Wellows' Citation Score vs our visibility score: two ways to read the same AI answers

Our whole service rests on one number: the share of an agreed set of buyer prompts in which AI engines name you. When a prospect shows us a different score from a different tool, we don't argue about which is right. We work out what the other number is counting, because two honest scores can move in opposite directions in the same month.

Wellows is a good case to work through. It is a citation-first AI visibility platform, and its two headline numbers are built on a different unit and a different denominator from ours.

Sources: wellows.com/pricing and wellows.com/llms.txt (last updated 23 September 2026), both read on 25 October 2026. Promptwatch facts are from its llms.txt, re-checked on 21 October 2026.

How Wellows scores visibility

Wellows collects answers from ChatGPT, Google AI Overviews, Google AI Mode, Perplexity and Gemini through each engine's public interface, logged out with web search on. It then reads every source in the citation list. Its scoring counts citations, not mentions.

AI Visibility Score is your share of citations against the competitors you track, across your tracked prompts. You can track up to ten competitors per brand.
Citation Score is a composite. It combines explicit citations (the engine cites your own page) and implicit citations (it cites a third-party page that mentions you), weighted by intent and by competitor share of voice.

Both are split into branded and non-branded prompts. Mentions are tracked, but separately from the scores.

How we define ours

We agree a set of buyer prompts with the client at the start. Each month we count the prompts in which AI engines name the client, divide by the total, and report it per engine and overall. If you're named in 14 of 40 prompts, the score is 35. Competitors don't enter the formula. They show up underneath, in a share of voice view, so the client can see who took the answers they didn't get.

Three real differences

The unit. Wellows counts citations; we count being named. They overlap but aren't the same event. An engine can name you in its text while citing pages that never mention you. It can also cite a review page that mentions you and still recommend a rival in the answer. Wellows' implicit citation is a citation of a page about you, which is different from the engine saying your name to the buyer.

The denominator. Wellows divides by citations across you and your tracked competitors. We divide by prompts. A relative score moves when competitors move. An absolute one only moves when you do.

The formula. The AI Visibility Score is a plain share. The Citation Score blends two citation types with weights for intent and competitor share. That makes it richer for analysts, and harder to explain to a board in one sentence.

A worked example

The numbers here are made up to show the mechanics. They are not a client result.

Say a brand is named in 14 of 40 agreed prompts: a visibility score of 35. Across the same prompts, the engines produce 80 citations that count for either the brand or one of three tracked competitors. Sixteen of them count for the brand, a 20% citation share.

Next month, one competitor's site starts blocking AI crawlers and loses 16 citations. The brand's own 16 haven't changed, but its share rises to 25% (16 of 64). Meanwhile the engines still name the brand in 14 prompts, so the visibility score stays at 35. Neither number is wrong. One says "you gained ground relative to rivals", the other says "your buyers saw you just as often as before". A client who only saw the first would think the program had worked.

When the competitor set changes. Add a fourth or fifth rival and your share can drop overnight with nothing changing in the answers.
When rivals stumble. As in the example, gains can come from someone else's problem.
When the answer names brands without citing them. A citation count can't see a recommendation that comes with no supporting link to a page about you.

Wellows gives some protection. Sentiment labels each cited page that mentions you as positive, neutral or negative, and the branded and non-branded split stops easy branded prompts from flattering the total.

When our visibility score misleads

We should be equally honest about our own number.

Small prompt sets are jumpy. With 40 prompts, each one is worth 2.5 points. Lose two prompts to answer variance and the score falls 5 points without anything real changing. That's why we never score a prompt on a single check.
Prompt selection decides the ceiling. If the agreed set is full of easy prompts, the score flatters. We keep branded prompts out of the headline for that reason, the same instinct behind Wellows' branded split.
Being named isn't the same as being recommended. Fifth in a list of five and the top pick both count as named. We read the stored responses before a monthly report goes out.
It says nothing about why. A score of 35 doesn't tell you which pages engines leaned on. That's where citation data, including Wellows-style citation analysis, earns its place.

Engine coverage and budget

Both scores depend on which engines are in the sample. Wellows doesn't track Claude or Microsoft Copilot. Our score includes both, because some buyers use them.

Budget matters too. On Wellows, one AI answer is one prompt on one engine on one day. Forty prompts on all five engines daily needs about 6,000 answers a month, which fits the Standard plan's 6,500 at $97/mo. Weekly cadence costs roughly a seventh as much, but a monthly score built on four checks per prompt is noisier than one built on thirty.

Where we measure it

We measure the score in Promptwatch. Prompt tracking runs our agreed set across ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, Google AI Overviews and AI Mode on every paid plan, so moving plans never changes which engines count. The stored responses let us open any prompt and check how the client was named. Citation analytics, down to page, domain, Reddit and YouTube level, explains the movement, and share of voice against the agreed competitors sits underneath. Crawler logs and visitor analytics answer the next questions: could the bots reach the page, and did anyone click through.

If a citation share is the number your team already reports, keep it. Just don't read it as the share of buyers who saw you. And whichever score you use, be wary of anyone who promises a particular number. We don't promise a score of 99, or any other figure.