The Count plugin for Claude flags wrong answers, dodgy assumptions, and plain hallucinations, so you know which answers you can trust.
Claude answers a question about a drop in revenue. Count highlights one claim, that a single account cancelled every product it had, and flags that the answer never checked it. Added to the chat, the finding becomes the next question for Claude.
Trusted by brands worldwide
Who it’s for
Claude can query millions of rows for you. Count checks the joins, the math, and the jump from what the data shows to what the answer concludes.
Claude will search and read for as long as you let it, but it won’t tell you which sources it made up. In 2025, Deloitte reports for the Australian and Canadian governments cited research that doesn’t exist.
Drafts come fast, and so does slop. Count catches the stitched quote, the stat with no source, and the claim your readers will check before you publish it.
Some are obvious. Most hide in a sentence that sounds finished. Pick one to see how Count flags it.
Step 1
Open Plugins in Claude Desktop’s settings and add Count.
Step 2
Ask the question you were already going to ask. Count runs alongside.
Step 3
Open a finding to dismiss it, or add it to the chat so Claude can fix it.
Yes. It follows the same security procedures as the Count platform. See the Trust Center.
It reads the question, the context you gave Claude, and the answer Claude gave back. Then it flags the claims that do not hold.
Yes. It is built to be more reliable than checking the answer by eye. It can still miss something.
Free to use. Pro is $49, and Enterprise is on the pricing page.
It is strongest on hallucinations. It is not perfect. It is better than trusting the answer as written, and it also flags assumptions, weak reasoning, and presentation issues.
No.
Add to Claude
