How Bee LLM works

What Bee LLM is

Bee LLM measures your brand's visibility inside AI answers. Every day it asks the questions a potential customer would ask ChatGPT, Gemini, Perplexity and the other engines, and analyses the real answers: whether you appear, in what position, with what sentiment, who appears before you, and which websites the AI pulls its information from.

That tells you three things Google Analytics cannot: whether AI recommends you when someone asks about your industry, which competitors get the recommendation instead of you, and what content to publish or which sources to work on so it starts citing you.

First steps

When you create your account, the setup assistant asks for your website and your market, then automatically generates your company, a first list of prompts and your competitors. You can edit everything afterwards.

• Company. Your brand: name, domain and name variations (for example your legal name or short name). Variations matter: AI sometimes names you in different ways, and without them those mentions would not count as yours.
• Prompts. The questions sent to the engines every day. The best prompts are what a customer who does not know you yet would ask ("best tool for X", "lawyer for Y in Barcelona"), not searches for your own brand.
• Competitors. The brands you are compared against in the ranking. Each plan includes a different number.
• First measurement. Creating a prompt launches its first run immediately; results appear within minutes. After that it repeats automatically.

Key concepts

• Monitored prompt. A specific question, in a country (and optionally a city), that Bee LLM sends recurrently to the AI engines.
• Run. Each pass of a prompt through the engines. One run produces one answer per engine.
• Mention. Every time a brand (yours or any other) appears in an answer. For each mention we store its position, sentiment and the role it plays in the text.
• Source. The URLs the engine cites or consults to build its answer. They are the main lever for improvement: if you gain presence on the sites AI already uses as sources, it is far more likely to mention you.
• AI queries (fan-outs). The real sub-searches ChatGPT or Gemini run on the internet while answering. They tell you literally what the AI searches for, which is gold for deciding what content to create.
• Project. A brand or client with its own prompts, competitors and settings. The Pro and Agency plans allow several projects in the same account.

Metrics

Visibility

Your brand's weighted presence as a percentage of all answers analysed in the period. It is not just "how many times you show up": each mention is weighted by how prominent it is in the answer, so appearing first in a list is worth more than a passing mention at the end. It is computed globally, per engine and per prompt.

Presence

The simple percentage of answers where you appear, unweighted. Useful as a quick "do they see me or not" read.

Average position

Where in the answer you appear on average, on a 0-10 scale where higher is better (10 would be always first). You can appear in many answers but always last: high presence, low position.

Best AI

The engine where your brand has the most visibility, provided there is enough sample to claim it. With little data it is shown as provisional.

Citability

A 0-100 score estimating how likely AI is to cite your domain as a source. It combines your real presence among the sources engines cite with your domain's authority. Recalculated weekly.

Sentiment

The tone the AI uses about each brand in each mention, from negative to positive. Aggregated per brand, per prompt and per engine.

Mention role

Beyond sentiment, each mention is classified by the role it plays: main recommendation, alternative, passing mention, comparison, negative example, and so on. Being recommended is not the same as being used as a counterexample.

AI engines

Bee LLM tracks up to 8 engines, depending on your plan: ChatGPT, Gemini, Perplexity, Google AI Overviews, Google AI Mode, DeepSeek, Qwen and Claude. The main engines are checked daily; some premium engines are checked on a slightly longer cadence to keep plans affordable, and their data feeds into all metrics the same way.

Each engine knows about you in two ways: its training memory (what it learned from the web up to its cutoff date, which you cannot change overnight) and live search (what it looks up on the internet while answering, where you can influence quickly with content and sources). The AI queries section shows you what it searches for live.

Global filters

At the top of the dashboard you can filter everything you see by country, city, date range and engine. Filters persist as you navigate between sections. The country filter only appears if you have prompts in more than one country.

Dashboard sections

Action plan

Every week the AI analyses your data and generates a plan of concrete, prioritised actions: what content to create, which sources to work on, what to fix in your presence. When you mark an action as done, your visibility at that moment is stored as a baseline, so over time you see the impact it had. There is also engine-specific advice.

