What is an AI visibility API, and which access path fits?
What is an AI visibility API?
An AI visibility API is a programmatic way to read the data an AI visibility tool collects: which prompts were sent to ChatGPT, Gemini, Google AI Mode, Perplexity, Claude and other engines, whether a brand was named or cited in each answer, in which position, which sources the engine relied on, and how competitors compared. The dashboards most teams start with show the same data, but a dashboard is a dead end once you want to join AI visibility with web analytics, revenue or your own reporting stack.
Access usually arrives in one of three forms. A REST API returns JSON for your own scripts, warehouses and internal tools. An MCP server (Model Context Protocol) exposes the same data as tools that an AI assistant such as Claude or ChatGPT can call in conversation. A connector (for example a Looker Studio community connector) skips code entirely and feeds a BI tool. Some vendors offer all three, many offer one, and several offer none at all on their entry plans.
API, MCP server or connector: which access path fits?
| Access path | What you get | Typical consumer | Watch out for |
|---|---|---|---|
| REST API | Raw JSON per project, prompt, answer or metric | Data engineers, internal tools, warehouses | Plan gating, rate limits, pagination, history depth |
| MCP server | Tools an assistant can call, usually read-only | Analysts and marketers working in Claude, ChatGPT or an IDE agent | Which tools are exposed, auth method, write access |
| BI connector | Pre-shaped report tables | Looker Studio dashboards, client reporting | Fixed report types, refresh cadence |
| CSV / scheduled export | Files on demand or on a schedule | Spreadsheets, one-off analysis | Manual steps, no incremental loads |
| DIY with LLM APIs | Your own answers from model APIs | Teams with engineering time and narrow needs | API answers can differ from consumer apps; you own sampling and storage |
The right choice depends on who consumes the data. A weekly client report is best served by a connector; a warehouse that joins AI visibility with GA4 sessions needs an API; an analyst who wants to ask “why did our citation rate drop on Perplexity last week?” gets the most out of an MCP server.
What should you check before building on one?
- Plan requirement. API access is often reserved for higher tiers while MCP access may start lower. Check the pricing page, not the marketing page.
- Authentication and scopes. Look for keys that can be scoped by project and by permission (read, write, export), and for OAuth where an assistant connects on a user’s behalf.
- Collection method. Ask whether answers come from the consumer apps (what users see) or from model APIs, and from which location and language. This decides what your numbers mean.
- Granularity and history. Can you fetch individual answers and cited URLs, or only aggregated scores? How far back does history go?
- Limits and units. Rate limits, pagination and whether API calls consume the same credits or prompt quota as the dashboard.
- Read versus write. Read-only access is safer for assistants. Write access (creating prompts or projects) matters if you want to automate setup.
Why "API answers" and "UI answers" are not the same thing
When you call a model API yourself, you choose the model version, the system prompt, whether web search is enabled and the location the search runs from. The consumer apps make those choices for the user, add personalization and change them without notice. Trackers that collect from the consumer interfaces measure what people actually see; trackers built on model APIs measure a controlled approximation. Both are useful, but they should not be mixed in one trend line without saying so.
Should you build your own tracker instead?
Building a basic tracker on LLM APIs with web search is a weekend project: loop over prompts, call each engine, store the answer and parse brand names and cited URLs. The work starts afterwards. You own prompt design, sampling frequency, entity matching (brand variants, product names, competitors), source normalization, history, and cost control as prompt counts multiply across engines, markets and days. Our build-your-own guide walks through the costs and caveats; for most teams a vendor API plus a small amount of glue code is cheaper than maintaining a parser.
How does Finseo expose its data?
Finseo offers all three access paths. It suits enterprise teams with SAML single sign-on, roles per project and an audit trail, is hosted in Germany (Frankfurt), GDPR-compliant with a DPA and ISO 27001 certified. The REST API lives at https://api.finseo.ai/v1 with endpoint groups for projects, prompts, metrics, competitors, sources, tags, fan-outs, attribution, tasks and export; it uses Bearer API keys that are stored SHA-256 hashed and scoped by project, with read, write and export scopes, and is included from the Growth brand plan and in all agency packages. The MCP server at api.finseo.ai/v1/mcp exposes 15 read-only tools covering visibility, competitor ranking, gaps and head-to-head comparisons, sources, prompts, answer texts, sentiment, tags, fan-outs and tasks; Claude and ChatGPT connect with OAuth 2.1 and PKCE, other clients with an API key, and MCP access is included from the Starter plan. A native Looker Studio community connector authorizes with a read-scoped API key and offers five report types: visibility per day, competitor ranking, top cited sources, prompt performance and query fan-outs. Details: finseo.ai/developers/api and finseo.ai/developers/mcp.
Every other vendor is covered in the guides with the same questions: does an API exist, is there an MCP server, how does auth work and which plan is required. Where a vendor does not publish an answer, we say “not published” rather than guess.