Prompt research
What prompts do buyers ask AI assistants, and how do you find them?
Last reviewed: · By Kraffic.ai
Short answer
Buyers ask AI assistants two broad kinds of prompts: short, search-style phrases and longer conversational questions that describe their situation. Prompts are also either branded, naming a company, or non-branded, asking about a category. Prompt research means collecting the prompts your buyers are likely to use, grouping them by topic and intent, and tracking a stable set across assistants.
Key takeaways
- Prompts range from a few words to several sentences, and the long ones carry details that search keywords never did.
- Non-branded prompts show whether you are discovered; branded prompts show how you are described.
- No AI company publishes its users' prompts, so prompt lists are built from research and inference, not from an official source.
- A good prompt set is specific to your buyers, mixed in length, and kept stable so results are comparable.
- Exact prompt volumes are not available, so prioritize by relevance to buying decisions.
Prompts this guide answers
- “what prompts do buyers ask AI”
- “prompt research for GEO”
- “how to find prompts people ask ChatGPT”
- “branded vs non-branded prompts”
- “How do I figure out what questions my customers are typing into ChatGPT and Gemini when no one publishes that data?”
- “What is the difference between keyword research for SEO and prompt research for generative engine optimization?”
- “AI prompt tracking list examples”
What kinds of prompts do buyers type into AI assistants?
Buyers type everything from two-word phrases to paragraph-length requests. A short prompt looks like a search query; a long one reads like a message to an advisor, with budget, location, constraints and preferences included.
Semrush reported, in its ChatGPT Search study of over 80 million clickstreams from the second half of 2024, that prompts averaged about 23 words without search and about 4.2 words with ChatGPT Search enabled. It also reported that roughly 70% of prompts did not fit classic search intents.
| Type | What it looks like | Illustrative example | What it tells you |
|---|---|---|---|
| Short, non-branded | A few words about a category | best payroll software small business | Whether you are in the default short list |
| Conversational, non-branded | One or more sentences with context | I run a 12-person landscaping company in Ohio and need payroll software that handles seasonal staff. What should I look at? | Whether you fit specific buyer situations |
| Short, branded | Your name plus a question | [your brand] reviews | How you are described |
| Conversational, branded | A comparison or a doubt about you | Is [your brand] a good choice for a company my size, or should I look at alternatives? | Your strengths and weaknesses as the assistant sees them |
Why does the difference between short and conversational prompts matter?
It matters because the two types produce different answers. A short prompt gets a general list. A conversational prompt gets a tailored answer in which the assistant matches brands to the stated needs.
A brand can do well on one and poorly on the other. If your site explains exactly which situations you serve, you give assistants what they need to match you to detailed prompts. Tracking only short phrases would hide that.
What is the difference between branded and non-branded prompts?
A branded prompt contains a company name; a non-branded prompt does not. Non-branded prompts test discovery: does the assistant bring you up when the buyer has not heard of you?
Branded prompts test reputation and accuracy: when someone asks about you directly, is the description correct and fair? Both belong in a prompt set, but most of the list should be non-branded, since including your name makes a mention close to certain and inflates scores.
How do you find the prompts your buyers use?
You find them by combining sources, because there is no official list. No AI company publishes its users' prompts, so every prompt list is a researched estimate of what buyers ask.
- Sales and support conversations: the questions buyers ask your team in their own words.
- Search data: queries from your search console and keyword tools, expanded into full questions.
- Your website: the products, services, use cases and locations you cover.
- Competitor comparisons: the alternatives buyers weigh you against.
- Community sites and reviews: how people describe the problem before they know the solution.
- The assistants themselves: ask what a buyer in a given situation should consider, and note the follow-up questions suggested.
How should you organize a prompt set?
Organize prompts by topic and by buyer stage, then check the mix. Each product or service line should have prompts for people learning about the category, comparing options and checking a specific brand.
Keep the set stable once it is built. If the prompts change at every scan, a change in your score may reflect the new list and not a real change in visibility. Add or retire prompts deliberately and record when.
Group by topic
One group per product, service or use case.
Cover each stage
Learning, comparing and deciding prompts within each group.
Mix lengths
Short phrases and full conversational questions.
Add market context
Include location or market, such as the USA or a state, where your buyers would.
Limit branded prompts
Keep them as a separate, smaller group.
Can you know how many people ask a given prompt?
Not exactly. AI companies do not release prompt-level volume, so any figure for how often a specific prompt is asked is an estimate from a third party, not a count.
A sounder way to prioritize is by closeness to a purchase. A prompt that asks which provider to choose matters more than a general definition, whatever its volume. Search demand for the related topic can serve as a rough guide to interest.
How does Kraffic.ai handle prompt research?
Kraffic.ai reads your website, records business facts for you to confirm, and discovers the prompts buyers type. You can add your own prompts or upload a CSV.
The tracked prompts are then run on Claude, ChatGPT and Google Gemini, and each scan reports which prompts name your brand and which were won or lost since the previous scan.
FAQ
Questions and answers
How is prompt research different from keyword research?
Keyword research uses search engine data to find short phrases and their volumes. Prompt research has no equivalent data source, because AI companies do not publish prompts. It works from customer conversations, search data and category knowledge to build full questions, including long conversational ones, and then tests them on the assistants.
How long are typical AI prompts?
They vary widely. Semrush reported that in the second half of 2024, ChatGPT prompts averaged about 23 words when search was not used and about 4.2 words when ChatGPT Search was enabled. So a realistic prompt set contains both short, search-style phrases and longer questions that describe the buyer's situation.
Should I track prompts that include my brand name?
Yes, but keep them separate and few. Branded prompts tell you how assistants describe you and whether the facts are right. They do not tell you whether buyers discover you, since the name is already in the question. Mixing many branded prompts into your main set will make your visibility score look better than it is.
Is there a tool that shows real ChatGPT search volume?
No tool has access to OpenAI's actual prompt data, because OpenAI does not publish it. Tools that show prompt volumes are presenting estimates built from other data, such as panels or search demand. Treat such numbers as rough indications and ask the provider how they were produced.
Should prompts mention the USA or a specific state?
Include location where your buyers would. Someone looking for a local service or a US-only product often names the place, and assistants adjust their answers to it. For products sold nationally, include a mix: some prompts with a location or market and some without, so you can see whether it changes who is named.
How often should a prompt list be updated?
Review it periodically and when your business changes, such as a new product, a new market or a new competitor. Avoid constant changes, because a stable set is what makes scans comparable. When you do add or remove prompts, note the date so that later shifts in your score can be explained.