How to Build a Promptwatch Knowledge Base for Content Agents

An agent that publishes on a schedule will publish whatever it believes. If all it knows is the open web, it writes the same competent, generic article every other tool in your category writes. AI models have no reason to cite that article, because they already know everything in it. The pages that earn citations say something specific: your numbers, your opinions, the claims your team has signed off on.

The Promptwatch Knowledge Base is where that specificity comes from. It is a private library the research agent searches before drafting, so content is grounded in your facts. It applies to manual generation, to Content Refresh rewrites, and to every automated piece from the Content Agent. Source: Promptwatch's Knowledge Base documentation, read October 15, 2026.

Plan check

Promptwatch's Knowledge Base documentation says the feature starts on Professional ($245/mo) with a per-project item cap, and that Explore, Essential, and agency plans don't include it. The public pricing summary on llms.txt lists a knowledge base on Essential ($95/mo) as well. The two sources disagree, so confirm on the live pricing page before you plan an Essential rollout around it.

What to feed it

Prioritize material only your team has. Promptwatch groups it into four kinds:

• Proprietary research and data. Internal studies, benchmarks, customer surveys, original numbers.
• Expert voice. Founder and specialist perspectives, point-of-view essays, strong takes.
• Brand and product truth. Approved messaging, product specs, pricing, positioning, signed-off claims.
• Customer evidence. Case studies, testimonials, interview transcripts, real examples.

Skip what the model already knows. Encyclopedic content, competitor marketing pages, and generic industry explainers add noise without giving the agent anything to say that a competitor's agent can't.

Our addition to that list: an approved claims sheet. One plain-text source that says what you can and can't claim, with numbers and dates. When an agent drafts at volume, the most expensive failure is a confident claim legal never approved. A claims sheet in the Knowledge Base is cheaper than catching that in review every week.

Add sources

Open Content in the left rail, then Knowledge Base, and click Upload data. Three source types:

• File. PDFs, Word documents, spreadsheets, presentations, CSV, markdown, plain text, and images. Up to 10 files at a time, 25 MB each. Drag and drop works.
• URL. A web page or a public document such as a hosted PDF. The page is fetched and its content extracted.
• Text. A title plus pasted content. The fastest path for meeting notes, positioning statements, or that claims sheet.

Each source is processed into a searchable entry with a generated title and summary. Only items with status Ready are searchable. If processing fails, the row shows why and offers a retry.

Check the agent can use it

The research agent has a Knowledge Base search tool among its enabled tools. Tool defaults live in AI Settings under Brand. During research it pulls relevant passages from Ready sources into the writing brief the writer drafts from.

Before you switch on Content Agent automation, Promptwatch recommends at least one Ready item as a readiness check. We'd go further and generate one draft by hand on a topic where your Knowledge Base has a clear, checkable fact. If the draft uses it, the wiring works. If it doesn't, fix that before the agent runs unattended.

Keep it current

The agent can only repeat what's in the library, so a stale library produces confidently outdated drafts. Three rules from the documentation, which we'd adopt as written:

1. Update on change, not on schedule. New pricing, a repositioning, or a new case study goes in the week it ships.
2. Delete what no longer holds. Deletion removes the source from the agent's reach immediately. Original files aren't kept after processing, so keep your own copy.
3. Review after odd drafts. When a draft states something wrong, check whether the library taught it that, then fix the source rather than only the draft.

The third rule is the one that makes an agentic content program sustainable. Fixing drafts one at a time is editing. Fixing the source is changing what the agent knows.

Review: Promptwatch. Product: promptwatch.com. Full directory.

What to do this week

1. Write a one-page approved claims sheet and add it as a Text source.
2. Upload your two strongest case studies and your current pricing page URL.
3. Confirm all three show Ready.
4. Generate one draft by hand and check it used a Knowledge Base fact.
5. Only then turn on the Content Agent schedule.