Find the growth opportunities worth prioritizing across market demand, the competitive landscape, Brand Value Props, and citation sources.
Busy AI answers don't mean big market demand. Answer counts follow your question set and sampling, and AI likes to list brands. Learn which layers of data to cross-check, and which signals justify a project versus a research lead.
Explore 12,000+ niche markets
Updated by
Updated on Oct 20, 2026
Not necessarily, and most of the time it doesn't. “Busy AI answers” usually mix five signals: rich answers, high prompt volume estimates, answers that change every time, brands mentioned again and again, and AI ads or shopping cards.
The first three mainly reflect how much content exists online, how the estimate was made, and the randomness of the model. The fourth only shows that a brand made it onto AI's shortlist. The fifth shows that someone is willing to pay for these questions. It is the closest to commercial demand, but it still does not equal market size.
To judge whether demand is big, you ultimately have to go back to search data, behavior on your own site, and real customers.
An AI answer is the output for one question. If you ask ChatGPT “Which smartwatches are good for swimming?” and it gives you a long list, that only proves it can answer. The only person asking may be you.
Separate the two kinds of signals and the picture becomes much clearer:
The problem is that the demand side is almost invisible. No major AI platform publishes its prompt logs, and both Ahrefs and Conductor acknowledge this when explaining their methods. So every demand-side number on the market is an estimate or an indirect observation.
Below, we take the signals one at a time and look at what each one can prove.
These five signals are often lumped together as “this market is hot in AI.” The table separates them. The further down the table, the closer the signal is to “someone is paying.”
| Signal | What you see | What it mainly reflects | What it can show | What it can't show |
|---|---|---|---|---|
| 1. Rich answers | Long answers that list many brands and cite many sources | How much public content exists; how mature the category is | AI has enough material to discuss the category | How many people are asking |
| 2. High prompt volume | A tool shows a topic is “asked a lot” | An estimate projected from panels or search data | How popular topics are relative to each other within the same tool | The absolute number of prompts; purchase intent |
| 3. Volatile answers | The same question returns different brands in a different order each time | Probabilistic generation; the size of the candidate pool | Whether the category's candidate pool is wide or narrow | That the market is changing |
| 4. Frequent brand mentions | Some brands appear again and again across many answers | The brand's position in AI's candidate set | The brand's visibility within the chosen question set | Sales, market share, clicks |
| 5. AI ads, shopping cards | Sponsored ads or product cards appear below the answer | Advertisers paying; the platform judging there is shopping intent | Companies are willing to buy exposure for these questions | Market size, conversion rate |
Long answers that list many brands first tell you there is a lot of content about the category online, not that many people are asking.
A study published by SparkToro and Gumshoe in January 2026 supports this. 600 volunteers used 12 prompts and collected 2,961 answers from ChatGPT, Claude, and Google AI. The researchers found that the number of brands AI recommended varied widely from topic to topic, and that this variation closely tracked how many entities exist for that topic in the material AI learns from. For example, Los Angeles has fewer than a dozen Volvo dealerships, while recent science fiction novels number in the thousands.
From this we can infer (this is our judgment, not the study's conclusion): for a consumer category with many reviews and ranking lists, AI will answer in confident detail. For an industrial category with strong purchasing demand but little public content, AI may give only a few lines. How rich an answer is has no necessary link to how big demand is.
How to check: open the citation sources behind the answer and sort them into brand websites, editorial media, community forums, and e-commerce platforms. If the citations are concentrated on a few ranking or affiliate sites, “rich” only means those sites write a lot.
Prompt volume (also called AI search volume) is the metric that looks most like “demand data” today, but it is entirely estimated. When you get a number, first find out which estimation method it comes from:
chatgpt.com and perplexity.ai in your own GA4, plus queries in Search Console. These are real counts, but they only cover the demand that has already reached your website.You also need to check whether these prompts have anything to do with buying. An NBER working paper written by OpenAI with Harvard economist David Deming analyzed 1.5 million ChatGPT conversations. According to Deming's summary, non-work messages made up 73% by mid-2025. “Seeking Information” (including looking up products to buy and recipes) accounted for 21.3% of all messages, below “Practical Guidance” at 28.3% and “Writing” at 28.1%.
