Why keyword research no longer tells the whole story
For the best part of two decades, keyword research meant one thing: find out what people type into Google, optimise your pages around those phrases, and earn traffic. Tools like Ahrefs, SEMrush, and Google Keyword Planner were the backbone of every digital strategy. They still matter. But they were built for a world where users type two or three words into a search box and scan a list of ten blue links.
That world is changing fast. ChatGPT crossed 100 million weekly active users within two months of launch. Perplexity now handles hundreds of millions of queries per month. Google's AI Overviews appear on a growing share of searches. People are not just searching differently; they are getting their answers differently. And the way they phrase their questions in these AI tools is nothing like a traditional search query.
This is where prompt research comes in. It is the practice of identifying the specific, conversational questions and requests that real people type into AI engines, and then structuring your content and data so those engines recommend your brand in response. Think of it as keyword research's more articulate sibling.
How prompts differ from keywords
The difference is not just length, though prompts are usually longer. It is intent, specificity, and structure.
A traditional keyword might be: best running shoes. A prompt in ChatGPT might be: I run 40 miles a week on roads and I have mild overpronation. What running shoes would you recommend under £150?
That prompt contains constraints, personal context, a budget, and an implicit preference for a direct answer rather than a list of links. The AI engine has to synthesise across many sources and produce a confident recommendation. For your brand to appear in that recommendation, you cannot simply rank for "best running shoes." You need your site to clearly communicate that your products match those specific criteria, in a format the AI can parse and trust.
Prompts also tend to follow certain patterns depending on the type of AI tool:
- Comparison prompts: "What's the difference between X and Y?"
- Recommendation prompts: "Which [product/service] is best for [specific situation]?"
- How-to prompts: "How do I [accomplish a task] without [common constraint]?"
- Validation prompts: "Is [brand] a reputable company for [product category]?"
- Shortlist prompts: "Give me three [product type] options under [price]."
Each pattern requires a different kind of content response. Recognising which patterns apply to your category is the first step in prompt research.
How to actually conduct prompt research
There is no single tool that does for prompts what Ahrefs does for keywords. That will change. For now, prompt research is a mix of direct testing, customer insight, and structured observation.
1. Use the AI tools themselves
Start by opening ChatGPT, Perplexity, and Gemini and typing in queries related to your product or service. Do not just test one phrasing. Try ten variations with different contexts and constraints. Notice which brands they cite, which they skip, and what kind of content or data they seem to draw from. Pay attention to citations in Perplexity especially, as it shows its sources clearly.
This gives you a direct view of what the AI considers authoritative for your category. If a competitor appears consistently and you do not, that is your gap analysis right there.
2. Mine your customer service and sales conversations
Your customers are already telling you what they want to know. Go through live chat logs, support tickets, sales call notes, and review responses. Look for the longer, contextual questions. These are almost word-for-word the prompts people will type into AI tools. A customer asking "do your protein bars contain soy? I'm lactose intolerant but not vegan" is showing you a real-world prompt pattern.
3. Use Reddit and Quora for natural language discovery
Reddit in particular is goldmine territory for this. Search your product category on Reddit and read the threads. People write in full, conversational sentences there. They describe their situation, their constraints, and their questions in the same way they would ask an AI. The language is unfiltered and specific.
4. Check "People Also Ask" for transitional insight
Google's People Also Ask boxes contain questions in near-prompt format. They are longer and more specific than keywords, and they reflect how conversational search is already evolving within Google itself. These are a useful bridge between traditional keyword research and full prompt research.
5. Run a systematic prompt audit of your brand
Ask ten to twenty prompts across different AI tools specifically about your brand. Things like "Is [Brand] a good option for [use case]?" or "What do people say about [Brand]?" The answers reveal how AI engines currently characterise you, and whether that characterisation is accurate, favourable, and specific enough to drive recommendations.
At FlinnSchema, this kind of audit forms part of what we do when assessing a brand's AI visibility. You can request a free AI visibility audit to see exactly how you are being described and cited across AI engines today.
Turning prompt research into content strategy
Once you have a list of prompts relevant to your category, you need to map them to content and structured data. This is where most brands miss the step. They do the research and then write a blog post. That is not enough on its own.
Match your content to prompt intent
Each prompt type needs a corresponding content approach. Comparison prompts need clear, factual comparison content, ideally with structured tables. Recommendation prompts need content that explicitly names who the product is for, under what conditions, and why. Validation prompts need trust signals: reviews, credentials, return policies, and brand signals that AI engines can read and cite.
