Prompt Research vs Keyword Research: What Changes When Customers Ask AI
Keyword research finds what people type into Google. Prompt research finds what they ask AI assistants. How the two differ and how to use them together.
· 4 min read
On this page
Keyword research finds the short phrases people type into Google and how often they search them. Prompt research finds the full questions people ask AI assistants, and, more importantly, which brands those assistants recommend in reply. Keywords tell you where demand is. Prompts tell you who is winning the answer. For AI search you need both.
Key takeaways
- Prompts are longer, conversational and full of context such as location and budget.
- There is no official prompt volume. Use keyword data and customer questions as proxies.
- The output of prompt research is not a ranking but a list of brands named and sources cited.
- Turn each prompt you lose into a fix: on-site content, facts or third-party mentions.
How prompts differ from keywords
| Keyword (Google) | Prompt (AI assistant) | |
|---|---|---|
| Typical form | “crm small business” | “What’s the best CRM for a 5-person agency that already uses Gmail?” |
| Length | 2 to 4 words | 10 to 30 words, often with follow-ups |
| Context | Little | Budget, size, location, tools used, preferences |
| Result | A ranked list of pages | A written answer naming a few brands |
| Data available | Search volume, difficulty, CPC | No public volume; you see answers, brands and sources |
| What you measure | Your ranking position | Whether you are named, where, and next to whom |
Why keyword research alone misses AI search
- The answer is the result. In Google you can rank tenth and still get some clicks. In an AI answer that names three businesses, you are either in or out.
- Context changes the answer. “Best CRM” and “best CRM for a small agency using Gmail” can produce completely different lists. Keyword tools group these together.
- Third-party pages decide a lot. AI engines often build answers from reviews, directories and comparison articles. Keyword rankings for your own pages do not show that.
Where keyword research still helps
- Sizing demand. Keyword volume is the best public signal for which topics matter.
- Finding the topic behind the prompt. High-volume keywords often map to whole families of prompts.
- Matching how engines search. When an AI engine searches the web, it runs queries that often look like keywords. Pages that rank for those queries are more likely to be retrieved.
For how classic SEO, GEO and AEO fit together, see GEO vs SEO vs AEO.
A combined workflow: from keywords to prompts
- Start with your money keywords. Take 5 to 10 keywords that bring paying customers, such as “emergency plumber Leeds”.
- Turn each into 2 or 3 prompts. Write them the way a customer would ask an assistant: “My boiler is leaking on a Sunday, who can come today in Leeds?” Add the context your customers really have.
- Add questions from real conversations. Sales calls, support tickets, reviews and FAQs are full of exact phrasings. Our guide to finding the questions customers ask ChatGPT covers the sources.
- Ask each prompt in several engines, several times. Record whether you are named, your position, the competitors named and the cited sources.
- Prioritise the gaps. High-intent prompts where competitors are named and you are not come first.
- Map each gap to a fix. Missing content on your site, unclear facts, or missing from the sources the engine cites.
Example: one keyword becomes eight prompts
Take the keyword “payroll software small business”. In an AI assistant, the same need shows up as many different questions, and each can produce a different list of brands:
| Prompt | Intent |
|---|---|
| What’s the best payroll software for a small business with 10 employees? | Best-of, size |
| Cheapest payroll software for a UK small business? | Best-of, price, country |
| Payroll software that works with Xero? | Integration |
| Do I need payroll software if I only have 2 employees? | Problem, early stage |
| Is [competitor] good for small businesses? | Competitor evaluation |
| Alternatives to [competitor] payroll | Switching |
| [Your brand] vs [competitor] for a restaurant | Comparison, industry |
| What is [your brand] and how much does it cost? | Brand, accuracy |
A keyword tool would treat most of these as one topic. AI answers treat them as separate questions, so you might be recommended for the Xero integration prompt and missing from the price prompt. That is exactly the detail prompt research is for.
Common prompt research mistakes
- Only tracking your brand name. Most new customers ask unbranded questions.
- Writing prompts like keywords. “crm agency” is not how people talk to assistants.
- Trusting invented volumes. No AI company publishes prompt volumes. Treat any exact figure as an estimate.
- Checking one engine. Results differ a lot between ChatGPT, Perplexity, Gemini and Claude.
- Drawing conclusions from one answer. Collect several before calculating a mention rate.
What good prompt research output looks like
For each prompt you should end up with: the engines checked, how many answers were collected, how often you were named, your typical position, the competitors named and the pages cited. A mention rate from one answer is not reliable; wait for at least three before drawing conclusions. To understand how engines assemble those lists in the first place, read how AI builds “best X for Y” recommendations.
Onxeera’s AI prompt research tool does this for you: it asks ChatGPT, Perplexity, Gemini and Claude through their official APIs with web search on, stores every answer with the brands and sources, and ranks prompts by opportunity so you know which gaps to close first.
Frequently asked questions
Is there search volume data for AI prompts?
Not in the way there is for Google keywords. AI companies do not publish prompt volumes, and any tool claiming exact ChatGPT prompt volumes is estimating. Use keyword volume, sales conversations and customer questions as proxies for which prompts matter.
Do I still need keyword research if I do prompt research?
Yes. Google search is still a major source of traffic, and keyword data is still the best demand signal available. AI engines also often search the web using queries similar to keywords. Use keyword research to size demand and prompt research to see who AI recommends.
How many prompts should I track?
Start with 10 to 20 that reflect real buying questions in your market, across local, comparison, best-of and brand questions. Add more once you know which ones you win and lose.
Should prompts be long or short?
Use the way your customers actually ask. AI prompts tend to be longer and more conversational than keywords, often with context like location, budget or company size. Track a few short and a few detailed versions of your most important questions.
The Onxeera team builds tools that help businesses get found, understood and recommended by AI search engines like ChatGPT, Perplexity, Gemini and Google AI Overviews.
See who AI recommends for your customers' questions
Ask ChatGPT, Perplexity, Gemini and Claude the questions your customers ask, and see who gets named.