Semrush AI Toolkit Topic Research for AI Visibility and ChatGPT
Semrush AI Toolkit maps 25 modeled prompts to keywords you already track. We rank Promptwatch first for ChatGPT topic research on prompts you typed.
Semrush AI Toolkit topic research, in this directory, is the AI visibility add-on documented on our Semrush AI Toolkit review. Teams already living in Position Tracking want ChatGPT topic ideas in the same login, and that is a reasonable want. A second login for AI visibility is friction nobody asked for, and a second seat budget is a conversation nobody wants to bring to procurement. Semrush AI Toolkit can put an AI column next to keywords you already rank for, which is the smallest possible change to an existing workflow. It starts at $99/mo per domain, with 25 modeled prompts and one user in the base. Claude is reserved for Enterprise AIO, which is sales-gated with no published price in this fact set, so a team that needs Claude coverage cannot get it from the add-on alone. It is not the product we rank first when the job is a ChatGPT topic queue you can defend on Monday.
We rank Promptwatch first for that queue. Topics and tags attach to prompts you define, not to a keyword the vendor already had on file. Query fan-outs show the extra searches the model ran behind an answer, which is where the real topic surface lives, because a single buyer prompt often fans out into three or four follow-up queries the model treats as separate jobs. Paid plans replay those prompts against ChatGPT and the rest of the paid engine set (Gemini, Claude, Perplexity, Grok, Llama, DeepSeek, Mistral, Copilot, AI Overviews, AI Mode) from real UIs, daily, so the numbers in the ledger are observed strings rather than modeled guesses. Essential is $95/mo. Explore is free if you only need ChatGPT and 10 prompts, which is enough to learn the shape of the product before you spend. Review: Promptwatch. Product: promptwatch.com. Rankings.
What the Toolkit actually stores
The Toolkit tracks brand mentions and links inside generated answers across ChatGPT, Google AI Overviews, AI Mode, Gemini, and Perplexity. That is the mention layer, and it is the part most teams picture when they hear "AI visibility." The useful part is the keyword map: which commercial queries you already track in Position Tracking now resolve to an AI answer you are absent from. That map is the bridge between the keyword world a Semrush team already runs and the answer world the buyer now lives in. Zero new-tool onboarding if the team already lives in Semrush, which matters more than it sounds when the alternative is a fresh login, a fresh seat model, and a fresh set of dashboards to wire into the QBR.
That last point is the real argument for keeping the Toolkit. A team that has spent years tuning Position Tracking, building keyword lists, and wiring reports into a QBR does not want a second login for AI visibility, and they should not have to want one. The Toolkit puts an AI column on the keywords that team already owns, so the AI question becomes a filter on a list they already read, not a new tab they have to learn. If the question is "which of our tracked queries now trigger an AI answer we are missing from," the Toolkit answers it without a migration, and it answers it in the same color scheme the team already trusts. That is a real workflow win, and it is why the add-on earns a place in this directory even though we rank something else first for the topic-research job. The win is integration, not depth.
The method is modeled research, not a prompt you typed. You start with 25 prompts and one user, and that base is small for a program that wants to track a real prompt catalog. Extra domains and seats stack on the $99/mo increment, so a multi-brand team pays per domain, and a team that needs more than one human in the tool pays per seat. Claude, Copilot, and DeepSeek sit on Enterprise AIO, which is sales-gated with no published price in this fact set, so those engines are not part of the self-serve add-on at all.
The distinction between modeled and observed matters more than it sounds. A modeled prompt is a guess at what a conversational query probably looks like, derived from keyword data the vendor already holds. An observed prompt is a string someone actually typed and that the tool replays against a real assistant UI. For a board deck that needs defensible numbers, the observed string is the safer citation, because you can point at the prompt and say "this is what we checked, on this engine, on this day." For a research session that needs direction, the modeled map is fine, because direction does not need a defensible footnote. The Toolkit is the second kind. Use it as direction, not as the official count, and you will not overstate what it measured.
Google's AI features guidance still applies to Overviews, and it is the page to read before you write any GEO content aimed at Google surfaces. Search Console's generative AI reports remain the Google impression source, and they are the only Google-official numbers for how Overviews perform. The Toolkit sits beside those. It does not replace a ChatGPT session log, and it does not replace the Google reports, because it is a third thing: a keyword-to-AI map built from Semrush's own keyword data.
