The bulk tool is a screening mechanism, not a decision. You supply a list of candidate coordinates; it returns one row of raw observations per site, measured over the radius you gave for that site. Nothing in the response ranks the rows, scores them, or declares a preferred site — those are conclusions the underlying data cannot carry, and the tool is built so it cannot imply them by omission. If you need a comparison rather than a measurement, the business location analysis page explains what each individual column means, and how to choose a business location covers the sequence a shortlist should follow.
| Column | What it is | What it is not |
|---|---|---|
| Label | The identifier you supplied, echoed back | A rank or an ordering |
| Status | Whether that row analyzed, and its error text if not | A quality score for the location |
| Places | Count of place records returned inside the row's radius | A count of competitors; the place set includes every business in the radius |
| Categories | The three most frequent business categories in that set | A market-share estimate |
| Mean rating | Arithmetic mean of published ratings, where a place has one | A quality judgement of the area |
| Reviews total | Summed review counts across the same set | Demand, footfall or revenue |
| Closed | Places recorded as closed or permanently closed | Turnover, failure rate or vacancy |
| Population | Gridded resident headcount summed over the radius | A customer count; see the methodology page for resolution limits |
The header must be exactly label,lat,lon,radius_m or
label,lat,lon,radius_m,biz, in that order. A label is required and
may not be empty. Latitude must fall between -90 and 90 and longitude between
-180 and 180. The radius is an integer number of metres between 100 and 20,000.
The optional biz field takes a single comma-free business-type key,
checked against the collector's own filter list; an unrecognised key is an error
rather than a silent fallback to all businesses.
Two limits are enforced before any collection is attempted. A file longer than
twenty rows is refused outright, and a request from a plan without the bulk
entitlement is refused before the run is created. Both refusals report
google_requests_sent: 0, so the claim that nothing was billed is
verifiable rather than assumed. Within a run, rows are processed one after
another rather than in parallel: a batch's cost is bounded by construction instead
of by a rate limit.
Allowance is metered per location, not per request, and is checked for the whole batch up front so a refusal costs nothing. Only rows that analyze successfully are consumed afterwards — a failed location is not billed as though it had succeeded. This is why a batch of twenty difficult coordinates can cost materially less than a batch of twenty easy ones, and why the consumption figure at the end of a run is worth reading rather than assuming it equals the row count.
Bulk analysis sits on the Agency tier of the plan ladder rather than being available on credit-metered plans, because one bulk site is among the most expensive single operations in the product. The pricing page carries the current allowances; they change, so the page is the authority and this description is not a quotation. Once rows are attached as dated snapshots, portfolio health reads across every campaign and shows which clients are going stale.
A run can be pointed at a campaign you already have. Each successful row is matched by label — case-insensitively, and optionally against a site-label prefix — to a site that already exists there, and the observation row is attached to it as a dated snapshot. A row whose label matches nothing is reported as unmatched and skipped: this endpoint never creates sites, so a typo in a label cannot quietly grow a campaign with rows you did not intend. Campaign-based comparison of two dated states is covered under business expansion analysis, and portfolio health reads the snapshots across every campaign once rows start accumulating.
There is no total, no average across sites, and no "best site" line, and that omission is deliberate. Averaging a population figure across sites measured over different radii produces a number that describes no actual place. Choosing one row over another is a judgement about your business, your customer and your risk tolerance — inputs this tool does not have and does not ask for. The measurement is the product; the decision stays with you, and the methodology page sets out in full where each figure comes from and where each one stops.