How we measure

What gets counted

• Questions are generic: trade plus suburb. They do not include the business name.
• If you edit a question so it names the business, that answer is stored but not counted.
• Each question is asked once. The first two questions are asked a second time.
• A business is “named” when the answer text, a cited link, or the phone number matches the name, website, or phone on file. Close spellings count. A different business with the same trade does not.

Engines

• ChatGPT: on · openai/gpt-5-mini
• Gemini: on · google/gemini-2.5-flash-lite
• Perplexity: on · perplexity/sonar
• Claude: on · anthropic/claude-haiku-4.5

Web search is on. ChatGPT is pinned to low reasoning effort so it does not run a long chain of searches. Claude is limited to two searches per answer. Perplexity often returns no source links through the gateway. When that happens, we still show the answer and say the links were missing.

The same trade and suburb question is shared across businesses in the same week. That keeps the cost down and means those businesses are scored against the same answers.

Limits

• Location hint: Australia, plus the suburb and state you entered. This is not a live map pin.
• Spend caps: US$8 per run and US$40 per week. A free check also stops at US$1. If the next call could pass a cap, we stop and say so.
• The free check uses fewer questions and engines than the weekly run. It is a sample, not the full report.
• We store the answer text so you can read what the engine said.