How to Use Keyword Research and Topic Volume to Estimate AI Prompt Demand
To estimate AI prompt demand from keyword research, work at the level of topics rather than individual prompts. Group related keywords into topics and add up their search volume. Use question keywords and "People also ask" data to see how buyers phrase things. Then estimate AI demand as a range, using low, middle and high assumptions about how much of your audience asks assistants instead of, or as well as, searching. No assistant publishes prompt counts. The prompt volumes sold by tools are estimates, usually modelled from panels of users. So use topic volume to rank topics against each other, weight them by intent and fit, and check the result against first-party data. That data includes Bing's AI Performance report, your sales conversations and your site search.
Why is keyword research still useful?
People ask assistants about the same needs they search for, though in longer, more conversational form. A topic with large search demand, such as "restaurant payroll", usually has substantial AI demand too. The wording differs, but the underlying need is the same. Keyword tools also give you something prompt data cannot yet give reliably: consistent, comparable volumes across thousands of topics and markets.
The limits are real. Some questions are asked mostly of assistants, such as complex comparisons and "help me decide" questions, and have little search volume. Some searches, such as navigational ones, are rarely asked of assistants at all. So treat search volume as a guide to relative demand, not a count of prompts.
How do you build a topic-level estimate?
- List topics, not keywords: the problems, categories and decisions your buyers research.
- Cluster keywords into each topic with your keyword tool, including long-tail and question keywords.
- Sum monthly search volume for each topic, in each market.
- Collect phrasing from question keywords, "People also ask", autocomplete and your own customer conversations.
- Apply scenario ratios for AI use: low, middle and high multipliers that reflect how much your audience uses assistants for this kind of question.
- Rank topics by the middle scenario, and keep the range visible.
Survey data can guide the ratios. BrightLocal's 2026 survey found 45% of consumers had used AI tools for local business recommendations, up from 6% a year earlier. B2B audiences and technical topics may differ. The point is to agree assumptions openly, not to claim precision.
Can you forecast GEO demand without reliable prompt volume data?
Yes, if you forecast ranges and relative priorities rather than exact prompt counts:
- Relative demand: topic volume tells you that topic A is likely to be asked about far more than topic B.
- Direction: rising search interest and rising AI usage together suggest growing AI demand for a topic.
- Scenarios: low, middle and high AI-use ratios give a range that decision-makers can plan around.
- Validation: compare the forecast with signals you can observe, such as Bing grounding queries, AI referral visits to topic pages, and how often prospects mention AI in sales conversations.
Avoid presenting third-party prompt volumes as facts. They come from samples of users, often through browser panels, and different vendors give very different figures for the same prompt. Why prompt volume data varies so much is covered in whether AI prompt volume data is estimated, unreliable or fabricated.
Should question volume be weighted differently for high-intent and informational prompts?
Yes. A thousand informational questions are usually worth less than a hundred buying questions. A simple priority score:
Priority = topic demand × intent weight × fit
| Intent | Example | Illustrative weight |
|---|---|---|
| Decision | "[Product] vs [competitor] pricing" | 1.0 |
| Shortlist | "Best payroll software for restaurants" | 0.7 |
| Solution | "How do restaurants automate tip payouts?" | 0.3 |
| Problem | "Why is restaurant payroll so complicated?" | 0.1 |
Fit is a 0 to 1 judgement of how well the topic matches what you sell and whom you serve. The weights above are a starting convention. Replace them with your own data where possible, such as the conversion rates of organic landing pages by intent. Keep informational topics in the plan, because they build the understanding and citations that later recommendations depend on. Just do not let their volume dominate the priorities.
How do you research AI SEO without search volume data?
Use sources that show what people ask, or come close:
- Bing Webmaster Tools' AI Performance report, which shows grounding queries: the phrases the AI used when retrieving your content for answers in Copilot and Bing's AI experiences. Since June 2026 it also groups them by intent and topic.
- Google Search Console, for long question-style queries (some of these come from rank trackers and AI visibility tools rather than people), and its generative AI report for pages gaining impressions in AI Overviews and AI Mode.
- Sales and support conversations, which contain the exact questions buyers ask.
- Site search logs, showing what visitors look for once they arrive.
- Reviews, where customers describe the problems they were solving.
- Forums and community discussions in your field, and competitors' FAQ and comparison pages.
Grounding queries are especially valuable. They show how Copilot and Bing's AI features turned questions into searches for content like yours, the closest first-party view of AI demand you can get, though they are a sample and include automated prompts from tracking tools.
How should the estimate be presented?
Decision-makers need to see the assumptions, not just the ranking. A simple table works well, with one row per topic and these columns:
- topic and example questions;
- monthly search volume for the topic cluster, by market;
- low, middle and high AI demand, using the agreed ratios;
- intent and fit;
- priority score;
- supporting evidence, such as Bing grounding queries or sales-call mentions.
Add a short note stating where the ratios came from and how confident you are in them. When real signals arrive, such as AI referral visits to topic pages or grounding queries for new phrasings, update the table rather than starting again. Over a few quarters, the estimate becomes a record of how AI demand in your market is changing, and which assumptions held.
What might an illustrative estimate look like?
Consider an illustrative restaurant payroll company comparing four topics. "Restaurant payroll software" has about 2,400 monthly searches in its cluster, "tip pooling rules" 9,000 and "payroll for food trucks" 400. "Restaurant payroll vs PEO" has 150. With a middle AI-use ratio, tip pooling looks largest. After weighting, the order changes. Tip pooling is informational (0.1) with good fit, which gives 900. Restaurant payroll software is a shortlist topic (0.7), which gives 1,680. Payroll vs PEO is a decision topic (1.0), which gives 150, and food trucks, a shortlist topic with a fit of 0.5, gives 140. Bing grounding queries confirm many long comparison phrasings around restaurant payroll software. The company prioritises the software shortlist topic, keeps tip pooling as a supporting guide, and adds the payroll vs PEO comparison because it is cheap to produce and close to purchase.
What do people ask about estimating AI prompt demand?
Do AI assistants publish prompt volumes?
No. Figures sold by tools are modelled estimates, usually from panels of users.
Is Google search volume a good proxy for ChatGPT demand?
For relative demand between topics, often yes. For absolute prompt counts, no.
How often should estimates be updated?
Quarterly, or when AI usage in your audience changes noticeably.
Which first-party source is most useful?
Bing's grounding queries, where available, because they show how AI systems searched for content like yours.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources