Key concepts

These are the terms that come up again and again in the dashboard. They are worth reading once: almost every billing question and every metric that looks confusing is answered here.

The life of a page#

Every page in the system always moves through the same sequence of states:

Page lifecycle

A page only reaches AI agents once it is Optimized.

Scraping#

The crawler visits the page and stores its content as text. It is a mechanical operation: no language model is involved and it spends neither optimizations nor tokens. A scraped page is ready to be optimized, but it does not serve any content to agents yet.

Optimization#

A language model rewrites the scraped page: it produces clean Markdown, a summary for llms.txt, a meta description and JSON-LD structured data. Each optimization spends one unit of your plan's monthly package.

A common mix-up

Scraped is not the same as optimized. A site can show 289 total pages and only 45 optimized ones: the remaining 244 are scraped and waiting. AI agents only ever receive content from optimized pages.

The two currencies#

AgentFi splits usage into two separate counters, both visible at the bottom of the sidebar. Knowing which one each action spends is what keeps the invoice predictable.

	Optimizations	Tokens
What it measures	AI rewrites of a page.	Every piece of LLM work that is not a page rewrite.
What spends it	Optimize All, optimizing a single page, re-optimizing with feedback, and every page auto-optimization processes.	Each visibility query × each configured engine; each fire of a recurring schedule; reputation deep-dive follow-ups; and auto-optimization once the package is used up.
When it runs out	Manual optimization pauses. Auto-optimization keeps going, but starts spending tokens.	Further spending is billed as Extra usage at your plan's per-token rate. On the Free plan there is no extra usage: scheduled runs pause until the next period.

The per-plan breakdown is in Plans, tokens and optimizations.

Site metrics#

Total Pages

How many pages the crawler has discovered, including the ones that have not been optimized yet.

Scrape coverage

Share of total pages that have stored content. It should be close to 100% after a healthy crawl.

Optimization coverage

Share of total pages already rewritten by the model. This is the metric that matters: it measures how much of your site is genuinely visible to AI agents.

Avg Content Length

Characters per optimized page. Very low values usually point at pages with little useful text, or at a scraping problem.

JSON-LD Coverage

Share of pages with generated structured data. It helps models identify products, prices and availability.

Visibility queries#

A query is a natural-language question AgentFi sends to AI engines to check whether your site shows up in the answer. There are four kinds:

Kind	What it is	How it is created
Brand	Questions that name your brand explicitly. They measure whether the model knows who you are.	Generated automatically from your content.
Discovery	Category questions that do not name your brand. They measure whether the model recommends you to someone who does not know you yet.	Generated automatically from your content.
Custom	Questions you write by hand for a specific case.	The Add query button.
Reputation	Follow-up questions the reputation analysis generates while studying how your brand is perceived.	Reputation analysis.

Tip

High brand visibility (the model knows you when asked by name) combined with low discovery visibility is completely normal. Discovery is the one to push up: it is the one that brings new customers.

Test outcomes#

Found in answer

The engine cited or named your site. When a citation position is available it reads Found at #N; low numbers mean the engine placed your site higher in its list of sources.

Not mentioned

The engine answered, but neither cited nor described your site. This is the metric with the most room for improvement.

Engine error

The engine did not return a usable answer: provider outage, timeout or content filter. It counts neither as found nor as not mentioned.

Cited / Text mention

Cited: the engine listed your URL among its sources. Text mention: it named you in the body of the answer, without linking.

Infrastructure#

Proxy#

The layer that sits in front of your site and decides what to serve based on who is calling. An AI bot gets the optimized HTML; a person gets the original page from your site. It can run as a Cloudflare Worker (managed by AgentFi) or be self-hosted in your own Nginx or Apache.

Discovery files#

The files an AI crawler uses to work out what is on your site:

llms.txt

A Markdown index of your pages and offerings, designed specifically for language models.

llms-full.txt

The extended variant, carrying the full content instead of just the index.

robots.txt

Standard crawl rules. AgentFi can manage the file or leave yours untouched.

sitemap.xml

The URL listing. AgentFi detects it, indexes it and can serve a version of its own.

OKF bundle

Open Knowledge Format: an open standard from Google Cloud. A set of interlinked Markdown knowledge concepts, one per page, that agents can walk. It is published under /okf/.

Referral#

A visit from a real person who reaches your site from a conversation with an AI. It is detected through the Referer header or the /agentfi/pages/ redirect. Not to be confused with bot visits, which are measured separately in Bot traffic.