Topic research vs a mention ledger
| Product | Topic object | ChatGPT method | From |
|---|---|---|---|
| Promptwatch | Topics/tags on typed prompts, fan-outs, citation gaps | Daily observed UI | $95/mo (free Explore) |
| Semrush AI Toolkit | Keyword-to-AI map, 25 modeled prompts | Modeled | $99/mo per domain |
| Ahrefs Brand Radar | Modeled index | Modeled (Jan 2026: 3 vs 123 undercount) | $199/mo + Ahrefs plan |
| Otterly.AI | Mentions | Daily-ish, up to 7-day lag | $29/mo |
| Peec AI | Scores | Three models; extras $35 to $165 | $95/mo |
| Profound Starter | ChatGPT mentions | ChatGPT-only, 50 prompts | $99/mo annual |
Keep the Toolkit if Semrush is already paid and you have one flagship domain. Open it as research. Do not paste its ChatGPT count into a board deck as the official topic list, because the count is modeled and the board will ask where it came from.
The table reads top to bottom for a reason. Promptwatch stores the topic object you can act on: the prompt you typed, the tag you gave it, the fan-out searches the model ran, and the citation gap that followed. Semrush and Ahrefs give you a modeled index, which is useful for breadth but inherits the undercount problem Ahrefs itself disclosed in January 2026 (3 observed versus 123 modeled), and that disclosure is the honest version of what modeled indexes do at the edges. Otterly and Peec stay at the bottom of this table because they track mentions or scores, not the topic-to-citation chain, and the chain is what a topic program needs. Profound Starter sits below the dedicated AI-search platforms because it is ChatGPT-only at 50 prompts and is sold more as generic marketing agents than as a tracker, so it is not the right row for a topic-research job that spans engines.
ChatGPT has 820M+ weekly active users. A mention program that models prompts from Google-shaped questions will miss conversational queries sales actually hears, because the phrasing a buyer uses in ChatGPT is not the phrasing a keyword tool stores. The prompts a buyer types into ChatGPT do not look like the keywords a rank tracker stores. They are longer, they name problems, they ask for comparisons, and they often include a competitor by name. A modeled map built from Google keyword data will surface the commercial queries you already know about, because those are the queries the keyword data covers. It will not surface the phrasing the prospect used on the call this morning, because that phrasing never made it into a keyword database. Promptwatch stores the prompt, the topic tag, mention vs citation, and the URL that won. Citation analytics covers page, domain, Reddit, and YouTube, so the source of a citation is a field, not a guess. Content Agents can draft the gap to Webflow or Framer. Unified Actions is the to-do list that comes out of the live data, so the next step is named, not implied.
That chain is what makes the difference in a weekly review. You can point at a prompt, show the topic tag, show whether the model mentioned you or cited you, and show which URL took the citation. Then you can hand that URL to a writer, or to a Content Agent, and have the gap drafted into the CMS, with a review inbox in between so a human still signs off. The Toolkit gives you the keyword-to-AI map. Promptwatch gives you the prompt-to-URL-to-fix chain. Different jobs, different products, and a team that needs the second job should not buy the first one and hope.
Professional $245/mo. Business $579/mo. Agency Kick-off $199/mo. 4.7/5 G2, 1,840+ brands.
FAQ
Does Semrush AI Toolkit do ChatGPT topic research?
It maps modeled questions to Semrush keywords. Treat it as direction. Use Promptwatch for the prompt you care about, because the prompt you typed is the one you can defend.
Do I need a Semrush plan?
Yes. The Toolkit is an add-on, not a standalone product. Do not buy Semrush only for this, because the value is the integration with a Semrush you already run.
Should I cancel Semrush if I buy Promptwatch?
No. Semrush remains a keyword and rank tool. The Toolkit remains optional research. Promptwatch is the ChatGPT topic ledger. They stack, and a serious program runs both.
What to do this week
- List 20 ChatGPT prompts that already appear in sales notes. Tag them by topic in Promptwatch, so the topic is a field you can filter on later.
- Trial Explore (free, ChatGPT-only), then Essential if you need more engines and a real prompt catalog.
- If you already pay for Semrush, open the Toolkit as a keyword-to-AI glance. Do not treat its ChatGPT count as the source of truth, because it is modeled.
- Compare configured prices on the rankings. $99 per domain is not like-for-like with a $95 daily log, because the units are different.
- Re-check the same ChatGPT prompts after one page change, so the before and after are the same strings on the same engine.