# KPIs and Realistic Expectations — AI Visibility and LLM Brand Discovery

Source: https://www.geekswithgeeks.com/en/ai-visibility/op-reporting

> Report a small honest scorecard and avoid promising what nobody can control.

## Honest numbers, not promises

Report: **mention rate** (with range) and **share of voice** by intent and language; **citation rate** and top cited domains; **accuracy** (percentage of key facts stated correctly); **sentiment/themes**; **AI-referred visits and conversions** (lower bound) and **crawler health** (200-rate on key pages). Link to business results where you can, such as demo requests from AI-referred visits, but be careful with causation. Be transparent that results vary by assistant and over time, assistants change without notice, and no one, including agencies and tools, can guarantee a placement.

## A scorecard

Numbers are illustrative. Include ranges and the date so readers see the uncertainty.

```text
Metric (Oct 2026, 60 prompts x 3 runs x 3 assistants)   Value          Trend
Mention rate (discovery prompts)                         34% (30-38%)   +6 pts
Mention rate (Hindi prompts)                             12% (9-16%)    +2 pts
Share of voice vs 5 named competitors                    18%            +3 pts
Our URL cited                                            9% of answers  +4 pts
Key facts stated correctly                               82%            +10 pts
AI-referred visits (lower bound)                         410 / month    +35%
GPTBot/ClaudeBot 200-rate on pricing + docs              99.2%          flat
```

## Show wins and misses

Include prompts where you are absent and wrong facts still present. Leaders trust a programme that shows its gaps and what it will do about them.

**Quiz:** Why can no one honestly guarantee placement in AI answers?

- [x] Assistants are non-deterministic, change often, and providers do not publish ranking rules
- [ ] Guarantees are illegal in all countries
- [ ] Because websites are not read
- [ ] Because Python forbids it

*Answer:* Assistants are non-deterministic, change often, and providers do not publish ranking rules. Outcomes depend on systems outside your control that vary run to run and update without notice.