A friend who runs an HVAC business in Phoenix called me in March. He had been on the first page of Google for “HVAC repair Phoenix” for two years, ran a 4.7-star Google Business Profile with 340 reviews, and could not figure out why his phone had gone quiet. He tested it himself. He asked ChatGPT, “who should I call for HVAC repair in Phoenix,” and ChatGPT recommended three competitors. None of them outranked him on Google. All three of them had press coverage he did not have. That conversation is the cleanest illustration I have of how AI search ranks service providers in 2026.
The five trust signals below are what separates the recommended HVAC company in Phoenix from the also-ran. They apply to every category of service business: legal, dental, accounting, home services, agencies, financial advisors, mental health practices, veterinary, IT consulting, the whole list. The signals are weighted differently across the engines (ChatGPT leans review-heavy, Perplexity leans citation-heavy, Claude leans schema-heavy), but the five are present in every engine’s source-selection logic.
Signal 1: distributed review presence, not just Google
The single biggest mistake service businesses make is concentrating reviews on Google Business Profile while ignoring the secondary platforms. AI engines weight a distributed review presence higher than a concentrated one, because the distribution itself is a trust signal. A business with 50 reviews each on Google, Yelp, and an industry-specific site (Houzz for contractors, Avvo for lawyers, Healthgrades for doctors) reads as “established across the category.” A business with 340 reviews on Google and zero elsewhere reads as “either review-pumping or only optimized for one channel.”
The Phoenix HVAC case was textbook. His Google reviews were strong, but he had no Yelp presence, no Angi listing, no BBB profile, and no presence on the regional contractor directories the engine pulled from when constructing the recommendation. The three competitors that won the citation had 30 to 80 reviews each on Google but also had active profiles on three to four additional platforms with 5 to 20 reviews each. The engine read that pattern as “category-established” and used it as one of the source-selection inputs.
The practical move is to identify the three to five review platforms that matter in your category, claim profiles on each, and route a portion of your review-request flow to each platform rather than funneling everything to Google. Distribution is the signal. Aiming for 100 reviews on Google after you have hit 50 produces less AEO lift than aiming for 15 reviews each on three other platforms.
Signal 2: independent press mentions across at least three publications
The second signal is third-party press coverage, and this is the one that most service businesses underinvest in because the ROI looks abstract until the recommendation starts citing it. AI engines weight a brand mention in a local newspaper, an industry trade publication, or a regional business journal as roughly 10 to 30x the trust value of a mention on a self-published page. The mechanism is simple. The engine has been trained on the convention that named-publication coverage indicates editorial vetting, and editorial vetting is the closest proxy the engine has for “this business is real and the work is acceptable.”
The threshold matters. One press mention is a fluke. Two is a coincidence. Three or more across different publications is a pattern, and AI engines read patterns. The Phoenix HVAC competitors each had between four and nine press mentions over the prior 24 months, ranging from local TV news segments about energy efficiency to trade pub features about a specific install. He had two. The gap was the recommendation.
Building this signal does not require a national press push. Local newspapers, regional business journals, industry trade publications, and niche blogs are the right targets for service businesses. The angle is what matters: a service-business press story works when it is framed around community impact, seasonal expertise, regulatory shifts in the trade, or a specific case the business handled. “Local HVAC contractor explains the heat dome” is a publishable story. “Local HVAC contractor offers great service” is not.
Signal 3: structured schema on every service page
The third signal is technical and unglamorous, but the impact-per-hour-of-work ratio is the highest of the five. Service businesses with LocalBusiness, Service, Offer, and Review schema implemented across their site got cited at roughly 3x the rate of competitors with no schema, holding the other four signals constant. The schema does not change what the page says. It changes how the AI engine reads the page.
The pattern that matters is which schema combinations correlate with citation. The minimum viable set is LocalBusiness on the homepage, Service schema on each individual service page (with sub-properties for name, description, area served, and price range), and Review schema embedded on the page that hosts testimonials. The next-level set adds FAQPage schema on every page that has a FAQ block, and BreadcrumbList on every category page.
