Prompt Visibility Heatmap, Brand Aliases, and Brand Book for AI Search Monitoring

A visibility score falls apart if the model wrote "Acme Inc" and your tracker only searched for "Acme". The reverse is just as messy: a generic word matches and you celebrate a mention that is not your company. AI search monitoring needs the brand name, the aliases people actually use, and a grid that shows which prompts and which models mentioned you. Without all three, the score is either undercounting your real presence or counting someone else's. The three pieces are not optional extras stacked on top of a score. They are what make the score mean something. A number on its own is a tile. A number plus the alias that produced it plus the model it came from is a diagnosis.

Promptwatch is the tool that puts those pieces in one login: Visibility Score, mentions over time, a position/rank chart, alias highlighting in answers, stacked competitor charts, a two-range compare, Brand Book, and a prompt visibility heatmap. Explore is free (ChatGPT, 10 prompts). Essential is $95/mo with a 7-day trial (50 prompts, 6,000 responses). Site: promptwatch.com.

What the score is counting

Visibility Score is the share of checked answers where your brand showed up. That phrasing matters. It is not a count of mentions, and it is not a count of prompts. It is a share, so a number that holds steady can still hide movement underneath it, because the denominator moves too. If you checked 40 prompts this week and 60 last week, the same share means a different number of mentions, and a program that reports only the share will miss that.

Mentions over time is that number as a series. A single week can look fine while the last 30 days are sliding, so you want the chart, not a tile. A tile tells you the score today. The chart tells you whether today is the top of a hill or the bottom of a valley, and that distinction is the whole reason to monitor rather than check once. A one-off check answers "are we visible". A chart answers "are we getting more or less visible", and those are different questions that pull different teams into the room.

The position/rank chart is the other half. You can be named and still sit under two rivals. "We got mentioned" is not the same as "we were the first brand in the answer," and a buyer who reads the first name and stops never sees the second. If your program reports only mentions, you will report a win the week you actually lost share of voice, because the model started naming a competitor above you. Rank is the number that catches that, because rank moves before share does. A rival jumping from third to first does not change your mention count, but it changes who the buyer actually reads.

Paid Promptwatch plans check the real UI on ChatGPT, Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, plus Google AI Overviews and AI Mode, and they refresh daily. Explore stays ChatGPT-only. That is enough to learn the score. It is not enough to trust a Gemini gap, because a Gemini gap you cannot see is a gap you cannot prove, and a gap you cannot prove does not get a ticket. The list of engines is also the list of places a brand can quietly disappear, and daily refresh is what turns "we think it dropped" into "it dropped on Tuesday".

Aliases and the Brand Book

Legal name, product name, and the short name buyers type are often different strings. Add them as aliases (Acme / Acme Inc). Promptwatch highlights matched aliases in the answer so you can see which form the model used. If the highlight never fires on the legal name, the model has already picked the nickname. Track that nickname, because the nickname is the string your buyers are going to type into the next assistant, and the one after that. Aliases are not a completeness exercise. They are how you keep the score attached to the string that actually represents you in the answer, rather than the string your legal team prefers.

Brand Book stores the name, aliases, and tone of voice. Monitoring reads that list so a mention is not dropped. Content workflows read the same book so a draft does not invent a second voice. The two readings matter for the same reason: a brand that calls itself three different things across three pages teaches the model three different names, and the model will pick one of them without asking. The Brand Book is the single source both sides read from, so the monitoring side and the writing side stop arguing about which name counts. You can set aliases when you create the project, which is the right time to do it, before the first answer comes back and locks in a name you did not choose.

Heatmap, stacks, and two date ranges

The prompt visibility heatmap is a prompt-by-model grid. One row can be solid on ChatGPT and empty on Gemini. That is the picture you want before you rewrite a page for "AI" as if it were one engine, because it is not one engine, and a fix that moves ChatGPT can leave Gemini exactly where it was. The heatmap is how you stop talking about "AI visibility" in the singular and start talking about the engine you are actually losing. A row that is green on one model and blank on another is a more useful signal than a blended score, because it tells you which engine to file the ticket against.

Stacked competitor charts put your mentions next to named rivals on the same prompts. Compare two date ranges after you ship a page. The before/after should be a chart, not a Slack memory, because a Slack memory is the version that gets edited to sound better in the standup and the chart is the version that survives the quarter. The two-range compare is also what separates a change that worked from a change that coincided with a model update. Without the before range, you cannot tell whether the page edit or the model moved your number.

Otterly.AI from $29/mo will show mentions on four base engines (Gemini and Claude are add-ons) with up to a 7-day lag. Peec AI from $95/mo scores prompts on three models. Neither is the heatmap-plus-Brand-Book setup. Ahrefs Brand Radar ($199/mo plus an Ahrefs plan) is an index, not a prompt list you typed. Method: how we rank. Tools list. Promptwatch is 4.7/5 on G2 across 1,840+ brands. Founded April 2025 in Amsterdam by Gijs de Groot and Klaas Foppen.

FAQ

Do I need every legal entity as its own brand?

Usually no. Put the trading name and the common short forms in Brand Book first. Split a second brand only when the answers treat it as a different company.

Does Explore include the heatmap across Gemini and Claude?

No. Explore is ChatGPT only. Multi-engine heatmap rows start on paid plans.

What to do this week

List the names people actually type for you, including the Inc / Ltd form.
Run 10 buyer prompts by hand on ChatGPT and Gemini. Mark which string appeared.
Load those prompts into Promptwatch and fill Brand Book.
Open the heatmap and write down every prompt that is empty on one model.
Compare this week to last week after you change one page, not five.