How to read citation and mention findings

Answer engines do not give you a rank URL. They generate a paragraph and, if you are lucky, they name someone. That makes the vocabulary easy to abuse. “We got cited” can mean the business was the whole answer, a name in a list, or a hallucination you would not want in a slide.

Before you brief anyone, separate prominence from rate.

Featured, mentioned, not mentioned

Treat each engine result as one of three states:

• Featured — the business is the primary recommendation or the subject of the answer.
• Mentioned — the name appears, but it is not the lead. A list of three dentists that includes you is a mention.
• Not mentioned — absent. Do not score this as “the model prefers a competitor” unless the answer actually named someone else.

Featured is not a gold star you can put on a proposal without context. A branded prompt (“tell me about [business]”) will featured-hit businesses that category prompts ignore. Mentioned on a comparison prompt can be more commercially useful than featured on a vanity prompt.

If you only report “appearances,” you will treat a passing name-check as equal to being the answer. It is not.

Citation rate is a percentage, not a grade

Citation rate is simple: of the engine results you actually collected, what share mentioned or featured the business?

It is only as honest as the prompt set. Ten branded questions and two category questions will print a flattering rate. Flip the mix and the same business looks invisible. Always say how many scenarios you ran, and keep branded results out of the category percentage.

Break the rate down twice if you can:

• By intent — branded discovery, local, category, service expansion, comparison, hire/buy/visit
• By engine — ChatGPT, Gemini, Google AI Overviews, and whatever else you queried

A 60% overall rate with 0% on AI Overviews is a different briefing than 60% everywhere.

Citation Index weights prominence

A citation rate counts a mention and a feature as the same “yes.” A Citation Index should not. In NovaScope we score featured higher than mentioned (two points vs one), then scale to 100 against a perfect featured result on every test. That is why Index and rate can diverge: lots of weak mentions lift the rate and barely move the Index; a handful of featured answers do the reverse.

You do not have to use our number. You do have to say whether you are counting presence or weighting the lead.

Share of voice is a different question

Share of voice asks: in the answers you collected, how often was this business named versus other businesses the model named?

It is not “share of the market.” It is share of names in this sample. If the model lists four competitors and you once, your share is small even if your citation rate looks fine on branded prompts. If nobody is named, share of voice is not a story — there is no voice to share.

How to put this in a client email

Lead with the split, not the headline number:

Branded recognition is reliable. Category and AI Overview citations are thin. Featured answers are rare outside the business name.

Then attach the counts: cited / total per cluster and per engine. If a competitor leads on mentions, name the competitor and the prompt types, not a vibe.

Verify the names before you send. Models invent firms, merge two listings, and cite a directory page that is not the client. Open the sampled URLs. Check that the competitor string is a real business. If a security or technical finding is in the same report, confirm it against the live site — generated write-ups drift.

A finding you cannot reconstruct from the underlying answers is not a finding. It is a sentence.