The single largest schema mistake is implementing schema once on the homepage and skipping it on the service pages. AI engines rarely cite the homepage of a service business. They cite the specific service page that matches the user’s query. If that page does not have its own schema, the engine often skips it for a competitor whose page does.
Signal 4: explicit service-area definition with city-by-city pages
The fourth signal is geographic clarity. AI engines recommend service providers when they can match the user’s location to a defined service area with confidence. The cleanest way to provide that confidence is a per-city or per-region page for each service area the business covers, with the city name in the URL, the H1, the meta description, and the schema’s areaServed field. A service business that covers Phoenix, Scottsdale, Tempe, Mesa, and Chandler should have five distinct city pages, not one “we serve the Phoenix metro area” page.
The risk people raise is doorway-page penalties. The risk is real if the pages are thin and duplicative. The mitigation is to make each city page substantively different: local case studies, location-specific pricing if it varies, regional regulatory notes, neighborhood-level details, photos of completed work in that city. A city page that lists six specific projects completed in Scottsdale with addresses (or named clients with permission) reads as substantive. A city page that swaps “Phoenix” for “Scottsdale” in template text reads as spam and gets filtered.
The Phoenix HVAC case illustrated this too. He had one “Areas We Serve” page listing 14 cities. His competitors had 5 to 12 individual city pages each, with photos, project lists, and city-specific testimonials. When a user asked ChatGPT for “HVAC repair in Scottsdale,” the engine matched to the dedicated Scottsdale pages and skipped his bundle page. The fix took him three weeks of writing and yielded a measurable lift in branded mentions across the engines within 60 days.
Signal 5: a citation footprint in industry-trusted sources
The fifth signal is the hardest to build and the most durable once built. AI engines weight citations from sources the engine considers authoritative more than citations from anywhere else. For service businesses, the relevant authoritative sources vary by category: trade association membership pages, state licensing board verification pages, accredited contractor directories, professional society rosters, hospital affiliations, bar association profiles, accountant certification listings, and the academic or government datasets that aggregate industry providers.
A business listed on the state licensing board’s website, an industry trade association’s accredited member directory, and a major insurance provider’s preferred-vendor list reads to the engine as “verified by structures the engine already trusts.” The trust transfers from the source to the business. Two or three such citations often outweighs a dozen reviews when the engine is breaking a tie between recommendations.
The work of building this signal is manual and slow. It involves filling out applications, paying membership fees, getting verified, and waiting for directory listings to crawl. Most service businesses give up halfway because the per-listing impact is invisible until the cumulative effect crosses a threshold. The threshold is real, though. The Phoenix HVAC competitors averaged 4 to 7 authoritative citations each. He had one. After eight months of building citations on the four directories that matter in HVAC (ACCA, NATE, BBB, and the Arizona ROC), his AI citation rate moved from “occasional” to “consistent.”
What changes when these five signals are in place
The recommendation engine treats service businesses as a portfolio of signals rather than as a single profile. No individual signal wins the recommendation. The combination does. A service business with strong reviews and weak press will lose to a competitor with moderate reviews and four press mentions. A business with strong schema and weak service-area definition will lose to a competitor with moderate schema and clear city pages. The pattern that wins is “balanced across all five, strong in at least two.”
The compounding kicks in around month four for most service businesses that work all five signals in parallel. Reviews accumulate, press mentions land and get indexed, schema is in place, city pages start ranking, and the citation footprint reaches threshold density. The engines re-evaluate their source pool as the new signals show up, and the recommendation pattern shifts. The Phoenix HVAC business got his next ChatGPT citation in month six. By month nine, he was the default recommendation for two of the five city queries he cared about. The total cost of the work was lower than three months of paid Google Ads, and the recommendation persists without ongoing spend.
The mistake is treating AEO as a single-channel play. Reviews alone, press alone, or schema alone will not move the recommendation needle for a service business that competes against operators who are working all five. The strategic move is to build all five in parallel, not sequentially. The signal density compounds; isolated investments stall out before they reach threshold.