Metrics, source, intent, SERP context, and current workspace signals are not collapsed into one score.
• Every imported row has a plausible use, owner, or manual list.
• The saved run remains attached to the handoff so later reviewers can recover its date and scope.