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

# The numbers explained

> What every number in the weekly report, Runs and Insights means, how it is counted, and what it does not tell you.

<Note>
  The weekly report and Runs are part of [Visibility](/visibility/overview), which some plans include.
</Note>

## How an answer is counted

Every number in the weekly report starts from the same thing: *one answer* — one reply an AI assistant gave to one of your customer questions during a test.

* *A citation* is a website the answer actually pointed to as its source. A site that only showed up in the assistant's search results, but was not used in the answer, does not count.
* *You* means your confirmed website, plus any other websites you added as yours in Search presence.
* Each site counts *once per answer*, however many of its pages were used.
* Questions where the customer *asks for you by name* are left out. Being found when someone names you proves little; the numbers show whether AI finds you when they don't.
* Answers where Aibuku could not see the sources are left out, rather than counted as "not cited".
* Without a confirmed website, there is nothing to count: confirm it in **Project settings** → **General**.

## The weekly report

Open **Insights** → **Visibility**. At the top, choose the period and the assistant.

| Period | What it compares |
| - | - |
| **1 day** | The latest day with answers, against the day before that had answers. The chart shows the last 14 days. |
| **7 days** | The last 7 days against the week before. The chart shows the last 8 weeks. |
| **30 days** | The last 30 days against the 30 before. The chart shows the last 6 months. |

**All assistants** combines Claude, ChatGPT and Gemini. Pick one to see only its answers.

### AI cited you in

The share of answers that cited your website, in large — for example *25%* — with *2 of 8 answers* underneath. The arrow shows the change in points against the previous period.

A period with no answers is never shown as 0%: you see **No answers recorded** instead. A real *0%* means there were answers, and none of them cited you. That is common for a new business.

### Your rank

Where you stand among *you and your competitors*, by how many answers cited each. Two sites with the same count share a place — *1, 2, 2, 4*. Other websites the assistants cited do not affect your rank.

### You vs competitors

One line per site: the share of answers that cited it, period by period. Your competitors are your competitor websites (up to five) — the ones you chose, or that Aibuku found when you set up. The three other sites assistants cited most over the whole chart are added too, marked **Other source**. Hover over a point to see how many answers it is based on. When you fixed something on your website and Aibuku confirmed the fix, a marker shows the date.

### What to do next

Up to three customer questions where another site is cited more than you, biggest gap first. Each shows how often you and the leading site were cited, and whether your website has a page that answers the question: **Your page answers it**, **Your page partly answers**, **No page answers it**, **Not measured** or **Not checked yet**.

### By assistant and By customer

* **By assistant** — your share for each assistant, with how many answers it is based on.
* **By customer** — your share for each of your customers, weakest first, and which site AI cites instead.

### Where AI misses you

A grid: your customers down the side, how plainly they ask across the top. Each square shows how often AI cited you. A square needs at least 5 answers before it gets a colour — fewer answers are too few to read. On **1 day**, most squares have too few answers; choose 7 or 30 days.

### What changed

Changes on your website that Aibuku confirmed from your saved [Site health](/visibility/site-health) checks: what was fixed, and what got worse.

### Sources AI trusts

The ten websites assistants cited most in the period — you, your competitors, and any other site. Being mentioned on these sites is one of the best ways to be recommended.

## Runs

Each row in **Runs** is one AI app's result for one part of a run.

### The verdict: how far it got

The verdict names *the furthest step any chat in the run got to*, and in how many of the chats: *Right tool · 2 of 3 chats*.

**Finding your website** — Claude, ChatGPT and Gemini are asked:

<Steps>
  <Step title="Found">The AI found your website in a web search.</Step>
  <Step title="Read">The AI read your website.</Step>
  <Step title="Pointed">Your website pointed the AI to your connector.</Step>
</Steps>

**Using your connector** (or **Using your MCP server**) — Claude and ChatGPT are asked:

<Steps>
  <Step title="Connector shown">Your connector was shown to the AI.</Step>
  <Step title="Connector chosen">The AI chose your connector.</Step>
  <Step title="Right tool">The AI picked the right one of your services.</Step>
  <Step title="Right parameters">The AI passed the right details to it — the last step.</Step>
</Steps>

If no chat passed any step, the verdict names the first step that could be watched, as a miss: **Not found**, **Not read** or **Not pointed** for your website; **Not shown**, **Not chosen**, **Not the right tool** or **Not the right parameters** for your connector. Most often it is the first one — some apps don't let us see the early steps, so their verdict starts later.

Two more verdicts:

* **Not scored yet** — the chats have not been checked yet.
* **Not measured** — this app could not be watched on this part.

### The other columns

* **Chats** — how many conversations were tested. Not the number of prompts: one conversation can hold several.
* **Prompt level** — how plainly customers had to ask. See [Writing good customer questions](/visibility/writing-good-questions).
* **Status** — whether the run is still going or **Done**.
* **Pinned** — pin a run to keep all its evidence, for every app, for longer than usual.
* **Lost details** — opens **Details the AI most often loses in your chats**: details a customer gave (a date, a number of people) that the AI dropped or changed along the way, over the last 30 days.

## Insights: real use

These numbers come from real assistants using your business, not from tests. Open **Insights** → **Connector**.

| Number | What it counts |
| - | - |
| **Total calls** | Each time an assistant used — or asked for — one of your services in the period. The change compares with the period just before, of the same length. |
| **Sessions** | Each time an assistant connected to your business. It can be higher than calls: an assistant can connect without using anything. |
| **Active users** | Different users the assistants reported. Not an exact count of real people. |
| **Success rate** | The share of calls that worked. A request for something you don't offer is not counted as a failure — you find those under **Asked for, not offered yet**. |

## What the numbers don't tell you

* *Tests are not customers.* The weekly report and Runs come from test questions Aibuku asks. They show how assistants treat your business, not how many people asked.
* *Small numbers swing.* With a handful of answers, one answer moves the share a lot. Prefer 7 or 30 days, and look at the trend.
* *A citation is not a sale.* It means the assistant pointed to you. What happens next is in [Insights](/your-business/insights) and your [Inbox](/your-business/inbox).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.