Topics and AI demand
A topic is what a market asks AI when it chooses, compares, fixes or learns about something in your category: “best payroll software for small businesses”, “how to switch CRM without losing data”. It is the unit Vercatus measures, the unit plays target, and the unit your plan counts. A prompt is one wording; a topic is everyone who asks for the same thing in their own words.
What a topic holds
Section titled “What a topic holds”- A name and a stage. The stage says where the person asking is: choosing, comparing, solving a problem, or learning.
- A market. A country and the language people there ask in, from the site’s markets. The same subject in two markets is two topics.
- Phrasings. The questions the poll asks. They vary the person, the length and the situation rather than rewording one sentence, because real people asking for the same thing barely share words. Your own questions, the prompts your workspace saved and your admitted Search Console searches join the topic they fit, the searches always as questions private to your workspace; see Which questions count.
- AI demand, below.
How topics are built
Section titled “How topics are built”The topic builder works from the buyer’s side, not from your pages. A topic you have no page for is exactly the one you most need to see. It reads:
- what Vercatus knows about the business: what it offers, who it serves, its competitors;
- the site’s markets;
- the Search Console searches that come from buyers (only those; see Which questions count);
- the prompts your workspace saved;
- the searches the engines themselves ran when answering your questions.
A site with no topics gets them built when its plan starts. You can rebuild at any time. A rebuild keeps a topic’s identity, and its history, when it names the same topic again; a topic the rebuild drops is retired, not deleted. You can also add, rename, pause and retire topics by hand. See Topics.
AI demand
Section titled “AI demand”AI demand is the estimated number of AI conversations a month on a topic in its market. Nobody who runs an assistant publishes how often a question is asked, so it is modelled, and the model is labelled so everywhere it appears: “modelled from adoption and search data, not observed prompts”.
It is always a range of likely values, never one number, shown in this form: “about 12,000 to 40,000 AI conversations a month (modelled; grade C; 4 sources)”. The inputs are public or first-party:
- how many people in the market use AI assistants, and how often;
- each engine’s share of that use;
- how much of that use is about this kind of subject;
- the topic’s share within it, from the search volume of the topic’s phrasings and of the searches the engines run for it, with seasonality where it is known.
Every input is listed beside the estimate with its value, its range, its source and its date. Inputs that are assumptions are labelled as assumptions.
Grades
Section titled “Grades”| Grade | Meaning |
|---|---|
| A | Calibrated against first-party AI data from several connected sites, and this site’s own data agrees with the pooled range. |
| B | Calibrated against first-party AI data from several connected sites. |
| C | Modelled from the inputs only, not yet calibrated. |
| D | A market the model has no parameters for, estimated from another market with a wider range. |
Calibration is what narrows the range over time. Connected sites contribute their first-party AI exposure: the AI impressions of Search Console’s uploaded AI report, Bing’s AI citations and assistant fetches from crawler logs. Vercatus compares that exposure with modelled demand times the site’s Answer Share there, and pools the ratio across sites. Only the ratios are pooled, never a site’s data, and nothing is pooled below three sites. AI-assistant sessions from Analytics are shown beside a topic’s estimate for that site only and are never pooled. A topic shows whether it is “calibrated with N sites’ first-party data” or “not yet calibrated”. See Vercatus Intelligence.
Not estimated
Section titled “Not estimated”When no search volume can be found for any of a topic’s words, the topic is not estimated, with the reason. That is shown as a dash, never as a range near zero: a topic nobody searches in a keyword tool can still be asked of assistants every day.
Demand tiers
Section titled “Demand tiers”Each topic also has a tier: high, medium or low. Until most of a site’s topics are graded B or better, the tier comes from where the topic’s search volume falls among the site’s own topics. A grade B or better estimate sets the tier from its midpoint. You can set a tier by hand on the topic.
The tier is what Answer Share weighs topics by, what the poll samples by, and part of how plays are ranked. Once most of a site’s topics are graded B or better, Answer Share weighs topics by their demand estimate instead, and the screen says which weighting it uses.
Coverage of demand
Section titled “Coverage of demand”Coverage of demand is the share of your market’s estimated AI demand that your topics cover. The rest is the searches the engines run for your questions that belong to none of your topics: demand you are not measuring. Demand you are not measuring on the Topics page lists them, largest first, each one a topic you can add.
The share is computed only over searches with a known volume, and it says how many lack one. When too few of them have a figure, it reads “cannot tell yet” rather than a share over a handful of searches.