Skip to main content

When to use this

Use this playbook when AI answers state a wrong price, a plan you retired, a feature you never had, or criticism that is no longer true. Engines repeat the pages they read. The answer is the symptom; the page behind it is what you fix. Some of those pages are yours and you can change them today. Others belong to someone else, and you have to reach them. You’ll need: edit access to the brand to add facts and use Quotable Agent, and tracking runs whose answers mention you. FactCheck reads only the prompts you choose for it, up to your plan’s limit.

Steps

1

Write down what is true

Open Settings and, under the brand, open Facts. These are the facts FactCheck treats as true about your brand. Only approved, active facts within their dates are used.
  • Add fact adds one fact. Facts you add are approved straight away.
  • Import takes one fact per line and adds them as suggestions to review.
  • Suggest facts drafts facts from your brand records for you to approve or reject.
Write facts that can be checked: prices, plans, limits, features, certifications and where you operate. Put one price on each line so each one can be checked on its own.
2

Find the wrong and outdated claims

In the Analytics section, open Sentiment.
  • Brand perception and Strengths and weaknesses list the aspects AI praises and criticizes most. Select one to read the answers behind it, with positive, neutral and negative passages outlined.
  • Attributes maps what engines say about you, aspect by aspect.
If your workspace shows the FactCheck tab beside Sentiment, open it. Click Choose prompts to pick the prompts whose answers are checked. Every claim those answers make about your brand is compared with your facts. Contradicted lists what AI says that one of your facts says is wrong, newest first. All claims groups every claim By fact or By group. See FactCheck.
3

Find the source behind each claim

Open a claim in FactCheck. Sources splits the pages three ways:Without FactCheck, open the answer from Sentiment or from the prompt’s Responses tab on Prompts, and read the pages it cites. Note the engine and the prompt: a claim on one engine may come from one page, and a claim on every engine from a page they all read.
4

Fix your own pages first

When the source is your site, update the passage the engine relied on. Old pricing pages, changelogs, help articles and comparison pages are the usual cause. Say the current fact plainly, near the top, in a sentence that stands on its own.Correct your brand records in Settings → Context as well. Quotable Agent and routines write from them, and the llms.txt button in the Context Manager builds a brand facts file from approved records for you to publish at the root of your site.To answer criticism that is unfair or out of date, hover a criticism under Strengths and weaknesses and click Respond. Edit the title, then click Write with the agent. Quotable Agent opens with a brief to answer it with honest evidence, and the draft lands in the Library.
5

Reach the sources you do not own

When the source is a review, a list, an article or a forum thread, ask its owner for a correction, or reply where replies are allowed. Give them the current fact and a page of yours that proves it.Quotable Agent can draft the request. Give it the page, the wrong sentence and the correct fact, then copy the draft and send it yourself. Quotable never sends email. See Close a source gap for the outreach steps.
6

Record the correction

In the FactCheck claim, add the page you published under Your corrections with Add link, so your team can see what was done and when. If an automatic verdict is wrong, change it under Verdict. A verdict your team sets counts everywhere and survives re-checks.

How to tell it worked

Fixes move in order: the page changes, engines read it, then answers change.
  • New answers. Tracking asks your prompts every day. Read the answers for the prompts where the claim appeared, on the engine that made it.
  • FactCheck. Watch Contradicted for the claim. A fact you add is checked against existing claims right away, so the counts can change before any engine does. Over the following weeks, look for the claim to stop appearing in new answers.
  • Sentiment. Sentiment score shows net sentiment per tracking day. Use it to see whether the criticism you answered fades.
Be careful about what you conclude:
  • Engines that search the web can pick up a corrected page once they read it again. Engines that answer from memory can repeat an old claim for much longer.
  • A claim can come from several pages. Fixing one may reduce it without ending it.
  • Answers vary from run to run. One clean answer is not a fix. See When numbers shift.

Automate it

From Routines, start from the Sentiment and facts check template. Every Friday it reviews the last 7 complete days of answers that mention the brand: the prompts where the tone is negative or mixed, with the sentence that shows it; claims that look wrong or outdated, with the engine and the cited source; and the sources most often behind them. It quotes the answers instead of paraphrasing them. See Routines.

Next steps

FactCheck

Claims checked against your facts.

Sentiment

How AI engines describe your brand.

Context Manager

The brand facts every agent writes from.
Last modified on November 18, 2026