Specificity is everything here. Vague content does not get cited. If your FAQ page says "our products are suitable for most dietary requirements," that will not surface for the prompt "I'm diabetic and looking for a low-GI snack bar." But if your page states "suitable for diabetics, with a GI rating under 35," it has a real chance.
Use structured data to make your answers machine-readable
This is the part most content strategists overlook completely. AI engines do not just read prose. They index structured data to understand what you sell, who you are, what your reviews say, and how your products compare. Schema markup in JSON-LD format signals this information in a standardised way that AI crawlers can parse efficiently.
For example, if your prompt research reveals that customers are asking AI tools for product recommendations by dietary requirement, then adding NutritionInformation schema to your product pages directly maps your product attributes to those recommendation queries.
Similarly, if validation prompts about your brand are common ("is this company trustworthy?"), then making sure your site uses proper Brand, AggregateRating, and MerchantReturnPolicy schema gives AI engines the verified data they need to speak confidently about you. You can read more about how AggregateRating schema influences AI recommendations for a practical breakdown of that specific implementation.
Write content in answer format
AI engines prefer content that answers questions directly, quickly, and without a lot of preamble. A paragraph that starts with a direct answer and then expands on it is far more citable than one that buries the answer at the end. This is sometimes called the "inverted pyramid" structure, and it applies more forcefully to AI-targeted content than almost anywhere else.
Each piece of content you create should be able to answer a specific prompt in its first one or two sentences. Test this yourself: paste the prompt into ChatGPT and see whether your content would be a satisfying answer if summarised into two sentences.
Measuring whether your prompt research is working
One of the harder aspects of AI search optimisation is measurement. There is no Google Search Console for ChatGPT, at least not yet. But there are practical proxies.
First, track your brand's share of voice in AI answers. This means systematically testing a defined set of prompts across multiple AI tools on a regular basis and recording whether your brand is cited. It sounds manual, but it is manageable with a spreadsheet and a consistent testing cadence. Our post on how to measure your brand's share of voice in AI answers walks through a repeatable framework for doing this.
Second, watch for referral traffic from Perplexity and other AI tools in your analytics. Perplexity and some Bing-powered AI features do send referral traffic, and you can segment this in GA4 to see whether it is growing.
Third, monitor whether your brand is being described accurately and favourably when it is cited. AI tools sometimes get details wrong or use outdated information. Structured data and regularly updated content are the main tools for correcting this.
The overlap with traditional SEO is real, but limited
Prompt research does not replace keyword research. Many of the same pages that rank in Google will also be cited by AI engines. Domain authority, backlinks, and content depth still matter. But the emphasis shifts in important ways.
Traditional SEO rewards pages that earn clicks. AI search rewards pages that provide clear, trustworthy, specific answers. A page with a high click-through rate but vague, fluffy content may rank well in Google and still be ignored by AI engines. The two scoring systems are increasingly divergent.
Brands that treat prompt research as a parallel workstream alongside traditional keyword research, rather than a replacement for it, will be best positioned. The goal is to appear in both environments, and the content requirements are different enough that you genuinely need to think about each separately.
If you want help identifying which prompts your customers are using and how your current site performs against them, the free AI visibility audit is a good starting point.
Frequently Asked Questions
What is prompt research and how is it different from keyword research?
Prompt research is the practice of identifying the specific, conversational questions and requests that people type into AI tools like ChatGPT, Perplexity, and Gemini. Unlike traditional keyword research, which focuses on short search phrases for Google, prompt research deals with longer, context-rich questions. The goal is to optimise your content and structured data so AI engines recommend your brand in response to those prompts.
Do I need special tools to do prompt research?
Not at this stage. The most effective prompt research currently involves directly testing AI tools with relevant queries, reviewing customer service conversations for natural language patterns, mining Reddit and Quora threads, and running a systematic brand audit across AI engines. Dedicated prompt research tools are emerging, but the manual approach is often more insightful anyway because it keeps you close to how real customers actually speak.
How does structured data help with AI search visibility?
Structured data, particularly schema markup in JSON-LD format, gives AI engines a machine-readable layer of information about your products, brand, reviews, and policies. Rather than inferring this information from prose, AI crawlers can read it directly and with confidence. This makes your content far more citable in response to specific prompts, especially recommendation and validation queries where the AI needs to verify facts quickly.
How long does it take to see results from prompt research optimisation?
It varies, but most brands see meaningful shifts in AI citation frequency within six to twelve weeks of implementing structured data and updating content to match prompt intent. The timeline depends on how often AI engines re-crawl your site, how competitive your category is, and how thoroughly the changes have been applied. Tracking your brand's share of voice in AI answers on a fortnightly basis is the most reliable way to measure progress.