What local tracking does and does not measure

Local tracking answers one specific question: when an AI assistant is asked this prompt from this city, does it mention you? It is a precise tool, and precise tools have edges. Knowing where those edges are stops you from reading a city result as something it is not.

It measures a requested location, not a guaranteed one

When you choose Chicago, AI Peekaboo requests the targeting supported by each model and records it with the check. The requested city and the effective city, state or country targeting can differ. A saved targeting record describes how the request was made; it does not prove that the answer is geographically correct or that a particular person would see it. AI answers also vary with personalization and history. Treat repeated checks over weeks as observations, not a guarantee.

Read the model and targeting precision

Availability depends on your plan and rollout. Some brands have Google AI Mode only; expanded local tracking can also include Google AI Overviews, ChatGPT, Gemini and Perplexity. Preview coverage before confirming: Google models can support city targeting, while other models may support a US state or country. An unavailable model has no supported targeting for that selection.

One city per prompt

A prompt can track one city at a time. Measuring the same question in two cities means a second prompt: it uses a prompt slot, runs your plan's standard analysis, and its city counts toward the 10-per-brand local limit. Rotating one prompt's city over time costs nothing extra. Changing the city starts a fresh history for the new city and keeps the old one intact, so switching never loses data, but it does not let you compare two cities side by side inside one prompt.

Local results stay out of your score

Your AI Visibility Score, the prompt table, the Overview charts and your competitor comparisons are all built from standard tracking only. Local answers are stored separately and never enter those numbers. That is deliberate: a brand that turns on ten local checks in a city where it is strong would otherwise inflate its headline score without anything changing in the rest of the market. If you want a local figure in a report, take it from the Local tracking view of the prompt.

Missing checks are not zeros

A local check can be pending, failed, or returned but not analysable. None of these count as "your brand was not mentioned". The Local performance panel counts brand appearances only across checks that completed and were analysed, and it says "No completed checks" rather than showing 0% when nothing has finished in the selected range. A low number of completed checks in a short window is a reason to widen the date range, not a verdict.

Capacity and cadence

Each brand can hold up to 10 prompts with local tracking. Expanded model coverage also consumes one additional prompt slot per enabled local location, so your remaining plan capacity may impose a lower limit. Local checks are queued with each scheduled analysis and processed by a shared worker, so they can arrive later than your standard results. Local checks do not retry on their own: if a provider request fails, the failure is recorded and the next scheduled run tries again. The REST API and MCP connector expose local results through dedicated operations. Ordinary visibility endpoints, standard exports and Looker Studio continue to report standard tracking; adding local API support does not add local data to those surfaces.

API report limits and recovery

A local API or MCP report selects one prompt, target and model. Use a 7, 14, 30 or 90 day window (14 by default), or an explicit range of at most 90 days. A report with more than 300 checks, or with different effective geographies in the same window, is rejected; request a shorter range. Old targets remain discoverable through the prompt's settings history after you change or stop tracking.

Reports show at most 10 answers, 20 competitors and 30 sources. Read the evidence completeness fields before treating these lists as exhaustive. Configuration lists and target/model history are also bounded to 100 entries and flag when more exist. These operations are not complete exports.

If you see COVERAGE_CHANGED, refresh coverage and review it again. For CONFIG_VERSION_CONFLICT, reread settings before deciding whether another change is needed. For LIMIT_EXCEEDED, check both available prompt slots and the 10-prompt local limit. A provider coverage failure can be retried later; EVIDENCE_UNAVAILABLE means saved evidence could not be read and needs support investigation.

When contacting support, include the operation or tool name, brand and prompt IDs, selected target and model, date range, error code, time of the request and REST metadata.requestId if present. Never include your API key or authorization header.

If local tools are missing, refresh your MCP connection after the server update. A missing route or tool may mean that deployment does not include local API support yet; an enabled: false response means local tracking is unavailable for that brand. Contact support if dashboard availability and the connection disagree.