Alerts and reports

On plans with alerts you can enable per-project email notifications (rises, drops and relevant news) and a periodic summary with your evolution. Configure them in Settings, with weekly, biweekly or monthly frequency.

Connect your AI, for any user

Every plan, including Free, includes a private URL that gives your AI assistant (ChatGPT, Claude, Gemini, Cursor…) access to your project data. Nothing to install.

The easy way

• In the dashboard, click "Data for your AI · copy URL" (at the top, in any section).
• The window that opens includes a ready-made message to copy: paste it as-is into your AI and it will read your data, summarise how you are doing and ask you what you want to know.
• The data URL returns an always-up-to-date report with your visibility, ranking, sources, AI queries and action plan. Any AI with browsing can read it.

Permissions

On Free the connector is fixed to read-only mode: your AI can view data but can never create, edit or delete anything. From Lite you can enable edit mode to create, edit, pause or delete prompts, add brand name variations and update plan actions. Edit operations have daily safety limits.

Connector technical reference

For developers and advanced users. All routes use the same bee_sk_… key and respect your plan's limits. Free exposes read operations only; write operations require Lite or above and edit mode enabled.

Data URL

https://beellm.ai/ai/<key> returns a self-contained markdown report of the project (visibility per engine, brand ranking, citability, prompts, AI queries, sources and action plan). Accepts ?project_id= for another project on the account. Not indexed, not cached.

MCP server

Remote MCP (Streamable HTTP) at https://beellm.ai/mcp/<key>, or at https://beellm.ai/mcp with an Authorization: Bearer <key> header. For Claude Code:

claude mcp add --transport http beellm https://beellm.ai/mcp/<key>

Read tools: get_projects, get_overview, get_prompts, get_ai_queries, get_sources, get_actions. With edits enabled, also: create_prompt, update_prompt, set_prompt_active, delete_prompt, complete_action, add_brand_variant.

REST API

The same data and actions over REST, with the key in Authorization: Bearer or X-Api-Key:

Limits and errors

• The active prompt limit is per account and per plan; when reached, creating or resuming returns 409. The AI must not pause or delete other prompts to free up room without the user's explicit confirmation.
• The assistant's create and delete operations have a daily quota; when exhausted they return 429, resetting at midnight (Madrid time).
• Missing or invalid key: 401. A write attempt from Free or with edits disabled returns 403. An account without access returns 402.
• The connector rate-limits requests per key to curb abuse. Free allows 1 active key and paid plans up to 3; manage and revoke them in Settings.

Plans and limits

Every new account starts free on the Free plan with no card, with daily ChatGPT tracking. Prices exclude VAT, no lock-in. Full details on the pricing page.

Privacy and security

• Your visibility data is yours: each account only sees and operates on its own projects, including through the connector and the API.
• Connector URLs are secrets: do not publish them, and regenerate them in Settings if you believe they were exposed.
• Deleting a prompt also deletes its entire history and recalculates metrics; it is irreversible.
• The dashboard session lasts 30 days; when it expires you are asked to log in again.

Frequently asked questions

AI describes me with old information, is it broken?

No: it is using its training memory. Engines only correct old data when they search live. Work on the sources they consult (Domains and URLs sections) and on content that answers their real searches (AI queries section).

Why does my visibility change from one day to the next?

AIs do not always answer the same way: the same question can produce different answers. That is why metrics are computed over multi-day windows and flagged as provisional with few measurements.

Can I monitor several countries or cities?

Yes: each prompt has its country (and optionally a city), and the dashboard filters by market. The Compare section puts two geographies head to head.

I ran out of prompt slots

The limit is on active prompts per account. You can pause the ones you do not need (they keep their history), delete the ones you no longer want, or upgrade your plan.

Where do I get help?

From Support in the dashboard or by writing to hola@beellm.ai. Product updates are published at /novedades.