About QueryHalo

About QueryHalo: what we measure, how we measure and what we do not claim

QueryHalo answers a single customer question: do ChatGPT, Claude and Gemini recommend my company, who is recommended instead, and what should I change first? This page describes the method precisely enough for you to interpret the result and, in principle, reproduce the measurement yourself.

QueryHalo is a one-time AI visibility audit: 6 neutral buying questions go to 3 AI models (GPT, Claude, Gemini) and all 18 answers are documented with the competitors they name. The website check is free; the full report costs €29 once (3 audits €59, 10 audits €149), usually takes 1–3 minutes and arrives as PDF and JSON.

What QueryHalo is

QueryHalo is a one-time AI visibility audit for companies and agencies. It consists of two steps: a free website check that needs only your homepage address, and a paid report that delivers 18 documented model answers, named competitors and a prioritized action plan. One audit costs €29 once, three audits €59, ten audits €149. There is no subscription, no account and no automatic renewal.

The primary language is German; all core pages also exist in English. The first market is the German-speaking region.

Method: questions without brand names

From your inputs on offer, audience and market, QueryHalo builds six neutral buying and recommendation questions – for example “Which providers for invoicing software are best suited to Austrian online retailers with 2–20 employees?”. Your company name appears in none of these questions. Only then is a mention an observed recommendation and not a confirmation of what we suggested.

Method: identical questions to three models

The same six questions go separately to three model families – currently GPT 5.6 Luna (OpenAI), Claude Sonnet 5 (Anthropic), Gemini 3.5 Flash Lite (Google) – via Vercel AI Gateway, without live web search. For each answer we record whether the company name or domain is mentioned, at which position the brand appears in the model's brand list and which other brands are named. All 18 answers are in the report in full; nothing is shortened or summarised before you see it.

The mentions produce the AI visibility value (share of the 18 answers with a mention). A model then generates a summary with strengths, gaps and four to six prioritized actions from the answers and the website check; this model too is instructed not to claim causality or promise rankings. If this synthesis fails, QueryHalo produces a rule-based plan from the measurements.

Method: server-HTML analysis of the homepage

The website check reads the server HTML of your public homepage – at most 700 kB, up to three redirects, private and local addresses are rejected – and scores eight criteria totalling 100 points: page title (14), meta description (12), primary heading (12), structured entity data in JSON-LD (20), brand name in text (12), citable facts (20), robots.txt (8) and llms.txt (2). This step calls no AI model and is therefore free. The rules are fully documented in the technical checklist.

The overall score weights 65% AI visibility and 35% website check. Both values are reported separately.

What QueryHalo deliberately does not claim

  • No rankings: there is no fixed position in an AI answer; answers vary between calls. The report is a snapshot of model knowledge on the creation day.
  • No live web search: we measure existing model knowledge, not search citations, AI Overviews or real prompt volumes.
  • No verdict on your quality: not being named is the normal case for smaller providers.
  • No guarantee: neither of future mentions nor of traffic or revenue. The actions are reasoned suggestions, not promises.
  • No full website crawl: the homepage's server HTML is checked. Heavily client-rendered pages can be underestimated.
  • No monitoring, no white-label, no team accounts: for trends you repeat the audit deliberately; agencies pass on the PDF.

How we check ourselves

We apply the same website check to queryhalo.app that customers see, on every public page. That is why a short definition paragraph with the key numbers sits under every main heading, every page carries structured data for organization, product and page, and a machine-readable description lives at /llms.txt. Anyone who enters our homepage into the free check sees the result themselves.

Operator

QueryHalo is operated by OptiRisk Consulting e.U., owner Thomas Michalik, Döblerhofstraße 10/167, 1030 Vienna, Austria, commercial register number FN 626043b (Commercial Court of Vienna). Contact for questions, invoices and complaints: info@optirisk.at. Further details are in the imprint; data processing and service providers are described in the privacy policy (German with an English summary); service, prices and withdrawal are governed by the terms (German).

Frequently asked questions

Is my brand handed to the models as the expected answer?+

No. The six questions contain neither company name nor domain. Your name is only used during evaluation to detect mentions in the answers.

Why exactly these three models?+

GPT, Claude and Gemini cover the three largest model families end customers meet in assistants. The specific versions are listed in the report and in llms-full.txt; we update them when providers retire models.

Can I repeat the measurement myself?+

Yes. The guide “Check your ChatGPT visibility yourself” describes the five steps; QueryHalo automates them and delivers the log as PDF.