How AI Assistants Decide Which Home Service Companies to Recommend

A homeowner's water heater fails on a Sunday night. Five years ago, they typed "water heater repair near me" into Google and scanned the map pack. Today, a growing share of them open ChatGPT, Gemini, or Perplexity and ask a different kind of question: "Who should I call to replace a water heater in Franklin, Tennessee? I want someone reputable who won't overcharge me."

The answer they get back is not an ad auction and not a map ranking. It's a recommendation — usually two to five company names, each with a sentence of justification. If your brand is in that answer, you get the call. If it isn't, you were never in the running, and no dashboard tells you it happened.

So it's worth understanding: where do those names come from?

Where AI answers actually come from

AI assistants build their recommendations from two layers.

The model's baked-in knowledge. Large language models are trained on a snapshot of the public web — websites, directories, reviews, forum threads, news coverage. If your brand appears consistently across that footprint, the model may "know" you: what you do, where you operate, and roughly how you're regarded. This layer changes slowly. It rewards brands that have been clearly and consistently described across many sources for a long time.

Live retrieval. For local, time-sensitive questions, most assistants now search the web in real time and synthesize an answer from what they find. ChatGPT browses. Perplexity is built on retrieval. Google's AI Overviews draw on its index and business profiles. This layer rewards something different: pages that are easy to fetch, easy to parse, and directly answer the question being asked.

Most home service brands underinvest in both layers, because neither one is quite the same thing as classic SEO.

The signals that decide who gets named

Across the major assistants, a consistent set of signals shapes who gets recommended:

Entity clarity. Can a machine state, unambiguously, what your company is called, what services you offer, and where you operate? Many contractor websites fail this test. The brand name differs between the site, Google Business Profile, and directories. Service areas are listed as a paragraph of town names on a footer. The "services" page is a lumped-together list with no detail. Assistants recommend what they can describe confidently — ambiguity reads as risk.

Service-line specificity. A homeowner doesn't ask for "plumbing." They ask about tankless water heater installation, sewer line replacement, slab leak detection. Assistants match questions to companies at the service-line level. A dedicated, substantive page per service line gives the retrieval layer something to land on. One generic services page gives it nothing.

Reputation across platforms. Assistants cross-reference. Google reviews matter, but so do Yelp, BBB, Angi, Facebook, and Reddit threads where locals ask "who did you use?" A brand with a strong rating in one place and silence everywhere else looks thinner to a synthesizing model than a brand with a consistent story across five sources. Volume, recency, and what the reviews actually say all feed the one-sentence justification the assistant writes about you.

Third-party corroboration. Being described by someone other than yourself is powerful. Local press, supplier and manufacturer directories (dealer locators, certification pages), chamber-of-commerce listings, "best of" roundups — these are the citations assistants lean on when deciding which names are safe to put in front of a user.

A machine-readable website. Retrieval-based assistants fetch your pages and parse them in milliseconds. Clean HTML, real text (not text baked into images), structured data (LocalBusiness, Service, FAQ markup), fast load times, and no wall of JavaScript between the crawler and your content. If a bot can't read the page, the page doesn't exist for this channel.

Freshness. Stale content decays in both layers. A site that visibly maintains its pages — updated dates, current service info, recent reviews — signals an operating business, not an abandoned one.

Why map-pack winners can still be invisible

The uncomfortable part: dominance in traditional local search doesn't transfer automatically. The map pack is driven heavily by proximity, Google Business Profile optimization, and review count on a single platform. AI recommendation is driven by the breadth and consistency of your description across the whole public web, plus how well your site answers specific questions.

We routinely see established brands — decades in business, top-three map rankings — get skipped in AI answers while a smaller competitor with cleaner service pages and broader citations gets named. The assistant isn't measuring who's biggest. It's measuring who it can confidently describe and justify.

What to actually do

If you run an established home service brand, the starting point is cheap: see what the assistants say about you today.

1. Run the prompts your customers run. Ask ChatGPT, Gemini, and Perplexity the questions a real homeowner in your market would ask — by service line, by city, with qualifiers like "reputable" and "licensed." Note who gets named, how you're described (or misdescribed), and which sources the answers cite.
2. Fix entity basics. One canonical business name, service list, and service area — identical on your site, Google Business Profile, and every major directory.
3. Build real service-line pages. One substantive page per service you want to be recommended for, written to answer the questions homeowners actually ask, in each market you care about.
4. Widen your review footprint. Keep earning Google reviews, but stop ignoring the second and third platforms in your category. Consistency across sources is the signal.
5. Earn third-party mentions. Manufacturer dealer pages, local press, community involvement writeups. Each one is a citation an assistant can lean on.
6. Re-test monthly. AI answers move. Competitors who invest here will move them in their favor.

None of this is exotic. It's the discipline of making your brand legible — to machines that are increasingly the first "person" your customer asks.

That's the work FlyBilly does end to end: benchmark how you're surfaced, described, and recommended across the major assistants, fix what's holding recommendation back, and tie the movement to calls and revenue. But whether you do it with us or on your own, do the benchmark. You can't win a channel you're not measuring.