This means (our inference): a large share of a topic's prompt volume may come from homework, copywriting, and learning, with nothing to do with buying. High prompt volume shows at most that “people talk about this topic.” It cannot show that “people want to buy.”
How we suggest using it: compare the relative level and trend of topics only within the same tool, and don't compare absolute values across tools. Look at it alongside Google search volume and Google Trends; it is more credible when they point the same way. In internal reports, talk about trends, not precise numbers.
If the same question gets a different answer today than last week, the most common reason is too small a sample, not a changed market.
The SparkToro study mentioned above gives concrete numbers: if you ask the same prompt 100 times, the chance that ChatGPT or Google AI returns exactly the same brand list is under 1%. The chance of the same list in the same order is even lower, about 1 in 1,000. Real users' wording is just as scattered: when 142 participants each wrote a prompt for the same intent, the average semantic similarity was only 0.081.
The study has another side too. With repeated sampling, a brand's appearance rate is relatively stable, and the researchers consider it a reasonable metric. “What rank you hold in AI,” by contrast, is meaningless.
How to check: first fix the market, language, model, question set, and time window, then sample several times and compare appearance rates, not two screenshots. Only changes measured under the same settings are worth following up as market signals.
When a brand appears again and again in AI answers, it has made it into the model's candidate set. That is a different thing from sales or market share.
The same goes for breaking results down by purchase intent: if one type of question makes up a large share of the tracked question set, that tells you which direction your content should take, not how many people are buying.
Even being mentioned does not necessarily bring visits. Pew Research Center analyzed real browsing data from 900 U.S. adults in March 2025, covering 68,879 Google searches. When an AI summary appeared on the page, users clicked a traditional result 8% of the time, compared with 15% when there was no summary. Only 1% clicked a link cited inside the summary. These are Google Search figures and can't be applied directly to ChatGPT, but they are enough to show a clear gap between “being shown by AI” and “bringing visits.”
The first four signals are all generated by AI itself. Ads are different: they are votes companies cast with their budgets.
According to OpenAI's official announcement, ChatGPT began testing ads on February 9, 2026, for logged-in adult users on the Free and Go plans in the U.S. In March, OpenAI announced it would expand the pilot to Canada, Australia, and New Zealand. In August, it announced that ads had launched in the UK, Mexico, Brazil, Japan, and South Korea. Ads are labeled Sponsored, shown separately from the answer, and matched by conversation topic, past chats, and ad interactions. OpenAI says ads do not affect answers and do not appear next to topics such as health, mental health, or politics.
Why this is closer to demand (our judgment): if advertisers are willing to buy exposure on questions like “how to choose a living-room TV,” at least some companies believe there are buyers there. That is far more convincing than “AI answers in a lot of detail.”
Three variables to watch:
Shopping cards are another clue: when a platform treats a question as a shopping question, it usually shows product cards. Which questions trigger product cards, and which competitors take those spots, is also worth recording.
The five signals can be arranged as a ladder. Look for evidence from the bottom up: no amount of low-level signals can replace higher-level evidence.
Evidence ladder: five levels, from “AI can answer” to “demand reaches you”
For a piece of content, reaching L2–L3 is usually enough. To decide whether to enter a new market, you must reach L5. Not every category has L4, and no ads does not mean no demand.
The order below starts with the cheapest checks. The further you go, the closer you get to real demand.
After these five checks, to decide whether to keep researching, run a small pilot, or pause, see the decision table in The Complete Guide to AI Search Market Intelligence. To judge specifically whether buyers have already reached the comparison stage, see Before Entering a New Market: How to Use Public AI Queries to Tell Whether Buyers Are Already Comparing This Type of Product.
