How Do You Get ChatGPT, Gemini, and Perplexity to Recommend Your Contracting Business?

You make the sources AI already trusts say the right things about you. There is no ad slot and no submission form. Get named in reviews, directories, local listicles, and community threads; keep your business details identical everywhere; and publish pages that answer buyer questions in quotable, direct sentences.

That is the whole strategy. Everything below is detail on how to execute it — and, just as importantly, which popular tactics the evidence says are a waste of your Saturday.

Enough to matter, not enough to abandon Google. Assistants are used most for the research half of the job — comparing options, sanity-checking quotes, learning what a fair price looks like — while the final 'who do I call' step still usually happens in Google Maps, a review site, or a referral text.

That split is the practical takeaway. AI search rarely replaces the phone call; it shapes the shortlist that leads to it. A homeowner who asks an assistant 'what should I look for in a roofing contractor and what should a full replacement cost in my area' arrives at your quote with expectations that were set by whatever the model read. If your pricing page, your reviews, and your FAQ were part of that reading, the conversation starts from your framing.

Different engines pull from different places, which is why one can rave about you while another has never heard of you. The most-cited public dataset on this is Yext's 2025 citation study, summarized below alongside independently reported source patterns.

Yext also reported that roughly 86% of citations in its sample came from brand-manageable sources — meaning your own site plus third-party profiles you can claim and correct. That is the optimistic read: most of what AI says about you is coming from places you are allowed to edit. Most contractors simply never edit them.

Question-shaped and comparison-shaped queries almost always do; short navigational queries usually don't. Seer Interactive's 2026 analysis of roughly 49,000 tracked queries across 53 accounts found AI Overview prevalence varies enormously by query type.

Read that table as a content map. Every row above 75% is a query type where an AI summary is standing between the homeowner and your website — and where being one of the cited sources is worth more than being the fourth blue link. Comparison pages, cost pages, and honest 'best of' style content are the highest-leverage formats you can publish.

A question as the heading, then the complete answer in the next 40-60 words, then the detail. Models lift self-contained passages. If your answer only makes sense after three paragraphs of setup, there is nothing clean to quote and you get skipped for a competitor who wrote plainly.

The pattern that gets quoted

• Heading is the exact question a homeowner would type or say — not a clever headline.
• First sentence answers it outright, including the number or the verdict. No throat-clearing.
• Next 2-3 sentences give the caveat or the range, so the passage is defensible on its own.
• Then the deep detail, tables, and examples for the humans who keep reading.
• Specifics beat adjectives: '$8,000-$21,000 for a 2,000 sq ft asphalt shingle replacement' is quotable; 'affordable, high-quality roofing' is not.

The pattern that gets ignored

• Marketing-speak headings ('Quality You Can Trust') that match no real query.
• Answers split across a paragraph, a bullet list, and an image caption.
• Key facts — prices, service areas, hours, certifications — that exist only inside images or PDFs.
• Pages that require JavaScript to render their main content.
• Walls of unstructured text with no headings for a model to anchor to.

Because an assistant treats your website as a claim and everything else as evidence. Your site says you're the best roofer in town; so does every competitor. Independent sources — reviews, directories, local press, community threads, trade associations — are what a model uses to decide the claim is credible.

This is why two contractors with identical websites get wildly different AI treatment. The one with 400 reviews across Google, Yelp, and Angi, a complete BBB profile, a few mentions in local roundups, and a real presence in neighborhood discussion threads is a well-evidenced entity. The one with a beautiful site and eleven reviews is a claim with no corroboration.

The citation stack, in priority order

• Google Business Profile — complete, correct categories, services listed, photos, Q&A answered, posts current.
• Review volume and recency across Google, Yelp, Facebook, and your trade's main platform. Recency matters; 200 reviews from 2021 read as a business that may not exist anymore.
• Major directories: Yelp, BBB, Angi, Thumbtack, Houzz, Apple Business Connect, Bing Places, Foursquare/MapQuest data.
• Trade-specific directories and manufacturer 'find a certified installer' listings — these are high-trust and under-claimed.
• Local listicles and roundups ('best HVAC companies in [city]'). Listicles are among the most-cited formats in AI answers; being inside one is worth more than a homepage rewrite.
• Genuine community presence — local subreddits, Facebook groups, Nextdoor — where you're actually helpful rather than spamming your link.

Heavily, and indirectly. Reviews and profile data are the most structured, most trusted public facts about a local business, so they anchor how models describe you. Volume, recency, star average, and — underrated — the actual words in your reviews all feed what an assistant says.

That last point is the one contractors miss. If your reviews repeatedly use phrases like 'emergency AC repair on a Sunday' or 'financed our whole roof,' those specifics become part of your machine-readable identity. A generic 'great service, highly recommend' review adds a star but no descriptive signal. When you ask for a review, ask the customer to mention what job you did and where — never scripted, just prompted.

Fix your evidence base first, then your content, then measure. The order matters — publishing twenty new pages while your directory listings contradict each other is building on sand. Here is the sequence we'd run for any home-services business starting from zero.

Build a prompt set and run it on a schedule. There is no Search Console for AI answers, so the honest method is manual sampling: fixed prompts, fixed cadence, logged results. Fifteen to twenty prompts checked monthly is enough to see real movement.

What to log every month

• The prompt, the engine, and the date — run in a logged-out or incognito session to reduce personalization.
• Were you named at all? Named in the first three? Named with correct details?
• Which sources were cited — this is the most actionable field. It tells you exactly which third-party pages to go improve.
• Any factual errors about your business (wrong hours, wrong service area, a service you dropped two years ago). Fix these at the source.
• Which competitors were named, and what they have that you don't.

Supporting signals worth watching

• Referral traffic from chatgpt.com, perplexity.ai, gemini.google.com, and claude.ai in your analytics — small numbers, but usually high intent.
• Server or Cloudflare logs for AI crawler user-agents (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended). If they never visit, nothing else matters.
• Direct and branded search volume — a rising 'people looked us up by name' trend is often the first visible effect of AI-assisted discovery.
• New-lead intake question: 'How did you hear about us?' with an AI assistant option. Crude, cheap, and more honest than most attribution.

Most of the 2026 AI-visibility gold rush is selling tactics with no evidence behind them. Skip these and put the hours into reviews and third-party presence instead.

• Paid 'get listed in ChatGPT' services. No AI provider sells answer placement. What you're buying is directory submissions you could do yourself.
• Stuffing pages with AI-flavored keywords ('AI-recommended roofer'). Models read meaning, not incantations.
• Building separate Markdown or bot-only versions of your pages. Google and Bing have both indicated this crosses into cloaking territory — real policy risk, no demonstrated upside.
• Treating llms.txt as a growth lever. It's cheap infrastructure worth having if it's automated, but large-scale studies have found no meaningful correlation with citations. See the technical checklist guide for the full evidence.
• Mass-generated thin content. Volume without substance gets you neither rankings nor citations, and it dilutes the pages that were working.
• Buying reviews or posting fake community recommendations. Policy violation, reputational risk, and increasingly detectable.

AI visibility is a trust and shortlist play with a long runway. Paid ads are a demand-capture play that works this week. They are complements, not substitutes — and the contractors who treat AI visibility as a reason to cut ad spend usually end up with neither leads nor citations.

The practical interaction: strong AI and organic presence lowers the friction on every paid click you buy. A homeowner who clicks your Meta ad, then asks an assistant 'is [your company] legit' and gets a confident, well-sourced yes, converts at a materially higher rate than one who gets a shrug. That's the compounding effect worth investing in.