The accuracy monitor reads what AI answers say about your brand and compares it with the brand facts you confirmed. It lives on the Accuracy tab of Sources, on the Pro and Agency plans. The tab has three views: Issues, Claims to check and Brand Facts.
What gets checked
- Each checked answer to a tracked prompt, on every platform, is read for claims about your brand: prices, dates, numbers, yes/no statements, addresses, phone numbers, opening hours and similar facts. An answer that never names your brand is skipped, because it cannot make a claim about it
- Each claim is compared with your brand facts. Only confirmed facts can flag an answer as wrong
- Prominara also reads facts from your own pricing, home, about and docs pages after a crawl. These wait as Unconfirmed in Brand Facts and are not used to check answers until you confirm them
- A claim no fact covers is not an error. It is counted as “no fact held” and, if it repeats, listed under Claims to check
- Opinions (“a trusted provider”) are not checked here; they belong to sentiment
At the top of the Issues view, Answers analyzed counts the answers read, Claims we could check shows how many claims one of your confirmed facts covered, and Accurate is the share of those claims that matched. The last two are kept apart on purpose: a high accuracy rate over a handful of checkable claims says little. Each fact you add lets more claims be checked.
How an error is decided
Values are compared by type first: “$49/month” and “$49 per month” are the same price, and an answer that agrees with any fact that applies to the claim is accurate. When the comparison cannot decide, an AI model reads the claim and your fact.
Every contradiction, whichever step found it, is checked a second time with a narrower question: can the answer and your fact both be true? It is reported only when the answer is no. If the two turn out to be about different things, the claim is not reported and can appear under Claims to check instead. If the review cannot run, nothing is reported.
Issues
Each issue card shows:
- The claim, in quotes, and the fact it was compared with (“Compared with your fact opening hours: …”)
- Severity (Critical, High, Medium or Low) and the platform that said it
- Nx seen - How many times this error has been seen
- Returned Nx - The issue was resolved and the same error came back
- Fixed on its own - Closed automatically after 3 correct answers in a row on the same platform, spread over at least 7 days
Open a card to see the Evidence: the question that was asked, the platform's own words, the fact it was compared with, when it was last seen and a View the answer link to the full answer. The card also carries a correction playbook and a Status control: Detected, Reviewing, Correcting, Resolved or Dismissed. Filter the list by severity or status.
One wording of a mistake and another are the same issue; two different wrong values are two issues. A resolved issue that comes back reopens and alerts again. A dismissed issue never reopens.
Alerts
A new error raises a Factual Error alert, at most once a week for the same wrong value of the same fact across all platforms. The alert quotes the answer and names the fact it was compared with, so a wrong comparison is visible without opening the app. By email, a site's factual errors from one delivery run arrive as one message. Choose how alerts reach you under Critical alerts in Notifications; see Monitors & Alerts.
Dismissing an issue
Choose Dismiss on an issue (or set its status to Dismissed) and pick a reason under Why are you dismissing this?
- The AI is right - What the answer said is correct. Its value is saved on the fact as another accepted value, so an answer stating it is counted as accurate from then on
- Compared with the wrong fact - The answer is about something else. That fact is no longer compared with claims about the same property
- Not worth tracking - The answer is wrong, but you do not need to act on it. Nothing is learned and the issue stays closed
The first two change a brand fact, so they need permission to edit brand facts (owners and admins). Moving an issue out of Dismissed, or choosing another reason, takes the earlier change back off the fact, so a wrong click can be undone from the card.
Claims to check
This view lists statements AI answers made about your brand that none of your brand facts covers: a price, date, number, address, phone number, opening time or yes/no value, said in at least two answers in the last 90 days. Each entry shows the property, the value, how many answers said it and on which platforms, and one example answer with a link to it. Properties a confirmed fact already covers are managed in Brand Facts and are not listed.
- Correct - Saves the statement as a confirmed brand fact. It is checked from then on, and answers that gave the same property a different value are read again
- Wrong - Asks What is the right value? and saves your value as a confirmed brand fact. The answers that made the claim are read again, and any that contradict the new fact open issues the usual way
- Ignore - The statement is not listed again
Answering needs permission to edit brand facts. Claims to check sends no alert of its own; the weekly digest shows how many are waiting.
Brand Facts
Facts are grouped by category (Pricing, Features, Founding, Leadership, Location, Metrics, Partnerships, Awards, Custom). To add one, choose Add Fact, pick a category, enter a key (for example monthly_price) and a value, and save. A fact you add is confirmed straight away, and facts you enter or edit are never overwritten by what Prominara reads from your site.
Facts read from your site carry an Unconfirmed label and a Confirm button; hover it to see the sentence on your site the fact came from. Editing an unconfirmed fact also confirms it. The Issues view shows a banner while facts wait, and owners and admins get reminder emails (see Monitors & Alerts).
Value types
Each fact has a type, which decides how answers are compared with it. The type is set to Auto by default and shows what was detected; you can set it yourself: Price, Number, Date or year, Yes / no, Address, Phone, Email, Website, Opening hours, List (partial), List (complete) and Text.
- Price, Number, Date or year and Yes / no values are compared directly. A price written as a floor (“from €40”) is contradicted only by a lower price
- Address, Phone, Email, Website and lists can be confirmed by a match (a shorter address, a reformatted phone number, part of a list). Anything else goes to the AI review
- Opening hours always go to the AI review. Text is accurate on an exact match and otherwise goes to the AI review
- For a List (partial), the review is told the list may be incomplete, so an answer naming something missing from it is not treated as wrong on that ground alone. Set List (complete) when the list is the whole set