The five checks above are not hard to do by hand once. The hard part is keeping the settings consistent every time and being able to look back at the evidence. Dageno's Market Intelligence records AI answers, citation sources, purchase intents, ads, and product cards within the same market, model, region, and time range, which maps directly to L1 through L4 of the evidence ladder. Below, we use the Smart Watches market as an example.
See whose content the answers rely on. AI citation sources and content lists cited domains, citation counts, and the share of responses that cite each one, and labels each domain by type, such as Official Sites, Editorial & News, and Community & Social. In Figure 1, youtube.com is cited in 19.7% of responses, and apple.com ranks 6th at 6.7%. Answers in this category are rich mainly because community and media content supports them.
Figure 1. Dageno cited source leaderboard, Smart Watches market
See which questions the activity concentrates on. AI Search Intents groups questions by purchase intent. In Figure 2, “Recommendations” has a 55.1% demand share. Within it, “Choose for your needs” accounts for 81.2%, while “Compare options & decide” accounts for only 0.3%. This suggests content should be organized around use cases. Demand share describes the tracked questions, not search volume.
Figure 2. Dageno search intent overview, Smart Watches market
See whether anyone is paying. AI Advertising Competition records the advertisers and ad placements observed in ChatGPT Ads and Google AI Overview Ads, and AI Shopping Product Visibility records the products, prices, and merchants in product cards. In Figure 3, across all platforms and all regions, this market has 200 ad placements and 99 advertisers. Whoop appears most often, with 13 ad placements. The number of ad placements does not represent spend or conversions.
Figure 3. Dageno advertising overview, Smart Watches market
Then add prompt popularity and your own data. Prompt Volumes prioritizes topics under the same settings, and Cross-Model and Regional Benchmarking compares differences across models, countries, and languages. After you connect GSC and GA4, your own queries, landing pages, and key events appear side by side with AI data, each keeping its data source label. This fills in L5.
If your team is short on people, Dageno also offers a GEO/AEO managed service covering diagnosis, content, sources, and ongoing delivery. To be clear, Dageno does not estimate market size or total demand (see the market entry research page). It makes every level of evidence reviewable and verifiable. Investment decisions still need first-party data and customer research.
The same “activity” can lead to opposite conclusions under different settings. Before you act, go through these:
No. How rich the answers are depends mainly on how much usable content exists online. To judge popularity, at least look at relative changes in search volume or prompt volume estimates, and ideally your own site data too.
Google search volume is built on Google's own data, so the way it is measured is much more stable. Prompt volume is always an estimate and is suited to comparing how topics rank relative to each other. When the two point the same way, the result is more credible. When they clearly diverge, the reason is worth investigating on its own.
Not necessarily. Estimation methods may undercount some groups, such as mobile and enterprise users. If your target buyers use AI to choose products and competitors already appear in the answers, it is worth prioritizing even if the estimated volume is low.
First confirm three things: which country the ads appear in, which questions they target, and whether the competitor keeps running them. Ads that keep appearing suggest the competitor sees a return, but you still need to decide with your own unit economics.
No. Dageno observes AI answers, citations, ads, and product cards to narrow your research scope and find competitive gaps. Market size and sales potential need independent business research and first-party data.
Busy AI answers are a lead worth following up, not a conclusion you can take straight to project approval. First identify which kind of activity it is, then look for evidence in this order: “AI can answer → buyers ask repeatedly → someone pays → demand reaches you.” That way, a nice-looking AI list won't lead your judgment astray.
To see the AI answers, competitors, citations, and ads in your own market, you can run a free lookup. It covers 12,000+ market segments.
Further reading
All pages were accessed on October 20, 2026.
Updated by
Dageno
Dageno is the research and insights team at Dageno AI, publishing industry reports and expert analysis on AI Search Visibility, Generative Engine Optimization (GEO), and AI-powered search discovery.
Dageno • Oct 19, 2026
Dageno • Sep 28, 2026
Dageno • Sep 28, 2026
Dageno • Oct 08, 2026