Knowledge profile

Upwork's agency profile is where an agency tells clients what it does: name, overview, hourly rate, skills, awards and portfolio, editable only by owners or admins. A knowledge profile tells the same story to a different reader, the model that decides whether a job is worth a proposal and then writes it.

What the AI reads, and where:

• Core services, tools and skills, projects to avoid, languages and locations go to both the scorer and the letter writer. A job that matches a project to avoid is unsuitable however well it scores.
• Case studies go only to the writer, chosen per job; see case study match.

Why it is required. A model asked to judge a job against an empty profile does not become cautious; it invents an agency and judges against that. So a job is not scored, and a campaign cannot be started, until a profile with real content is assigned, and a profile that exists but holds almost nothing counts as none. The writer is told never to claim what the profile does not support, so every fact a client reads in a letter should trace back to it.

Writing one well is mostly about being specific. "Web development" gives the scorer nothing to separate a fit from a near miss; "Next.js storefronts on Shopify, no WordPress" does. Projects to avoid matter as much as services: they are the cheapest way to stop Connects going to jobs the agency would turn down. A first draft can be assembled from the agency's own Upwork profile and edited from there.

How Upsight uses it. Each campaign names one profile, the editor shows the exact text the AI receives, and a review lists what to fix first; the same tools are available through Upsight's MCP server, and the profile is what the suitability score is judged against.