75% of Home Service AI Searches Are Emergency Queries. Audits Miss Them.

Most HVAC and plumbing AI visibility advice -- and most audits, including how we've been running ours until this week -- starts from the same assumption: the customer is researching. They're comparing options. They type "best HVAC company Dallas" or "top plumber Toronto" and evaluate the list.

That describes 25% of homeowners.

The query type driving 75% of AI home service searches

In our June 22, 2026 methodology investigation (`methodology-recs/2026-06-22-incident-framed-b-queries.md`, Scout session 55), we reviewed query behavior data from the 5W HVAC & Plumbing AI Visibility Index 2026, which analyzed 350,000 queries across home service AI interactions.

The central finding: 75% of homeowners typed keyword shorthand identical to Google search -- not deliberate category research. The format was "[emergency situation] [city]," not a comparison prompt. "My AC stopped working, who do I call in Dallas." "Burst pipe emergency plumber near me." "Power outage electrician Toronto tonight."

The trigger is the incident, not the research session.

This sounds obvious in retrospect. People in an emergency don't browse -- they triage. But the visibility implications are significant. The query that 75% of your customers type when they actually need you is structurally different from the query type that standard AI audits measure. We've been scoring businesses on their category-query performance and treating that score as a proxy for overall AI visibility. The 5W data says that proxy covers one quarter of what actually matters.

45% of sessions end after the first response

The same 5W study surfaces a second finding our June 22 rec flags as equally important: 45% of home service AI sessions are single-prompt.

The homeowner asks one question. Gets an answer. Calls someone. Done.

For a home service business, this is a binary gate, not a ranking competition. If your business is in the first AI response to an incident query in your service area, you're in the running. If you're not, there's no second page, no refinement prompt that might surface your name. The customer calls whoever the AI named first.

This is distinct from the research context. A homeowner evaluating kitchen renovation contractors iterates -- multiple prompts, multiple businesses checked, website visits before a decision. An HVAC call at 10pm in July doesn't work that way. The first response is the shortlist.

Why incident and category queries may pull from different source pools

Our June 22 rec identifies a hypothesis -- not yet confirmed in Sourcepull's live query environment, because live validation requires an auth token absent for 38 consecutive sessions -- about why incident framing might surface different businesses than category framing.

The mechanism analysis draws on how each platform retrieves information.

Perplexity runs a real-time web search for every query. An emergency-framed search ("burst pipe emergency plumber near me") may activate different source types than "best plumber Dallas": Yelp's emergency service categories, local news coverage of service outages, 24/7 availability directories. These sources may appear in incident retrieval that don't show up in the general category source pool.

ChatGPT defaults to training data but activates live web search when it detects a query requires recent or localized information. Incident framing -- "tonight," "now," "stopped working" -- signals immediacy in a way category research doesn't. That signal may be enough to push ChatGPT toward live retrieval, which means its citation pool for those queries starts to resemble Perplexity's live results rather than its training cache.

Our June 20, 2026 update to `knowledge/platform-citation-behaviors.md` (session 53) documents the per-platform architecture that makes this plausible: Perplexity's 2-7 day structural fix ingestion vs. ChatGPT's 7-21 day window reflects exactly this distinction -- Perplexity is live-retrieval-first, ChatGPT is training-cache-first with selective live lookup. The question the rec opens is whether emergency framing is a reliable trigger for that live lookup on ChatGPT, and whether the resulting source pool differs materially from the category pool.

We don't know the answer yet. What we do know is that the mechanism is different enough to warrant testing.

What the June 12 citation data already confirms

Before getting to what might differ by query type, it's worth being clear about what doesn't change.

Our June 12, 2026 home services citation investigation (`knowledge/home-services-ai-citation-data.md`, session 45) documented that 87% of independent HVAC and plumbing contractors in the US have zero AI citation share across major platforms. The 1.2% ChatGPT citation rate for local contractor locations was not query-type-specific -- it's the baseline across all query framing.

For most independent contractors, the incident query finding doesn't change the first problem to solve. A business that's invisible on category queries is almost certainly invisible on incident queries too. Entity infrastructure is still the prerequisite: directory presence, consistent NAP across sources, schema markup that confirms the business exists as a stable entity. You cannot appear in Perplexity's live retrieval for an emergency query if you're not in the directories Perplexity indexes.

What the incident query data adds is a layer of diagnostic specificity for contractors who have completed entity work and still score lower than expected. If a business has verified listings, consistent NAP, and schema, but still shows weak citation numbers on category queries -- incident-framed query testing may explain the gap. The business may be positioned as a general services provider rather than an emergency responder, and that positioning distinction matters when 75% of queries are incident-triggered.

What we're changing in how we audit

Based on the June 22 investigation, we're adding incident-framed queries to Sourcepull's standard B-query set for home services categories.

Current B-queries test category research: "best HVAC company [city]," "top plumbers [city]," "electrician [city] reviews." These measure performance when a customer is comparing options. They cover the 25%.

Proposed additions for home services: "my AC stopped working, who do I call in [city]"; "burst pipe emergency plumber [city]"; "roof leak repair [city]." These run alongside the existing category queries -- they don't replace them. The audit output should distinguish between category-query citation results and incident-query citation results, since the citation pools may differ.

If empirical testing confirms the pools are the same businesses, we retire the additional query set and note the hypothesis failed. If the pools differ -- different businesses appearing, different source types retrieved -- the distinction becomes a standard part of what a home services audit measures.

What contractors can do with this now

Even before the empirical picture is settled, the 5W finding has practical implications.

If 75% of home service AI interactions start with an incident, the language in your directory profiles and on your service pages should reflect how incidents are described. "24/7 emergency service," "same-day dispatch," "available nights and weekends" -- these aren't marketing filler. They're the terms that match against the queries triggering the majority of AI searches in your category.

Emergency service categorization in your directory profiles carries specific weight. Yelp's emergency availability settings, HomeAdvisor's "urgent" tags, Google Business Profile's "24 hours" designation -- these are data fields that appear in the source types emergency queries retrieve. A profile missing emergency availability data looks identical to one that doesn't offer emergency service, which is not a position you want to be in when 45% of the sessions that matter end after the first response.

Schema treatment of emergency services deserves the same attention. A `hasOfferCatalog` entry for "HVAC repair" is a weaker match against "my AC stopped working, who do I call" than a named service entry for "emergency AC repair" or "24/7 HVAC emergency response." Specificity in schema naming maps to specificity in how queries are framed.

The single-prompt finding is the argument for urgency. Most local search advice assumes you have multiple opportunities to appear -- show up on the first page, appear in the map pack, be findable if someone digs. The 45% single-prompt rate eliminates that assumption for the emergency context. One shot at the first response. The businesses appearing there have positioned themselves for that query, not just the researching one.

---

A Signal Check at sourcepull.ca shows your current citation rate across ChatGPT, Perplexity, Gemini, and Claude on category queries. Once we roll incident-framed query testing into the full audit, we'll be able to show both sides of the picture. In the meantime, if you're a home service business and your category scores look reasonable but you're still not seeing leads come through AI channels, emergency query positioning is the place to look.