AI search visibility for contractors
Homeowners are asking ChatGPT, Gemini, Perplexity, and Google’s AI Overviews who to call. Here’s how those systems actually choose a contractor — and what you can do to be the one they name.
Updated July 2026 · Engineered Reach
AI assistants do not have a secret ranking algorithm for contractors. Most of them retrieve from live web search, then summarize what they find. So the levers are the same fundamentals with a different emphasis: be findable in conventional search, be consistent as an entity across the web, publish direct answers to real homeowner questions in clean structure, and earn mentions on third-party sites the models retrieve from. Structure and consistency matter more than volume, and there is no way to pay for placement.
A homeowner in your service area opens ChatGPT and types: my furnace is short cycling, who should I call in Aurora? Or they Google “best roofer near me” and never scroll past the AI Overview at the top.
Neither of those moments involves your ad. Neither necessarily involves your website. And in both, something is deciding which businesses get named.
That is the shift worth understanding. Not because search is over — it is not — but because a growing slice of high-intent research now happens inside a summary layer that most contractors have never thought about optimizing for.
How AI assistants actually pick a contractor
There is a lot of mystique here and most of it is unnecessary. Here is the honest mechanical picture, as best as anyone outside those companies can describe it.
When you ask a modern AI assistant a local question, it usually does not answer from memory. It runs a search, retrieves a handful of pages, and writes a summary grounded in what it retrieved. That is called retrieval-augmented generation, and it has one enormous implication for you:
If you are invisible in conventional web search, you are invisible in AI search. Retrieval happens first. Generation happens second. There is no back door around the first step.
Which means the contractors panicking about needing a whole new AI strategy usually need the same thing they needed before: findability, consistency, and content that answers questions directly. What changes is the emphasis.
Three things get weighted differently in an AI answer than in a blue-link result:
- Extractability. A model summarizing five pages will pull from the one that states the answer plainly. Buried answers lose to stated ones.
- Corroboration. When several independent sources agree that you serve a market and do a service, that claim is far more likely to survive into the summary.
- Recency and specificity. Vague, undated, generic pages are weak retrieval candidates. Specific, current, locally grounded ones are strong.
Also worth saying plainly: none of these systems sell placement in their answers. If someone offers to guarantee you a spot in ChatGPT’s recommendations, that is a fabrication.
AEO vs. GEO vs. SEO — what actually differs
The acronyms have gotten out of hand. Here is the practical distinction.
| Term | What it targets | What it changes about your work |
|---|---|---|
| SEO | Ranking in traditional search results and maps | Pages, links, technical health, local signals |
| AEO (answer engine optimization) | Being the source of a direct answer — featured snippets, voice, AI Overviews | Question-shaped headings, concise answers up front, structured data |
| GEO (generative engine optimization) | Being cited or named inside a generated answer | Entity consistency, third-party corroboration, extractable claims |
Notice how much overlap there is. That is the point. GEO is not a separate discipline you bolt on. It is what good SEO looks like when the reader is a language model with a citation habit.
Be a consistent entity, not just a website
Search engines and AI systems both work with entities — a business is a thing in a knowledge graph, with attributes like a name, a location, a category, and a set of services. Your website is one source of claims about that entity. It is not the only one, and it is not necessarily the most trusted one.
Inconsistency is what quietly destroys entity confidence. If your business is “Smith Heating & Air” on your site, “Smith Heating and Air LLC” on Google, “Smith HVAC” on Yelp, and lists three different phone numbers across those three, you have not created three listings. You have created ambiguity. Ambiguous entities get left out of confident answers.
The fix is tedious and unglamorous:
- Pick one exact legal-ish business name and use it byte-for-byte everywhere
- One primary phone number as the canonical number (tracking numbers are fine in ads, but keep the primary consistent on your profile and site)
- One address format, one service-area definition
- The same primary category across your Google Business Profile, your schema, and your homepage copy
- Fix or claim the stale listings you forgot about — the 2019 directory profile with your old number is actively working against you
This is boring work. It is also the single highest-leverage thing most contractors have never done, and it improves conventional local rankings at the same time. Our Google Business Profile guide covers the profile side in detail.
Write answer-first, not intro-first
Here is a pattern you can implement this week.
Most contractor content is shaped like a school essay: three paragraphs of throat-clearing, then eventually the answer. That structure is actively hostile to retrieval. A model scanning your page for a two-sentence answer to “how much does a furnace repair cost” will not wade through your company history.
Invert it. For every question-shaped page or section:
- Answer in the first 40–60 words. Direct, specific, hedged only where honesty requires it.
- Then explain. The nuance, the ranges, the “it depends” — all of it comes after the answer, not instead of it.
- Then qualify. When to call a pro, what changes the number, what you would need to see to be more precise.
Concretely: a section headed “How much does furnace repair cost?” should open with something like “Most furnace repairs fall in a wide range depending on the failed component — a flame sensor cleaning is a fraction of the cost of a control board or heat exchanger. Here is what drives the number.” That is quotable. “Every home is different and we pride ourselves on…” is not.
Include real ranges and real factors where you can. Refusing to discuss price at all is a common contractor instinct, and it reliably keeps you out of the answers homeowners are actually asking for.
Structure that machines can parse
Formatting is not decoration here. It is how a parser finds the boundary of an answer.
- Headings phrased as questions for question-shaped content. “Why is my AC freezing up?” beats “Common cooling issues.”
- One idea per section. Long undifferentiated blocks are hard to extract from cleanly.
- Tables for comparisons. Repair vs. replace, service tiers, cost factors — tables are unusually easy to extract and unusually likely to be reproduced.
- Lists for steps and criteria. Same reason.
