Finding Audience Prompts for AI Search: Proxies and Tracking Methods

As AI-driven search reshapes how people ask questions online, SEOs must adapt their research methods. Search Engine Land’s recent piece, “Proxies for prompts,” highlights practical ways to discover the prompts audiences use on AI platforms when direct query data is unavailable. The original article outlines several proxy techniques — from People Also Ask (PAA) to userbots, Google Search Console regexes, Perplexity follow-ups and Semrush’s Prompt Research — and explains how each can reveal prompt language and intent. Read the original article on Search Engine Land.

Because many AI tools do not expose the raw prompts users submit, proxies become essential. These methods produce usable insights that inform content strategy, on-page optimization, and topic modeling for AI visibility.

Why proxies matter

AI search answers tend to be conversational and context-rich. If your content doesn’t reflect that phrasing or lack grounding (RAG), it may be omitted as a source. The article paraphrases Mike King’s definition of RAG: “RAG is a mechanism by which a language model can be ‘grounded’ in facts or learn from existing content to produce a more relevant output with a lower likelihood of hallucination.” Cited in the Search Engine Land piece, this idea underscores why being listed as a source matters — not every AI answer requires external sources, but when it does, grounding increases the chance your pages are cited.

Effective proxies you can use today

1. People Also Ask (PAA)

People Also Ask remains a straightforward proxy for prompt phrasing. PAA questions mirror conversational search and frequently resemble AI prompts. To scale discovery, use tools like AlsoAsked (http://alsoasked.com/) to extract related questions and expand topic coverage. Treat PAA results as candidate prompts to test in AI platforms and to seed FAQ-style content on your site.

2. Userbot visits and RAG detection

Userbots (e.g., ChatGPT‑User, Perplexity‑User) ping pages used as sources during Retrieval‑Augmented Generation. These visits often appear in server logs. When you see userbot traffic, pair those pages with keyword and PAA analysis to infer which prompts led to your content being used. If you can access server logs, tag userbot hits and map them back to content themes to prioritize pages for RAG optimization.

3. Long queries in Google Search Console (GSC)

Ziggy Shtrosberg’s long-regex approach isolates extended, instructional queries from GSC that look more like AI prompts than typical keywords. Filter by Desktop Search Appearance and apply the provided regex to surface queries with prompt-like verbs and phrasing. Be careful: some long queries may originate from LLM tracker tools rather than humans, so validate patterns against engagement metrics.

4. Perplexity follow-ups and platform-specific cues

Perplexity’s “Related” follow-up prompts offer a look into how prompts evolve mid-session. These follow-ups can reveal the iterative steps people take when refining requests, which helps shape multi-part content and step-by-step guides. Also note that Perplexity and other platforms may return country-specific follow-ups, so run local checks when relevant.

5. Semrush Prompt Research and topic grouping

Semrush’s AI Visibility Tool groups prompts into topics and surfaces brands, sources and full AI responses — a practical way to avoid tracking thousands of unique prompts individually. Grouping prompts into themes lets you measure topic-level performance, prioritize content updates, and create atomic content that covers the variations AI might require.

How to act on proxy data (actionable steps)

1. Combine signals: cross-reference PAA, userbot hits and GSC long queries before making content changes. Signals that overlap are higher-confidence prompts to target.
2. Use topic clusters: group prompt variations into themes and optimize a hub page/peripheral articles rather than chasing each unique prompt.
3. Harden grounding: add clear, authoritative sources and structured data so your content is more likely to be cited during RAG processes.
4. Filter out noise: look for patterns like high impressions with zero clicks (a red flag for LLM trackers) and deprioritize prompts that show no human engagement.
5. Test directly: run representative prompts in ChatGPT, Perplexity and other AI platforms. Inspect network traffic (look for “search_queries” and “search_prob” in DevTools) or use the ChatGPT search-extractor bookmark script to see background searches and RAG probability for a prompt.

Measurement and reporting

Measure impact by tracking topic visibility and pages cited by userbots. Look for increases in brand mentions, referral traffic from sources that frequently appear in AI answers, and eventual click-throughs. If your AI tracking tool supports prompt grouping or topic-level reporting, use it to report on performance changes over time.

Final thoughts

Proxies aren’t perfect, but they’re the best available method for understanding AI prompt behavior until platforms provide dedicated reporting. By triangulating PAA, userbot visits, GSC long queries, Perplexity follow-ups and tools like Semrush, you can reconstruct many of the prompts your audience uses and optimize content for higher chances of being cited. As the Search Engine Land article concludes, these techniques require constant reevaluation as AI models and usage patterns evolve. Review the original Search Engine Land article for more examples and links.

“RAG is a mechanism by which a language model can be ‘grounded’ in facts or learn from existing content to produce a more relevant output with a lower likelihood of hallucination.” — Mike King, paraphrased in Search Engine Land

Original article: https://searchengineland.com/proxies-for-prompts-466351