Miscited · Writing

AI visibility and answer accuracy are different things

Both categories sell dashboards about AI answers, so they look like the same product bought for the same reason. They are not, and one worked example separates them permanently.

The example

Ask any assistant what Slack's free plan keeps. Many will answer something close to this:

"Slack's free plan keeps your 10,000 most recent messages, so nothing is lost while you stay under that limit."

The brand is named. The tone is positive. One of the citations is Slack's own pricing page. The 10,000-message limit ended on 1 September 2022; the free plan keeps 90 days of history. You can verify both halves in under a minute.

Now score that answer with each kind of tool. A visibility tool records a mention, positive sentiment and a citation to the brand's own domain, which is the best result it knows how to record. An accuracy tool records a defect: the claim contradicts a fact whose expiry date has passed, and the cited page does not support the sentence attached to it.

Same answer. Opposite scores. That is the whole distinction.

What each category actually measures

Visibility monitoringAnswer accuracy
UnitA mentionA claim, in an intent family, on a surface, in a market, at a time
Core metricsMention rate, sentiment, share of voiceDefect rate against dated facts, citation support rate
Needs from youA list of promptsA registry of your facts, with effective dates and expiries
Catches a stale priceNoYes
Catches a citation that does not support its claimNoYes
Catches total absence from a category answerYesYes
Proves a fix workedRarely, and usually by before/after aloneRequires matched controls and a stated effect size

The three questions that tell them apart on a demo call

Ask any vendor in this space these, in this order.

1. "What is the sample size behind this percentage, and its interval?"

If the answer is a number with no denominator, the dashboard cannot tell a real change from noise. We worked through what each sample size actually earns you in how many prompts an AI visibility number needs. Short version: 25 prompts supports the observation that something happened, not a rate.

2. "Show me an answer that mentions us positively and is still wrong."

A visibility product has no way to represent this state, because nothing in its data model holds what is true. If the demo cannot produce the case, the tool cannot detect the case.

3. "When we fix something, how do you know the fix caused the change?"

Watch for a control group. Models update on their own schedule. Without a set of questions deliberately left untouched over the same window, a before-and-after comparison attributes every model update to your content team.

Which problem do you have?

This is genuinely situational, and vendors in each category have an obvious incentive to tell you it is theirs.

The honest summary: visibility tells you whether you are in the room, accuracy tells you whether what is being said about you in that room is true. The second question only exists once the answer to the first is yes, and for most established B2B companies the answer to the first is already yes.

Questions people ask about this

What is the difference between AI visibility monitoring and answer accuracy monitoring?

Visibility monitoring measures whether your brand appears in AI answers and how the mention reads: mention count, sentiment and share of voice. Accuracy monitoring measures whether the statements in those answers are true, by checking each claim against a dated registry of your own facts and checking whether each citation supports the claim attached to it. A confident answer that names you positively and states a price you retired scores as a success in the first category and a defect in the second.

Do I need both AI visibility and accuracy tracking?

Visibility is an input to accuracy: you cannot check a claim in an answer that never mentions you. The practical question is which failure costs you more. If buyers cannot find you in AI answers at all, visibility is the binding constraint. If they find you and act on something false, accuracy is.

Why does share of voice not catch a wrong AI answer?

Because share of voice counts mentions and weighs their tone. It has no representation of what is true. A wrong claim delivered in a positive tone with a citation to your own domain increases share of voice and sentiment at the same time as it costs you the deal.