We use Qwairy to track an influencer marketing platform in five European markets (Germany, UK, France, Spain, Italy). Each market is its own workspace, close to 1,000 prompts in total, running on ChatGPT, Gemini, Google AI Overview and Copilot. The monthly runs feed our SEO and GEO reporting to management.
What made the difference for us is first of all the granularity. Every answer is stored and can be broken down by engine, market, topic and tag. That matters because the engines behave very differently. Over the summer almost all of our movement came from one engine while the others stood still. A tool that only shows one blended visibility score would have hidden that.
The second point is that the data can be checked. Through the REST API and the MCP server we get the raw prompts and answers, so every figure that goes into a management deck is recalculated outside the UI with the exact date range and the same set of providers. I know few tools in this category that allow this. The MCP also lets us ask questions about our own tracking data from an AI assistant, which has replaced most of our CSV exports.
Third, the data model is flexible enough to build our own structure. We use topics for industry verticals and a second layer for task-type questions such as measurement, creator management or ROI. Topics then drive the GEO matrix, the competitor comparison and the content opportunities. The feature we use most in day-to-day work is the citation source analysis. It shows which domains the engines rely on in our category and where competitors get cited while we do not, and that goes straight into our content and outreach list.
A few things beyond the product. Credits are charged for stored answers rather than attempts, and the pricing simulator in the dashboard tells you what a setup will cost before you change it. Monthly billing without a lock-in was important for us. Boost, which collects several answers per prompt and run, is the right way to deal with the variance of AI responses. And the team replies within hours and ships regularly. Copilot and a Claude model were added during our first months with the tool. Review collected by and hosted on G2.com.
There is no cross-market view. A workspace is one domain in one country in one language. For a brand in five markets that means five dashboards and five shared links, and any aggregate across markets is something you have to build yourself. Shared links without a login only cover the overview and the brand perception, not a filtered prompt list.
The API and the MCP are read-only. Re-tagging a few hundred prompts or moving them to other topics is manual work in the UI. Two of the MCP tools returned internal errors when we tried them, and the monitoring field in the API does not reflect an active schedule.
You need to read the metric definitions before you report them. Brand mention visibility, for example, is calculated over the answers that mention any brand at all, not over all answers, and that gives a very different number. The dashboard does not make this obvious.
Premium engines are expensive in credits. The Anthropic models via API cost 12 to 20 credits per prompt against 1 for the core engines, so covering a full prompt set with them does not fit into any plan. The new Claude core model in beta helps, but beta models can be paused at any time.
The interface is dense for new users. It takes a few sessions to know where to look, and because the product moves fast, things occasionally change place. Review collected by and hosted on G2.com.