site: and inurl: still work. They just don't matter as much. The research methods that defined SEO for fifteen years are being replaced by prompt mining inside ChatGPT, Claude, Perplexity, and Gemini — because that's where the buyer is doing the research now. Ahrefs founder Dmitry Gerasimenko, Semrush CEO Oleg Shchegolev, Surfer's Sławek Czajkowski, and the founders of the newer GEO tools — Profound, Otterly, AthenaHQ — have all built around the same shift.
The operator era
Google search operators were the SEO researcher's instrument set since Danny Sullivan was running Search Engine Land. site: for index inventory. inurl: for URL pattern matching. intitle: for title-tag prevalence. allintext: for content presence. cache: for indexed-version inspection — retired by Google in January 2024. related: — also discontinued. link: — broken years before that. info: — gone. Each operator answered a specific structural question about how Google had ingested and ranked the web.
The operators are still useful for diagnostic work. They are not useful for understanding what a buyer is going to see when they research a product or a brand in 2026. The buyer is not running a site: query. The buyer is asking Claude which CRM to pick. The buyer is asking ChatGPT which protein powder is cleanest. The buyer is asking Perplexity which hotel in Tokyo has the best omakase counter.
What still works
site: is the most durable operator. Use it to verify indexation and to find pages that should not be indexed — a problem that Lily Ray, Aleyda Solis, Marie Haynes, and Cyrus Shepard each cover in their public SEO writing whenever a client surfaces it. inurl: finds slug patterns and parameter pollution. intitle: surfaces title-tag stuffing or duplication. Quoted strings still force exact-match retrieval where the algorithm allows it, though looser matching has crept in even on quoted queries. filetype:pdf and filetype:doc continue to surface lead-magnet content that wasn't supposed to be public. Most other operators have been deprecated or weakened to the point of unreliability. Treat them as legacy tools, not primary research.
The prompt-mining method
Prompt mining is the systematic process of testing how AI engines respond to category queries — and recording the answers. The method has four parts. Build a prompt set covering the queries a buyer would actually run, drawn from Search Console queries, sales-call transcripts, customer-support tickets, Reddit threads, and the long tail of how-questions that map to your category. Execute the prompts across multiple engines on a fixed cadence — daily for high-stakes categories, weekly for most B2B SaaS, monthly minimum for industrial and B2B niche. Parse responses for cited brands, cited sources, sentiment, and rank position when the engine returns a list. Track the change over time as a time-series, not a snapshot.
The data structure that comes out of prompt mining is similar to a SERP report but richer. It captures source-publication citations alongside brand mentions, which traditional SEO never quantified because the Google SERP did not display sources the same way an AI engine does. A buyer query for 'best CRM for small business' on ChatGPT will name HubSpot, Pipedrive, Salesforce Essentials, Zoho, and Freshsales — and will cite specific source publications including G2, Forbes Advisor, Capterra, and TechRadar. Each cited publication is a leverage point. Each cited brand is a competitor. The grid that emerges is a research artifact no operator-era tool produces.
Engine-by-engine differences
ChatGPT pulls heavily from Bing's index plus its own training data plus selected real-time sources surfaced through web search. The Bing index dependency means BingBot crawl health matters for ChatGPT visibility, not just Googlebot. Claude tends to cite primary sources more often when given the opportunity, qualifies its answers more aggressively, and shows differential treatment of news sources versus brand-owned content. Perplexity surfaces sources visibly in every answer and is the most transparent about its retrieval; the same query run twice on Perplexity sometimes returns different source sets, which is itself diagnostic data. Gemini integrates Google Search results and Google's Knowledge Graph; the source pattern looks Google-organic-adjacent. Google AI Overviews are derived from a different retrieval layer than standard Google organic — the ranking signals are related but not identical, as Lily Ray's documentation of AI Overview citation patterns through 2024 and 2025 made clear.
A brand can be cited heavily on ChatGPT and barely visible on Perplexity. The reverse happens too. The Notion-Linear-Asana-Trello example: on ChatGPT, Notion at 42%; on Claude, 51%; on Perplexity, 28%. Engine-by-engine asymmetry is the single most important structural finding from any honest prompt-mining program.
The translation table
Old operator question to new prompt question. Where is my brand indexed — replaced by which engines cite my brand. What pages rank for this term — replaced by which sources are cited when a buyer asks this question. How does Google see my title tags — replaced by how does an AI engine summarize my brand in a sentence. What is my competitor's content footprint — replaced by which brands does the engine consider competitive with mine. Where do I have broken backlinks — replaced by which authoritative sources fail to cite me when the engine retrieves on my category. Each shift is a substitution, not a deprecation. The operators still answer the technical-SEO question. The prompts answer the buyer-discovery question.
The 2026 research workflow
A modern discovery research workflow runs three layers in parallel. Google Trends, Search Console, Ahrefs, and Semrush for residual organic-search demand and competitive backlink analysis. Operator-based site audits — Screaming Frog under Dan Sharp, Sitebulb, Lumar (the rebranded DeepCrawl), Ahrefs Site Audit — for technical indexation issues. Prompt mining for AI engine citation patterns through Profound, Otterly, AthenaHQ, Daydream, or an in-house framework. The outputs feed three different teams — SEO, GEO, and PR — and converge in a single content and earned-media plan.
Most agencies still run one layer. The buyers see all three. The brand showing up cleanly across Google organic, AI Overviews, and ChatGPT is winning a different competition than the brand showing up only on Google. Operators were the language of the old web. Prompts are the language of the new one. Build the prompt set before the planning cycle, not after.
Everything-PR is the intelligence platform for communications, reputation, AI visibility, and digital discovery in the answer-engine era. Thirty-plus publications. Publishing since 2009. Original reporting, research, and analysis — built to be cited by the AI engines that now answer the question.
