Skip to main content
Plans: All plans, limits vary · In the app: Sidebar → Analytics → Sentiment → FactCheck
FactCheck compares what AI engines say about your brand with a list of facts you approve. When an answer says your Pro plan costs $15 and your fact says $12, the claim is marked Contradicted, with the prompts, engines and pages that carry it. In the sidebar, open Sentiment under Analytics, then the FactCheck tab.

Set it up

1

Add your facts

Click Edit facts, then Add facts, and write one fact per line.
2

Choose the prompts to check

Click Choose prompts, tick the prompts whose answers should be checked, then Fact-check these prompts. Only answers to these prompts are read.
3

Read the results

Claims are checked within minutes. The headline says where you stand, for example “3 claims contradict your facts”.

Facts

A fact is one sentence about your brand that is either true or false: a price and what it is per, a plan limit, a certification, an integration, a market, a founding year. A paragraph, a slogan or an opinion cannot be checked.
Write denials explicitly, and put each price on its own line. Each fact gets a type: Price, Plan, Free tier, Feature, Integration, Location, Founded, Headcount, Customers, Availability or Other. A claim is compared only with facts of its type; price, plan and free tier facts are read together.

The Facts sheet

Edit facts opens the Facts sheet beside the results. Before you add, each line is compared with your list. A repeat (“Already in your facts.”) blocks the add; two wordings of the same value count as one fact. A conflict (“Disagrees with …”) changes the button to Keep both. Brand context › Facts holds the full editor. Suggest facts drafts suggestions from your approved Brand Records and product catalog, which you approve or reject. Add fact there also takes a Plan, Location, Valid from, Valid until and Source link. Only facts that are approved, switched on and inside their dates are used.

How a claim is checked

Quotable pulls each factual statement about your brand out of the answers to the chosen prompts and gives it a type. Then:
  1. Find the facts that apply. Same type, in force on the answer’s date, valid in the answer’s location, and about the plan or product the claim names. A fact about the Pro plan does not judge a claim about the Team plan.
  2. Settle what rules can. Prices, years, counts, and whether a listed feature or integration is present are compared directly. Prices are compared only like for like: same currency, billing period and plan. Currencies are never converted. A “plans start at” claim is compared with your lowest plan price.
  3. Ask a model for the rest. A model reads the claim next to the facts that apply. If it cannot decide, the claim stays Being checked and is retried.
  4. No fact, no verdict. A claim no fact covers is Not covered, never Contradicted.

Worked example

Your fact: “The Pro plan costs $12 per user per month.”

When checks run

  • New answers are checked when tracking saves them. Claims the rules settle get a verdict at once; the rest are checked in the background, usually within minutes.
  • When you change a fact or the chosen prompts, the affected claims are checked again right away, and the page refreshes while verdicts land.
  • Answers from the last 180 days are checked.

Read the results

Filter by date range (the last 30 days by default) and model. Four counts sit on top: Claims (supported plus contradicted, a repeated claim counted once), Answers read, Contradicted and Supported.
  • Contradicted lists the newest 100 contradicted claims with Models, Claim, Your fact says, Answers and Last seen.
  • All claims shows them By fact (what each fact settles) or By group (every wording of one claim together, with its Contradicted share). Filter by All verdicts, Contradicted or Supported.
  • Export downloads the supported and contradicted claims as a CSV.

The claim panel

Click a claim to open it. The panel has its own link. AI repeats the pages it reads, so fix the claim on those pages: your own first, then third-party ones. Changing facts, prompts, verdicts and links needs edit access to the brand. See Team and roles.

Plan limits

Your plan sets two limits per brand: how many facts can be active, and how many prompts FactCheck reads. See FactCheck for each plan. Active facts are approved and switched on; suggested, rejected and archived facts never count. The fact limit applies when you add, approve or switch on a fact. The prompt limit applies when you add prompts; a selection above a lowered limit stays, but you cannot add more until it is under it. Pitch brands do not include FactCheck. See Plans and limits.

With Quotable Agent

Two buttons open Quotable Agent beside the page:
  • Build facts with AI reads your brand profile, current facts and your own site (pricing, plans, FAQ, security, integrations, about). It drafts facts with the page each came from, asks you to confirm every number, adds the ones you agree to and proposes prompts to check.
  • Work with AI on a claim reads the claim, its sources and your facts, and checks your own site first. If your site agrees with the claim, it offers to reword or switch off your fact. If it agrees with your fact, it names the pages carrying the wrong statement. It can also set the verdict or add a correction link.
Every change appears as an approval card listing each fact or prompt in full. Nothing changes until you approve it.

API

Paths are under /projects/{project_id} (a project is a brand). See the API reference for authentication and errors.

Next steps

Sentiment

How AI engines describe your brand, aspect by aspect.

Context Manager

The Brand Records facts are suggested from.

Citations

Every page AI engines cite for your prompts.
Last modified on November 18, 2026