Buyers ask AI for a shortlist. Developers let AI pick. Be the pick.

The share of sandbox tasks in which a coding agent chose your package, API or MCP server over a rival’s, unprompted. We give Claude Code and Codex a realistic task such as “add payments” or “add auth”, run each three times, and record what they installed, what they fell back to after an error, and why. It sits beside share of voice in the same report.

ChatGPT, Gemini, Claude, Perplexity, Grok and Google AI Overviews, on the models customers use, with live web search on. For software and developer products we also run Claude Code and Codex on real tasks and record what they build with. Each assistant gets the same fixed panel of questions so results compare week to week, and a sample is checked against the consumer interface so you can see how far the two diverge.

Yes. On Enterprise plans questions are asked by digital twins of the shopper segments from your insight work, in each market you sell in. The same question gets a different answer for a budget-minded parent in the UK than for an ingredient-focused athlete in the US, and the report keeps them apart.

Tracking is a monthly plan per product, with Brand and Enterprise plans for teams tracking many products and markets, and a free tier for one product. Projects, studies and tests are paid from prepaid credit packs; the AI Recommendation Project starts at $12,000. Plans and prices are on the pricing page.

One scan, four outputs. Brand and insights teams get the tracker and the slide for leadership. Innovation teams get the unmet needs behind each answer, ranked as a brief, and a pre-launch read on whether the assistants will recommend a new product on day one. Product managers get the feature and claim gap list and the listing audit. The CEO gets one number, its trend, and what a share point is worth.

We size it with the same opportunity sizing model used in the sprint: your share of the answer against category demand and the conversion the assistants drive, so the slide carries a dollar figure, not a percentage.

Visibility tools count mentions for a generic logged-out user and suggest content. This project asks as your shoppers’ digital twins, reads the sources behind each recommendation, drafts claims, content and concepts, and tests them against the assistants, the twins and real respondents before you ship. It sits beside your brand tracker and covers what the tracker cannot see.

AI answers vary run to run, so watch and project scans ask each question three times on every assistant and report a range. One-off scans ask once and are directional.

Yes. The Simporter MCP server connects to any MCP client with a personal token, so your team can run a scan, pull this week’s share and read the why without leaving the tool they work in. Weekly alerts can post to a Teams or Slack channel.