The IndexHot GEO Measurement Standard v1.0

1. Why a Standard Matters

Most GEO tools return a single “visibility score” with no visible prompts, answers or arithmetic. If you cannot reproduce a number, you cannot trust it, diagnose it, or optimize against it. A shared standard makes AI-search visibility measurable the same way click-through rate and domain authority became measurable for SEO.

Black-box scores cannot be acted on

• Today’s tools typically expose one aggregate score but hide the underlying question set, the raw answers and the weighting, so a change in score cannot be traced to a cause.
• Without fixed inputs, two vendors testing the same brand on the same day can return different numbers, and there is no way to tell who is right.
• Teams end up optimizing for the vendor’s hidden formula rather than for real visibility, which is exactly the problem the original GEO research (arXiv:2311.09735) warned against by publishing an open benchmark.

The industry is converging on governed, reproducible measurement

The AMEC GEO Principles and Practitioner’s Guide to GEO Measurement (2026) explicitly call for measurement that avoids opaque scores and vanity rankings. A core requirement is a governed, reproducible question library: a fixed, versioned set of prompts run under logged conditions, rather than ad-hoc questions chosen to flatter a result.

Citation count alone is misleading

Peer-reviewed evidence shows that “being cited” and “being absorbed into the answer” are different outcomes. The study From Citation Selection to Citation Absorption (arXiv:2604.25707, 2026) analysed 602 controlled prompts, 21,143 search-layer citations and 18,151 fetched pages across ChatGPT, Google AI Overviews/Gemini and Perplexity.

• Perplexity cites the most sources per prompt on average, yet shows the shallowest average absorption per source — breadth without depth.
• ChatGPT cites fewer sources, but each cited page shows substantially higher average influence on the final answer.
• High-influence pages are longer, modular, semantically aligned with the question, and rich in extractable evidence such as definitions, statistics, comparisons and procedural steps.

The implication is direct: counting citations rewards the wrong behavior on Perplexity and understates real influence on ChatGPT. A credible standard must measure mention, citation and absorption separately on every platform.

2. The Measurement Framework

The framework has three layers: a governed question library that fixes the inputs, citation metrics that define what is counted, and cross-platform scoring that combines platforms without hiding the components. Every layer produces a number you can recompute from the raw answer logs.

2.1 Governed Question Library

The library is the controlled variable of the whole measurement. It is versioned, frozen for each measurement cycle, and split into two prompt types that answer different business questions.

• Branded prompts contain the brand or exact product name (e.g., “Is IndexHot accurate?”). They test defensive coverage: does the AI describe you correctly when asked directly?
• Non-branded prompts express category intent without naming you (e.g., “best AI search visibility tool”). They test competitive discovery: are you surfaced when the buyer does not already know you?

Library construction rules:

• Seed prompts from real query logs, support tickets and keyword research, then stratify by intent: informational, commercial, comparative and navigational.
• Freeze and version the set for a full cycle; never edit a prompt to improve a result mid-cycle.
• Log the model version, platform, date, region, language and settings (web search on, no logged-in personalization) for every run.
• Run each prompt at least three times to capture non-determinism, and report the median plus the spread.

2.2 Citation Metrics: Mention, Citation, Absorption

Each answer is coded against three nested, binary-per-answer metrics. They are nested in the sense that absorption implies a mention, but a mention does not imply a citation and a citation does not imply absorption.

Absorption is additionally given a depth score of 0–2: 0 none, 1 a generic or paraphrased idea, 2 a distinctive claim or figure traceable to the source. The first brand mention’s rank position (1 = first named) is recorded separately.

Coding edge cases are decided by fixed rules so different reviewers agree:

• A misspelled or partial name counts as a mention only if a unique identifier (product name, domain or trademark) makes the reference unambiguous.
• A link through an aggregator or review site counts as a citation only when it points to a page you own or an official profile you control; third-party listings are recorded separately as “earned” mentions.
• When you and a competitor are named in the same answer, both receive a mention, but rank and absorption decide who the answer actually favors.
• Any answer a second reviewer would code differently is flagged and arbitrated; the target inter-reviewer agreement is at least 90%.

2.3 Cross-Platform Scoring

Rates are first computed per platform, then combined with explicit, adjustable weights. Per-platform rates are:

• Mention Rate (MR) = answers mentioning the brand ÷ total prompts.
• Citation Rate (CR) = answers citing the brand domain ÷ total prompts.
• Absorption Rate (AR) = answers with depth ≥ 1 ÷ total prompts.

The per-platform score weights absorption most heavily, reflecting the evidence that influence matters more than raw citation count:

Default weights reflect audience reach and citation influence, but teams should reweight to match their own traffic mix. Weights must always be published alongside the score.

3. How to Reproduce

You can reproduce a full baseline in an afternoon with a spreadsheet and three browser tabs. The goal is not a one-off number but a repeatable cycle you can run monthly to track trends.

1. 1

   Build a 50-question library.

   Create 10 branded prompts and 40 non-branded prompts, stratified across informational, commercial, comparative and navigational intent. Freeze the list and tag every prompt with its type and intent.

2. 2

   Run each prompt on all three platforms.

   Execute the same 50 prompts separately on ChatGPT, Perplexity and Gemini (with AI Overviews where available), web search enabled and personalization off. Run each prompt three times and keep the median.

3. 3

   Record mentions, citation sources and rank.

   For every answer log: mention (yes/no), citation (yes/no plus the source URL), absorption depth (0–2), the rank of the first brand mention, and which competitor was cited instead when you were absent.

4. 4

   Calculate the IndexHot Score.

   Compute MR, CR and AR per platform, apply S_platform = 100 × (0.30·MR + 0.30·CR + 0.40·AR), then combine platforms with the published 0.40 / 0.30 / 0.30 weights. Save the raw logs so the score is fully auditable.

A complete run should leave you with three artifacts: the frozen question library, a raw answer log with one row per prompt × platform × repetition, and a one-page scorecard. Use the scorecard’s bands only as a starting point for conversation, never as a verdict:

• 0–25 — Invisible: absent from most non-branded answers; focus on indexing and entity clarity first.
• 26–55 — Mentioned: named often but rarely cited or absorbed; invest in evidence-rich, quotable content.
• 56–80 — Cited: a regular source with competitive presence; optimize absorption depth and rank.
• 81–100 — Recommended: frequently the first, most deeply absorbed answer; defend and broaden coverage.

4. Frequently Asked Questions

Short answers to the questions teams ask most often. The same answers are published as structured FAQ data for machine readability.

Cite this standard

If you reference the IndexHot GEO Measurement Standard in research or reporting, please use one of the following formats.

APA

IndexHot. (2026). The IndexHot GEO Measurement Standard v1.0. https://indexhot.com/methodology/

BibTeX

@misc{indexhot2026standard,
  title={The IndexHot GEO Measurement Standard v1.0},
  author={IndexHot},
  year={2026},
  url={https://indexhot.com/methodology/}
}

Version history

Related resources