Prompt Visibility Heatmap, Brand Aliases, and Brand Book for AI Search Monitoring
A visibility heatmap without aliases will miss the name your buyers actually type. The Brand Book is the context layer Content Agents and scoring should share, not a PDF in Drive.
Prompt rows in a table are easy to ignore. A visibility heatmap is the same data in a shape a human will actually argue with: which prompts you own, which you share, which you never appear on. The trap is the brand string. If the model says "PW" or a former product name and your tracker only matches the legal entity, the heatmap is green in the wrong cells, and green in the wrong cells is worse than red in the right cells because it feels like a win.
The wrong-green problem is the one that does the most damage. A red cell in the right place is a problem you can see and fix. A green cell in the wrong place is a problem you cannot see, because it looks like success. The team celebrates a heatmap that says the brand is visible, and the brand is not visible under the name buyers actually use. The green is a measurement error dressed as a win, and the win is what stops you from looking for the real number. A red cell provokes a question. A green cell ends one, and the green cell that ends the question is the one that hides the gap.
Aliases and a Brand Book sound like branding hygiene. They are measurement settings. Get them wrong and every downstream agent works off a misspelled company, and a misspelled company is a company you are not tracking.
The framing matters because it changes who owns the setting. Branding hygiene is a marketing job. Measurement settings are an analytics job. A Brand Book owned as a brand document gets polished and forgotten. A Brand Book owned as a measurement setting gets checked against the data, because the data is what tells you the setting is wrong. The same document, two different owners, and the second owner is the one who keeps the heatmap honest.
Heatmap and the charts around it
Promptwatch's visibility score, mentions over time, and position chart are how you see movement instead of a single scrape. Stacked competitor charts show who took the cell. Period compare is the "did this ship do anything" view. The heatmap is the overview that tells you where to open a row, and the row is where the actual work lives.
The heatmap and the row are two different zoom levels, and both are needed. The heatmap tells you where to look. The row tells you what you are looking at. A heatmap alone is a map without a destination. A row alone is a destination without a map. The two together are what turns a grid of colors into a list of tickets, and the tickets are what the team works on. The heatmap is the index. The row is the work.
Use it after you have a stable prompt list. Twenty purchase questions beat two hundred curiosities. Fan-outs will expand the list; you still pick what is money, because a money prompt that moved is a meeting, and a curiosity prompt that moved is a curiosity.
The money-versus-curiosity split is the split that keeps the list useful. A curiosity prompt is interesting. A money prompt is consequential. A heatmap full of curiosities is a heatmap that produces no meetings, and a heatmap that produces no meetings is a heatmap that gets ignored. Twenty money prompts produce a heatmap that produces meetings, because a money prompt that moved is a thing someone with a budget wants to talk about.
Trends that explain what changed between checks matter more than a prettier heatmap. A cell that went cold because a Reddit thread appeared is not a "we need more content" cell until you look at citations, and looking at citations is what stops you from writing a page against a problem that was actually a forum thread.
Aliases: the unglamorous matching layer
Brands accumulate strings. Legal name, spoken name, old product, common misspelling, abbreviation sales uses on calls. Highlighting those in answers is how you stop undercounting. It is also how you avoid overcounting a generic word that is not you, and overcounting is the error that inflates a report the client later disputes.
The two errors point in opposite directions and both break the report. Undercounting makes a healthy brand look absent. Overcounting makes an absent brand look healthy. Undercounting produces a false alarm. Overcounting produces a false comfort. The false alarm is annoying. The false comfort is dangerous, because it stops you from working on a gap you actually have. Aliases are the setting that holds both errors down, and the setting is unglamorous because it is a list of strings, not a feature with a name.
Set aliases when you create the project, then again after the first week of answers. You will discover a nickname in Claude that never appears in ChatGPT. Personas can change which name shows up. Do not merge persona rows into one average and then wonder why the heatmap looks stable, because stable across personas is often two opposite moves canceling out.
The persona-merge error is the error that hides movement by averaging it away. Two personas move in opposite directions, the average stays flat, and the flat average looks like stability. The stability is fake. The two personas moved, and the movement is the thing the heatmap is supposed to show. Merging the rows hides the movement behind a number that did not move, and a number that did not move is a number that tells you nothing.
This is not a map-pack rank tracker. Country is on paid plans; state and city are on Professional and above. Local GEO is still prompts plus Google Business Profile, not this heatmap pretending to be BrightLocal, and pretending to be BrightLocal is how a national tool gets asked to do a local job it cannot do.
Brand Book as the shared brain
The Brand Book is project context: who you are, how you sound, claims you will stand behind. Content Agents read it so drafts do not invent a tone. Monitoring still needs it so "the brand" in an answer can be scored as you, a competitor, or a generic category, and that scoring is what makes a mention count as a mention of you.
The scoring is the part that makes the Brand Book a measurement document, not just a writing document. A mention of a word that matches your brand is not always a mention of you. It could be a competitor with a similar name. It could be a generic word. The Brand Book is what lets the scoring tell those apart, and the scoring is what makes the mention count meaningful. Without the Book, the count is a count of strings. With the Book, the count is a count of you, and the difference is the difference between a number and a number that means something.
Keep it short enough that a model will use it. A 40-page manifesto becomes ignored context, and ignored context is the same as no context. Put the forbidden claims in writing: no invented funding, no fake customers. Review inbox exists because the Book will still be incomplete, and incomplete is the normal state of a Brand Book, not a failure of it.
The shortness is a constraint that makes the Book work. A model reads a short Book and uses it. A model reads a long Book and skips it, because the long Book exceeds the context the model will actually attend to. The forbidden-claims list is the part that does the most work per word, because it is the part that stops a draft from inventing a funding round or a customer, and a draft that invents those is a draft that produces a legal problem. The forbidden list is short, and the short list is the one that gets read.
Jasper Brand Voice is a cousin for writers. It is not a visibility heatmap. Otterly.AI will monitor mentions from $29 without this workspace. Fine. Different object, and a different object is not a worse object, it is a different question.
How this turns into work
Unified Actions should not fire "write article" on every red cell. Red plus no ChatGPTBot fetch is a technical ticket, and the fix is in robots.txt, not in the content. Red plus a competitor URL in citations is a page-level fight, and the fight is the work. Red plus an alias miss is a settings ticket, and those are the cheap wins, because a settings ticket fixes a whole column at once.
The three reds are three different tickets, and treating them the same produces the wrong fix every time. A red cell with no fetch is not a content problem, and writing an article for it is writing an article no bot will read. A red cell with a competitor URL is a content problem, and writing an article is the right fix. A red cell with an alias miss is a settings problem, and fixing the alias is a one-minute fix that corrects a whole column. The three fixes live in three different places, and Unified Actions is the layer that sorts them, because a "write article" ticket on a robots.txt problem is a ticket that wastes a writer's week.
Explore, free with 10 ChatGPT prompts, is enough to see whether aliases were wrong. Essential at $95 per month is the first paid heatmap that includes more than ChatGPT. Professional at $245 per month adds the city layer and 25 million crawler logs, and the city layer is what turns a national heatmap into a local one without changing tools.
A heatmap is a reporter until aliases are true and the Brand Book is the same document the agent writes from. Then it is a control panel. That is the agentic reading of a feature that looks like a pretty grid, and the grid is the surface, not the substance.