> ## Documentation Index
> Fetch the complete documentation index at: https://docs.mavel.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Surfaces

> The eight AI engines Mavel tracks, how to enable them per project, and why per-country collection matters.

A surface is an AI engine that Mavel queries with your prompts, capturing the full answer, the cited sources, and any brand mentions. Mavel tracks eight surfaces:

| Surface             | Provider      |
| ------------------- | ------------- |
| ChatGPT             | OpenAI        |
| Perplexity          | Perplexity AI |
| Gemini              | Google        |
| Claude              | Anthropic     |
| Google AI Overviews | Google Search |
| Google AI Mode      | Google Search |
| Microsoft Copilot   | Microsoft     |
| Grok                | xAI           |

Each surface has different answer styles, source citation habits, and update cadences. A brand that appears consistently in ChatGPT may barely register in Google AI Overviews, and vice versa. Running your prompts across all eight surfaces tells you where you have coverage and where you have gaps.

## Enabling surfaces for your project

Surfaces are enabled at the project level: you choose which engines to track, and Mavel runs every active prompt against each enabled surface, daily.

The engines available to your project depend on your plan. See [Plans](/account/plans) or [mavel.ai/pricing](https://mavel.ai/pricing/) for the current limits. New projects start with a default set of engines enabled; you can add or remove surfaces from your project settings.

<Note>
  Your plan determines the maximum number of surfaces you can track simultaneously. Locked surfaces show an upgrade prompt in the interface.
</Note>

## Per-country answer collection

Each project has a **default country** you set during onboarding. When Mavel runs your prompts, it collects answers from that country, using a residential browser or proxy in that location. Your prompts inherit this country setting automatically.

This matters because AI search answers are geographically variable. The same question asked from Germany and the United States can return different brands, different cited sources, and different framing. A brand with strong US coverage may be invisible in European results, or the dominant source in a local market may never appear globally. Mavel anchors each collection to your target market so your measurements reflect what your actual buyers see.

## How each surface is queried

Mavel uses different collection methods per surface, depending on what the engine requires:

* **Browser-based surfaces** (Perplexity, Gemini, Google AI Overviews, Google AI Mode, Grok) run through a real browser to collect answers as a user would see them.
* **ChatGPT** uses a specialized scraping path suited to its answer-time behavior.
* **Microsoft Copilot** uses a third-party browser automation service because of its security challenges.
* **Claude** is queried directly through the Anthropic API.

You do not need to configure any of this; it is handled automatically. The distinction matters only when you see variation in collection success rates: some surfaces have higher anti-bot ceilings than others, so first-pass collection rates differ. For most surfaces, Mavel retries failed collections on a fresh IP, which recovers the majority of them; Microsoft Copilot is the exception: its failures are not IP-retried, since its anti-bot wall requires the specialized collection path rather than rotation.

## Interpreting cross-surface data

<Warning>
  AI engines are non-deterministic: the same prompt asked twice may return different answers. Mavel measures trends across daily runs. Judge your visibility by the trend over days and weeks, not by any single answer.
</Warning>

Compare a brand's trend within one surface over time before comparing across surfaces. Cross-surface comparisons are directional: a surface that cites sources heavily (like Perplexity) will produce different citation-level metrics than a surface that generates conversational answers without inline links. The metrics are designed to be comparable, but the underlying answer style differs.

For a full description of what each metric measures, see [Metrics overview](/results/metrics-overview).