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

# AI Visibility Truth: How We Measure

> How PokPok's AI Visibility Truth at geo.pokpok.ai measures a brand in AI answers, disclosed item by item against the IAB's Provider Disclosure Framework (Measuring Visibility in the AI Era, August 2026).

AI Visibility Truth asks AI engines your buyers' questions **every day** and records **whether each answer names your brand**. This page answers each of the IAB's required disclosures, so you can judge the numbers for yourself.

01 // What is being measured

## Which engines, which questions, from where

| IAB disclosure | What PokPok does |
| - | - |
| **Platform coverage** | **ChatGPT**, **Gemini**, **Google AI Mode**, **Google AI Overview**, **Grok** and **Claude**, each as a buyer gets it in the app, logged out. Other engines on request. The model version is the one the app uses on the day; the apps do not report it, so it is not recorded. |
| **Prompt library construction** | From **50 questions per brand**, asked once a day; larger sets on request, with no upper limit. A new brand starts from the buyer questions PokPok's report engine writes from its website: six groups of buyer moments plus two sharp questions. You can replace or add questions. Every question counts equally; none is weighted. Each brand is asked in one market (its country); questions are not varied by demographic or cultural framing. |
| **Query sourcing** | Provider-defined (synthetic): written from your own website by PokPok's report engine, plus any questions you write. Google search demand per question can be measured alongside, but it does not choose the questions. |
| **Source type classification** | The pages each answer cites are stored and listed by website. They are not sorted into brand-owned, retailer, editorial or institutional types, and they are not weighted. |

02 // How it is being measured

## Asked like a buyer, counted by plain rules

| IAB disclosure | What PokPok does |
| - | - |
| **Data collection architecture** | **Active query simulation**: PokPok sends the questions to the engines and captures the answers. No panel, no platform-supplied data, no mix of the two. |
| **Data collection methodology** | Automated collection of the answers the consumer apps give, through a third-party collection service, with **web search on** (as the apps answer buyers). Each question is asked **once a day per engine**, from the brand's market. A failed ask is retried at least 6 hours later, at most twice more that day. |
| **Panel validity** | Not applicable: no panel is used. |
| **Data validation** | No external validation (no platform-native or first-party analytics). Entity rules: a brand or competitor counts as named only where its name stands as a **whole word** (not "Intel" inside "intelligence"); a brand's name comes from the set-up research, never from its web address. Each brand's first measurement is its baseline. No outliers are removed. |
| **Mention detection, attribution and sentiment** | Code checks each stored answer for the brand's names and its tracked competitors' names, by the whole-word rule. No model reads or labels the answers. **Sentiment and position are not measured.** |
| **Factual accuracy classification** | **Not measured.** Answers are shown as the engines gave them; PokPok does not judge them as accurate, inaccurate or hallucinated. |

> [!NOTE]
>
> One run per question, per engine, per day. Engine answers vary by nature, so AI Visibility Truth shows each day's answer as given and draws the line over days; it does not report ranges for a single day.

03 // How the data is maintained

## History, and what happens when rules change

| IAB disclosure | What PokPok does |
| - | - |
| **Historical data and versioning** | History starts at each brand's first measurement and is kept. Engine model updates are not recorded, because the apps do not report a version. When PokPok changes its counting rule, or a brand's or competitor's names change, **every stored answer of that brand is counted again** with the new rule, so the whole history stays comparable; each re-count is recorded and can be undone. Demo brands carry generated sample history, shown as sample. |

04 // The standard

## Read the framework this page follows

IAB, Measuring Visibility in the AI Era (August 2026): Provider Disclosure Framework, pages 27–29.

[IAB framework (PDF) →](https://www.iab.com/wp-content/uploads/2026/07/IAB_Measuring_Visibility_in_the_AI_Era_August_2026-1.pdf)


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