AI Visibility Tools for Topic Ideas in ChatGPT and Perplexity: AI Search Visibility Content Topics

AI visibility tools for topic ideas get sold as if a dashboard will invent the next brief. That is not how we plan content for ChatGPT and Perplexity. The topics that move citations are the questions buyers already ask, the extra searches the model fans out into, and the pages it already footnotes. We write those down. We do not export a keyword list and relabel it "AI search visibility content topics."

The pitch that a dashboard invents topics is the pitch we hear most often and the one we push back on hardest. A keyword tool answers "what are people typing into a search box." A chatbot topic list answers "what are people asking an assistant, and what does the assistant cite back." Those are different questions, and answering the second with the first is how teams end up with a content calendar full of pages no model ever cited. The topics that move citations come from three places: the prompts buyers actually ask, the fan-outs the model runs from those prompts, and the URLs the model already footnotes. The job is to write those down and to stop at the ones that are yours to write.

The tool we use to hold that list is Promptwatch. Topics and tags there are labels we attach to prompts we typed (campaign, persona, country). Query fan-outs show the related searches the model ran. That is the idea queue. Explore is free, ChatGPT only, 10 prompts. Paid plans add Perplexity (and Gemini, Claude, Overviews) daily from the UI.

How we turn a visibility log into a topic list

Start from sales language. Fifteen prompts the account team already hears. "Best X for Y," a comparison, a "should I switch" question. Not a seed-keyword dump.
Read the fan-outs. If ChatGPT fans a lawn question into herbicides and prevention, those are separate pages only when you have something original to say. Google's own optimization guide already says not to spawn a thin URL per variant. We apply the same rule to chatbot topics.
Read the cited URLs. Citation analytics name the page that won. If it is a competitor comparison you are absent from, the topic is "get mentioned there," not "write our own roundup of roundups."
Tag by persona and market. A CFO prompt and an operator prompt are not one brief. Country-level targeting starts on Essential. State and city targeting require Professional or an agency plan when the answer changes by location.
One new page only when the miss is yours. How to get cited by ChatGPT is the edit checklist. On Professional, Business, or an agency plan, Agent Analytics is the first ticket when the model never fetched you. Essential has no listed crawler-log allowance.

The five steps are an order, not a checklist. Sales language first because prompts that come from the account team are prompts that map to revenue. Fan-outs second because they show the model doing its own research, which is where the long tail of chatbot topics lives. Cited URLs third because the page that already won tells you whether the topic is "write a new page" or "get into the page that exists." Persona and market fourth because a CFO answer and an operator answer are different briefs even when the prompt reads the same. The new page last, and only when the miss is yours, because most of the time the right move is to get cited in an existing page, not to publish a new one.

Essential is $95/mo for a country-level prompt and citation program. Professional is $245/mo with state and city targeting plus 25 million crawler logs. Kick-off is $199/mo with those features, unlimited projects, and 10 seats. Business is $579/mo.

What we keep from the SEO suite

Ahrefs Brand Radar is $199/mo plus an Ahrefs plan. It clusters a modeled corpus. That can be a quarterly "is this category even showing up" slide. An independent January 2026 check reported 3 ChatGPT mentions where 123 were verified. We keep Ahrefs for crawl and links. We do not treat Brand Radar topics as the ChatGPT idea list.

The January 2026 check is the number we keep coming back to. Three mentions where a manual count found 123 is a modeled index that undercounts by a factor of forty. A slide built on that index will say the category is quiet when it is not. The index is fine for a quarterly directional read. It is not fine as the source of a content calendar, because the topics it suggests are the topics the model can see in the model, not the topics buyers ask.

Semrush AI Toolkit is $99/mo per domain on modeled prompts. Useful next to keywords you already track. Poor as the only source of Perplexity topics. We keep Semrush. We write the Promptwatch list in sales language.

Semrush sits in the same role as Ahrefs here. The AI Toolkit is a column next to the keyword data you already pay for, which is useful when you want to see whether a keyword you track has an AI side. It is poor as the only source of Perplexity topics because modeled prompts are a guess about what an engine would say, not a record of what it did say. We keep Semrush for keyword and ad work. We write the Promptwatch list in the language the sales team uses, because that is the language buyers use when they ask an assistant.

Search Console generative reports show Overview impressions by page. They do not suggest chatbot topics. Keep them for Google. See our AI features reading.

Otterly at $29 can hold a cheap ChatGPT and Perplexity glance. Gemini and Claude are add-ons. Seven-day lag. Fine as a proof. Not the editorial calendar.

Otterly has a place, and the place is narrow. A $29 glance at ChatGPT and Perplexity is enough to prove to a stakeholder that the channel exists. The seven-day lag and the add-on engines mean it is not the tool you open on a Monday to plan the week. The proof and the calendar are different purchases.

What we will not do

We will not bill a client to produce a URL for every fan-out phrasing. We will not call a Semrush or Ahrefs topic cluster a ChatGPT content plan. We will not invent a "50 language topic engine."

The refusal is the method. A fan-out is a list of related searches the model ran, not a content calendar. Spawning a thin page per variant is the exact failure Google's own optimization guide warns against, and it produces pages no crawler fetched. A modeled keyword cluster answers a different question than the one a buyer asked an assistant. The topic list that moves citations is the one built from sales language, tagged by persona and market, and checked against the URLs the model already cites.

FAQ

Are Promptwatch topics the same as Ahrefs topics?

No. Same English word. Different object. Ours are labels on prompts we typed.

Can we skip the log and brainstorm topics in a workshop?

You can. You will not see which of those ideas the models already cite, or whose URL they use.

Does Explore give Perplexity topic ideas?

No. Explore is ChatGPT only. Paid Promptwatch is the ChatGPT-plus-Perplexity start.

What to do this week

Write 15 buyer prompts from sales, not from a keyword export.
Load them into Promptwatch Essential ($95) for a country-level topic list, or Professional if state or city targeting or crawler logs are required. Turn on Perplexity on the paid plan.
Export fan-outs and cited domains. Mark three topics you will actually ship.
Keep Ahrefs and Semrush. Do not pause them to "focus on GEO."
If you want us to run the topic triage, email hello@1001seomedia.com with the prompt list. We will not send a 40-URL content calendar from a modeled index.