Prompt Generation
AI cites the page that answers the buyer's exact question. So start with the real question.
Buyers don't type keywords into AI. They ask long, specific questions, and AI cites the pages that match them most closely. Rocksalt finds those questions where buyers actually ask them: LinkedIn, Reddit, your sales calls and support tickets. It scores each one for market signal and buying intent, then groups them into themes you can own.
Why real questions matter
The question you track decides what gets cited
If you track prompts nobody types, you optimize content nobody's question will retrieve. Two findings from recent research explain why the prompt list is the foundation of everything else in AEO.
Finding 01
Relevance to the question decides who gets cited
ChatGPT retrieves dozens of pages per answer but cites only about half of them. In Ahrefs' study of 1.4 million prompts, the cited pages had titles measurably closer in meaning to what the user asked.1
Title-to-prompt semantic similarity (cosine), Ahrefs, April 2026
Authority isn't the shortcut. Surfer measured roughly 5 million AI citations across AI Mode, AI Overviews, ChatGPT and Perplexity and found domain authority had essentially no correlation with being cited.2
Finding 02
Buyers ask AI long, specific questions
Semrush's analysis of 80 million clickstream records found the average ChatGPT prompt runs 23 words, against 4.2 words when ChatGPT itself searches.3
23
words in the average ChatGPT prompt
4.2
words in a keyword-style search
A generated prompt
“best AI content tools for LinkedIn”
A question a buyer actually asked
“How do I build a workflow that turns the raw inputs I already have — voice memos, bullets, client conversations — into finished LinkedIn posts, keeping the source note visible beside each draft, instead of generating generic content from nothing?”
Optimize for the first and you can still miss the second. The second is what your buyer typed.
How Prompt Generation works
From what buyers say to the prompts worth winning
01
Start where buyers actually talk
Rocksalt reads the questions buyers ask in public on LinkedIn and Reddit, and in private on your sales calls and support tickets.
Public channels show what the market is asking. Internal sources show what your own buyers ask right before they buy.
02
Keep only the questions the market responded to
Not every post is a signal. On Reddit and LinkedIn, a question only makes the cut if it earns more engagement than posts in that space typically get. One-off chatters fall away. What's left are the questions other buyers recognized as their own.
03
Prioritize by market signal and by money
Two scores tell you what to win first.
Market signal (1–5). A question that shows up across several sources, or again and again in your calls, is stronger evidence than one heard once in one channel.
Recommendation prompts. Money questions, where a buyer asks AI to recommend or compare vendors, get answers that name companies. Early research questions get answers that cite sources. Both matter; the money questions come first.
04
See themes, not just a pile of prompts
Rocksalt clusters related questions into themes, so you can see the patterns in what your market cares about. Your content and AEO strategy runs at the level of themes you want to own, instead of chasing one tactical prompt at a time.
You can also add themes that you want to be known for. Rocksalt starts tracking questions related to that topic.
Generated prompt lists vs. Rocksalt Prompt Generation
| Typical AEO tool | Rocksalt | |
|---|---|---|
| Where prompts come from | An LLM reads your website, or prompts are built from keyword volume | Real questions from LinkedIn, Reddit, sales calls and support tickets |
| What they look like | Short, keyword-style | Long and specific, in the buyer's own words |
| Quality filter | Volume for keywords | Public questions must beat typical engagement for their community |
| Prioritization | Every prompt treated equally | Market signal plus recommendation (money) prompts first |
| Evidence | Hard to trace | Every question linked to its source |
| Freshness | Set once, ages quickly | Continuously running; new questions surface as buyers ask them |
“I knew our prompts weren't realistic but didn't know how to correct that. Rocksalt showed me what buyers actually ask.”
CMO, 500-person company
Questions about Prompt Generation
How is this different from the prompts in my AEO tracking tool?
Most tracking tools generate prompts by having an LLM read your website or by converting keywords into questions. Rocksalt starts from questions buyers have actually asked, so you track what your market types, not what a model imagined.
Which sources does Rocksalt read?
Public conversations on LinkedIn and Reddit, plus your internal sources such as sales call transcripts and support tickets.
What does market signal mean?
It scores how strong the evidence is for a theme, from 1 to 5. Questions seen across multiple channels, or repeatedly in your calls, score higher than questions seen once in one place.
What is a recommendation prompt?
A question where the buyer asks AI to recommend or compare vendors. AI answers to these name specific companies, which makes them the highest-value prompts to win.
Can I track a topic Rocksalt didn't surface?
Yes. Add any theme you want to be known for and Rocksalt scans all connected sources for real buyer questions about it.
What happens after I track a theme?
Tracked themes stay permanently. They drive your AI visibility tracking and feed content creation, where Rocksalt drafts original content from what your company knows.
Stop guessing prompts. Start with what buyers ask.
See the buyer questions Rocksalt finds for your company, and which ones are worth winning first.
Sources
- Ahrefs, Louise Linehan, “Why ChatGPT Cites One Page Over Another (Study of 1.4M Prompts),” April 2026. ahrefs.com
- Surfer, Joshua Hardwick, “We Analyzed 5 Million AI Citations. Domain Authority Didn't Predict Any of Them.” surferseo.com
- Semrush, “New Semrush Study Reveals ChatGPT Search Trends: Insights from 80 Million Clickstream Records,” February 2025. semrush.com
- Kevin Indig, Growth Memo, “Growth Intelligence Brief #17,” April 2026 (citing Semrush clickstream data). growth-memo.com
- Kevin Indig, Growth Memo, “Growth Intelligence Brief #16” (citing AirOps analysis of 15,000 prompts). growth-memo.com