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

The first week of AI search monitoring is setup, not content. If aliases are wrong, the heatmap is wrong, and the weekly slide argues about spelling. We do this work in Promptwatch before we change a page.

We run client programs on paid plans (Essential at $95/mo for one brand, Kick-off at $199/mo when we hold several). Explore is free, ChatGPT only, 10 prompts. Fine for a stakeholder demo. Not a program.

The distinction matters because the work in the first week is unglamorous and easy to skip. Aliases and Brand Book are the kind of setup that does not produce a chart, which makes them the first thing a rushed program drops. We do not drop them, because everything downstream reads from them. The heatmap reads aliases to decide what counts as a mention. The Content Agents read the Brand Book to decide how the brand should sound. A program that skips the setup spends the rest of the quarter arguing about whether a mention counted, which is the worst use of a reporting call.

What we load before the first check

Brand Book. Name, aliases, and tone of voice live in one place. Content Agents and briefs read that context later. If legal has a house style, it goes here. If marketing has a nickname sales uses on calls, it goes here too.

Brand aliases. Legal name, product name, common misspellings, the acronym people type. Promptwatch marks matched aliases in the response so we can see what the model wrote. A mention of the product line that never says the parent brand still counts if we listed the line.

We do not add every historical DBA from 2011. Noise aliases inflate mention counts and make the stacked competitor chart look like a win. We add the strings buyers and models use.

The test for an alias is whether a buyer would recognize it and whether a model would write it. A defunct trade name from a decade ago fails both tests. It does not help the heatmap find real mentions, and it inflates the count so the share-of-voice chart looks better than it is. The same logic applies to abbreviations that only insiders use. If the model would not write that string, tracking it produces a number that has nothing to do with visibility.

Competitors. Same rule. Direct names only. Auto-fill from a URL is fine for the first pass. We still delete the publisher that is a citation source, not a rival.

How we read the heatmap

The Prompt Visibility Heatmap is the Monday scan: one row per prompt, one column per model. Mentioned, cited, absent. We do not average ChatGPT, Gemini, Claude, Perplexity, and the rest into one "AI score." Claude can skip a page ChatGPT cites. That disagreement is the finding. A single blended number is the thing that hides the finding, because the disagreement is where the work is, and the average smooths it out until it looks like nothing is happening.

Reading the heatmap by engine is what makes it useful. A row that is cited on ChatGPT and absent on Claude is not a 50% win. It is a Claude problem, and the Claude problem has its own fix that has nothing to do with the ChatGPT column. When you blend the row into a single score, you lose the column that told you where the work was. The heatmap exists to keep the columns separate, so the reading has to keep them separate too.

Paid engines we start with: ChatGPT, Gemini, Claude, Perplexity, plus Google AI Overviews and AI Mode. Grok, Llama, DeepSeek, Mistral, and Copilot sit on the same paid plans. We add them when the buyer uses them, not because a slide needs more logos.

Next to the heatmap we keep three charts, and we keep them separate:

• Mentions over time, so a one-day flip does not become a rewrite.
• Average position across the prompt list, so we see whether we are first-named or a footnote.
• Stacked competitor mentions, so share of voice has a denominator.

Each chart answers a different question, which is why they stay separate. Mentions over time tells you whether a change held or was a spike. Average position tells you whether the brand is named first or buried in a list. Stacked competitor mentions gives the share-of-voice number a denominator, so a mention count means something relative to the field rather than in isolation. Combining them into one chart loses the question each one answers.

Period compare is the A/B we run after a change, not before. Two weeks of daily checks (paid plans refresh daily) beat a screenshot from someone's laptop. Otterly at $29 is a cheap mention check with Gemini as an add-on and lag. Peec starts at $95. Profound Starter is ChatGPT only at $99 on annual billing. None of those replace a heatmap that already has aliases and a frozen list.

What we refuse to report

We will not treat a highlighted competitor name in one answer as a crisis. We will not rename the brand in the Brand Book mid-quarter without a note on the slide. We will not claim Promptwatch sends instant alerts. It does not. Daily checks, then we read the heatmap.

G2 rates Promptwatch 4.7/5. That is a review score, not a forecast for your prompts.

FAQ

Do we need Professional at $245/mo just to see the heatmap?

No. Essential at $95/mo is enough for one project, 50 prompts, and the visibility views we use in week one. We move to Professional or Business ($579/mo) when we need a second project, more seats, or state and city tracking. Kick-off at $199/mo is the agency SKU: unlimited projects and prompts, 10 seats.

What if the legal name and the product name are different?

Both go in aliases. The Brand Book keeps the official name as the primary. The heatmap will miss the product if we only tracked the holding company.

How we run it

1. Write the Brand Book with the client in the kickoff, not after the first report.
2. Load aliases and two or three direct competitors. Freeze the prompt list.
3. Run a quiet week. Read the heatmap by engine. Do not change pages yet.
4. Pick one prompt that is absent on two engines. That becomes the ticket.
5. If you want us in the seat, hello@1001seomedia.com. We set aliases. We do not invent mention rates.