Skip to content
askround

What askround is

People ask assistants which tool to use, and the product’s makers never see the answer. askround asks OpenAI, Gemini, Anthropic and Perplexity the same question through their APIs, with web search on, several times each, and reads what comes back: whether a product is named and at which position, which rivals sit around it, what the answers claim about it and which of its pages they cite.

It is a command line and an MCP server. The second is the point: a coding agent working on a product can ask the assistants itself, instead of asking its user to type a question into ChatGPT and paste the answer back.

What it measures: the APIs, not the apps

Section titled “What it measures: the APIs, not the apps”

An API answer with web search is close to what a chat app says, but it is not the same page. The apps add memory, personalisation and their own choice of model. askround records what produced each answer (the model asked for and the one that answered, the search tool, the settings, the time), and every number it reports is about the API. Say so when you report its numbers.

Answers also vary from one run to the next, so askround never reports a single answer as the truth. A product is “named in 2/3 runs, positions #1, #3, -”.

The laws

Section titled “The laws”

Every design question in askround is settled by one of these:

  1. Context is cost. An agent pays for every token it reads, so a round comes back as a brief of a few hundred tokens, and the text of an answer only when it is asked for.
  2. Refuse before paying. Every check that needs no call runs before the first one: a question that names its product, personal data, a missing key, a round above the cap.
  3. Ask instead of read. The agent asks the assistants itself rather than asking a person to paste what an assistant said.
  4. Facts trace to a passage. Every claim, position and rival carries the answer it came from and the character range of its passage, so it can be checked and cited.
  5. The person sets the money. The day and month budget lives in a file no tool writes; an agent cannot raise it.
  6. It measures the APIs, not the apps. What produced each answer is recorded, and every report is about the API.

Where to go next

Section titled “Where to go next”