Your Service Page Wins One AI Query. Here's Why It Loses the Others.

When a homeowner looks for an HVAC company on ChatGPT, they're not running one search. They're running variations: "emergency furnace repair Denver," "how much does furnace installation cost," "furnace making banging noise fix," "HVAC company near me Denver." Each of those is a structurally different query with different vocabulary, different intent, and different citation criteria.

Most AEO advice treats service page optimization as a single outcome: does your page show up for your service category or not. The research says it's more granular than that. A page optimized to win one query variant frequently underperforms on the others -- not because the other queries are harder, but because the content changes required to serve them conflict with what you've already written.

The multi-query conflict problem

In our July 3, 2026 review of arXiv:2601.13938 -- "IF-GEO: Conflict-Aware Instruction Fusion for Multi-Query Generative Engine Optimization" (Liu et al., January 2026) -- we documented a framework that formalizes something practitioners run into without naming it.

The paper starts from a constraint it calls the "content budget." A service page has limited space for vocabulary, topic emphasis, and evidence structure. When two query variants want contradictory things from that budget -- one wants emergency response language ("weekend availability," "burst pipe," "immediately"), the other wants decision-context language ("cost estimate," "how much," "price range") -- optimizing for one displaces the other. The paper frames this as the "multi-query conflict": heterogeneous queries impose competing revision requirements on a single page under a fixed content budget.

The practical implication from the IF-GEO framework isn't to create more pages. It's to map the distinct vocabulary preferences of your top query variants before editing anything, then design content coverage that serves each without sacrificing the others.

For a home services business, four query types cover most of the buyer surface:

1. **Category queries**: "HVAC company [city]" -- the buyer who hasn't decided what they need yet 2. **Incident queries**: "furnace blowing cold air," "AC stopped working" -- the buyer with a specific symptom 3. **Cost queries**: "how much does furnace installation cost," "HVAC estimate" -- the buyer in decision mode 4. **Urgency queries**: "emergency HVAC [city]," "furnace repair tonight" -- the buyer who needs service now

A page that leads with emergency response language wins urgency queries and incident queries. It typically loses cost queries because it uses no cost vocabulary. It may also lose category queries because it's too narrowly framed. The content budget spent on one type comes at the cost of the others.

Why vocabulary is the mechanism

This matters for AI search specifically because of a finding our research vault has tracked since May 2026. The Discovered Labs study ran regression analysis on AI citation rates with domain fixed-effects controls -- stripping out domain authority to isolate what page-level factors independently predicted citation. One signal survived: vocabulary alignment (beta=+0.37, significance at q approximately 10^-73).

What buyers type into an AI engine needs to match what's on your page. That's the operative mechanism. Not keyword density, not heading count -- proximity between the query vocabulary and the page vocabulary. The multi-query conflict maps directly onto this finding: if your page has strong vocabulary coverage for urgency and incident queries, it has weak vocabulary coverage for cost queries, and vocabulary alignment is the surviving signal.

The practical test is straightforward. Take your four query types. For each, list five words a buyer would use that you wouldn't use in any other context. "Emergency" and "tonight" belong to urgency. "How much," "cost," and "estimate" belong to cost queries. "Noise," "blowing cold," and "not working" belong to incident queries. Now check how many of those vocabulary sets appear in your existing service page content. The absent sets are the citation gaps.

Why Perplexity amplifies the problem

The multi-query conflict hits hardest on Perplexity. In our May 2026 investigation of platform citation behaviors, we found that Perplexity operates closer to literal keyword retrieval than ChatGPT. ChatGPT rewrites queries internally -- expanding intent and tolerating vocabulary distance between what you wrote and what the buyer asked. Perplexity searches closer to the literal query text. Vocabulary gaps between your page and a specific query variant are more damaging on Perplexity.

The Yext 2026 AI Visibility study (17.2M citations across ChatGPT, Perplexity, and Gemini) adds a structural layer to this: Perplexity trusts niche expert directories, not general-purpose ones. For service categories, Perplexity synthesizes from specialized industry resources -- resources that are carefully targeted at specific query types. If your content doesn't align vocabulary to both "emergency" queries and "cost estimation" queries, Perplexity's literal retrieval fails to match your page to whichever vocabulary set you're missing.

ChatGPT's tolerance for vocabulary distance doesn't eliminate the problem -- it softens it. Your citation probability on cost queries is still lower if your page has no cost vocabulary, even if ChatGPT's query rewriting partially bridges the gap.

What to do before editing anything

The IF-GEO framework's core insight is that uncoordinated multi-query optimization creates more conflicts, not fewer. The right sequence is: map first, then write.

For any service page, audit which of your four query types has explicit vocabulary and evidence coverage. "Explicit" means the vocabulary is present, there's at least one H2 that answers that query type directly, and there's at least one specific claim (attributed statistic, timeframe, price range) for that query type. Implicit is not enough. If your HVAC page mentions cost only in a "request a quote" CTA, you have no cost vocabulary and no cost evidence. That's a citation gap on cost queries.

The fix doesn't require new pages. It requires expanding each H2 section to serve one query type explicitly. The urgency section gets urgency vocabulary and a specific response-time claim. The cost section gets pricing vocabulary and at least a range or a "starting at" figure with a source. The incident section gets symptom language. The category section gets the city, the service scope, and a verifiable fact about the business.

Each section should open with a direct answer to that query type's implied question. That's the structure AI retrieval systems extract from -- not from meta descriptions, not from the page title, but from the first sentence beneath each heading.

The pattern in AI visibility audits

When we run Signal Check audits for service businesses with mid-range scores, the most common pattern isn't a low score across the board. It's strong performance on one or two query types and near-zero on others. An HVAC company might score well on urgency queries -- Perplexity and ChatGPT both cite them for "emergency furnace repair" -- and score zero on cost queries because the service pages contain no pricing vocabulary, no cost ranges, and no attributed cost data.

Fixing the zero-score query types doesn't require a site redesign. It requires adding one H2 section and two or three evidence-dense sentences that target the absent vocabulary set. That's the scope of work the IF-GEO framework points toward: not a full content overhaul, but deliberate coverage of the query variants you're currently failing.