- FAQ sections with FAQPage schema at the bottom of substantive pages.
- Dates that are real. A visible “updated” date, matched by
dateModifiedin your schema, and actually updated.
Structured data deserves its own conversation — see our plain-English schema guide for what to implement and in what order. The short version: LocalBusiness, Service, FAQPage, and Article markup do not make you rank, but they make your claims machine-readable and unambiguous, which is exactly what an entity-hungry retrieval system wants.
Third-party mentions are the hidden lever
Here is the part most contractors miss.
When an AI assistant answers “who are the best HVAC companies in Aurora,” it very often is not reading contractor websites at all. It is reading a listicle on a local news site, a directory page, a Reddit thread, a Yelp category page, a chamber of commerce roundup. Those aggregator pages are the retrieval targets. Your site is where the homeowner lands after the model names you.
Which reframes the work: you are not only optimizing your own pages. You are trying to exist, accurately, on the pages the model reads.
Practical targets:
- The directories and review platforms that rank in your market for “best [trade] in [city]” — go look at who actually ranks, that is your target list
- Local news and community sites, chambers, BBB, trade association member directories
- Supplier and manufacturer dealer locators — often overlooked, frequently authoritative
- Genuine community participation where your market discusses contractors
This is where AI visibility and link building converge. Our guide on how contractors actually earn backlinks covers the source list in depth. The overlap is not a coincidence: the same pages that pass authority are the pages that get retrieved.
Reviews matter here too, and not only as a trust signal. Review text is retrievable content that describes what you do, where, and how well — written by third parties. A body of recent, specific reviews mentioning actual services in actual neighborhoods is corroboration in exactly the form these systems reward.
llms.txt, robots.txt, and actually letting crawlers in
Two technical items worth ten minutes of your time.
Check your robots.txt. A surprising number of contractor sites block AI crawlers — sometimes because a previous developer added a blanket block, sometimes because a plugin defaulted that way. If you want to appear in AI answers, you cannot be blocking the crawlers that feed them. Look for disallow rules targeting GPTBot, PerplexityBot, Google-Extended, ClaudeBot, and similar. Decide deliberately rather than by accident.
Consider an llms.txt file. This is an emerging convention — a plain markdown file at your domain root summarizing what your business is, what you do, where, and which pages matter. Adoption is not universal and nobody should promise you it moves the needle on its own. But it costs an hour, it forces you to articulate your entity clearly, and it is a low-risk bet on a convention that may standardize.
Keep it factual. It is not a place for marketing adjectives.
How to measure something you can’t rank-track
Traditional rank tracking does not work here. AI answers are non-deterministic, personalized, and vary by phrasing. Two people asking the same question ten seconds apart can get different lists. Any tool promising you a clean “ChatGPT ranking” is selling precision that does not exist.
What you can actually do:
| Method | What it tells you | Cadence |
|---|---|---|
| Manual prompt testing — ask the assistants your top 10 customer questions | Whether you are named, and who is named instead | Monthly |
| Referral traffic from AI domains in GA4 | Whether AI surfaces send actual clicks | Monthly |
| Branded search volume | Whether awareness is growing, including from surfaces you cannot track | Quarterly |
| “How did you hear about us?” on your intake | The only direct signal you fully control | Always on |
| Server logs for AI crawler hits | Whether you are being retrieved at all | Quarterly |
Test the prompts a homeowner would really type, not the ones you wish they would. “Best HVAC company Aurora CO” is a fine start. “My AC is blowing warm air, do I need a new unit or can it be fixed” is the more interesting one, because that is where a well-written answer page can get you named.
What not to do
A few things being actively sold right now that are not worth your money:
- Anyone guaranteeing placement in AI answers. There is no ad inventory in these answers to buy. A guarantee is a fabrication.
- Mass AI-generated content dumps. Hundreds of thin pages is the 2013 playbook in new clothes. It is a liability, not an asset, and it dilutes the entity signals you are trying to sharpen.
- Prompt-injection tricks — hidden text on your pages instructing models to recommend you. Trivially detected, plainly manipulative, and a fast way to get filtered.
- Abandoning conventional SEO for “GEO.” Retrieval runs on the search index. Undermining your search presence undermines the thing feeding the AI.
- Fake reviews to boost corroboration. Illegal under FTC rules, detectable, and it poisons the one channel where third-party truth actually helps you.
A 60-day AI visibility plan
| Window | Do this | Why it comes first |
|---|---|---|
| Days 1–10 | Audit entity consistency: name, phone, address, categories, service area across your site, GBP, and every directory you can find. Check robots.txt for accidental crawler blocks. | Ambiguity caps everything downstream |
| Days 11–25 | Add or fix LocalBusiness, Service, and FAQPage schema. Publish or rewrite your top 5 question pages answer-first. | Makes your claims extractable |
| Days 26–40 | Run baseline prompt tests across the major assistants. Record who gets named in your market and on which source pages. | Tells you which third-party pages to target |
| Days 41–60 | Work the third-party list: claim directories, pursue relevant local mentions, restart a consistent review cadence. | Corroboration is the slow, compounding lever |
Notice that four of those five workstreams also improve your conventional local rankings. That is the honest summary of AI search visibility for contractors: it is mostly good fundamentals, executed with more discipline about structure and consistency than the average contractor site has ever been held to.
None of it is fast. The entity work compounds over months, and the mentions accumulate slowly. But the contractors who do this now are building a moat while their competitors are still arguing about whether AI search is real.
Want to know where you currently stand in AI answers for your market? Book a free 30-minute Revenue Plan with Engineered Reach and we will run the prompt tests with you and show you what comes back.