Book a Call

Does ChatGPT Recommend Contractors? Honest Answer

Reading about it is step zero. Check whether your service area is still open: about a minute of questions, then you pick a time. Check if your area is open.

Yes, it names companies. No, that is not a recommendation.

Ask ChatGPT who to hire for gutter cleaning in your town and it hands back a short list of names. Nobody judged the work to build that list. It came out of listings, reviews and directory pages, because that is all the assistant has to read. It has never seen your trailer, your before photos, or the Saturday you drove back out because a downspout was still slow.

You cannot make an assistant like you. You can make yourself easy to find, easy to quote and hard to get wrong. Every number below carries its source, and where nobody knows, this says so.

Table of contents

  1. What happens when someone actually asks
  2. It names you, it does not vouch for you
  3. How many people are asking
  4. Where the answer comes from
  5. What actually moves the needle
  6. How to check, and why the check lies
  7. What nobody knows yet
  8. The one thing you can buy
  9. Frequently asked questions

What happens when someone actually asks

Two kinds of question get completely different answers.

"How do I get the dog smell out of artificial grass?" is a how-to question, and the assistant answers it like a magazine: rinse it, an enzyme cleaner with a ten to fifteen minute dwell time or a one to one vinegar mix, repeat weekly. The pages behind that answer are mostly cleaning product makers, because they wrote the how-to pages and almost no service company did. Questions like that trigger a Google AI Overview, the answer block above the normal results, 92 percent of the time, and cost questions 97 percent of the time (Whitespark, 540 local searches, May 2025).

"Who is the best window cleaner around here?" is a hire question, and the assistant behaves like a directory instead. Across a 330-response sample of ChatGPT Search and Google AI Mode, the names coming back most often were Fish Window Cleaning in 21 percent of answers, Angi in 20 percent, Window Genie in 19 percent and Yelp in 15 percent (parse.gl, 2026-08-25). Two of those four are not window cleaners. They are places where window cleaners are listed.

Losing the hire question to Angi and Yelp is the normal outcome, not proof that something is broken on your site.

It names you, it does not vouch for you

An assistant has no idea whether you show up on time. It has read what other people published about you and repeated the parts that fit the question. That is a citation, not an endorsement.

It also gets things wrong. One 2026 review found ChatGPT and Perplexity return correct business details only 68 percent of the time, against Gemini at 100 percent, because Gemini is grounded in Google Maps data (cited.so, 2026). It can name your company and hand the caller a phone number you gave up two years ago.

Treat any agency selling a "ChatGPT recommends you" badge as selling a claim the product does not make.

How many people are asking

More than last year. Fewer than the people selling this would like you to believe.

  • 45 percent of US consumers used an AI tool to find a local business in the past year, up from 6 percent (BrightLocal 2026, via The Valley Marketing Group).
  • On hire searches like "turf cleaning near me", the block of local map results showed on 93 percent of them and an AI Overview on 15 percent (Whitespark, 540 local searches, May 2025).
  • A wider read of local business queries a year later put AI Overviews at 68 percent (Whitespark via Search Engine Journal, July 2026).

Those last two look like a contradiction. They count different query sets at different dates, so anybody quoting you one precise share is quoting whichever study flatters the pitch. The honest version: it is rising fast, it is far higher on how-to and cost questions than on "near me" hire questions, and the map is still where the homeowner with an overflowing gutter ends up.

A click out of an answer can show up in your analytics. A name read on screen and then dialed from memory records nothing anywhere, so some share of your calls already starts inside an assistant and nothing you own can tell you which.

Where the answer comes from

For Google's AI answers there is real documentation: no extra requirements and no special markup to appear in AI Overviews or AI Mode, only an indexed page eligible to be shown as a snippet, plus an up to date Business Profile (developers.google.com). The work is ordinary search work.

ChatGPT is a different product, and OpenAI has published no equivalent rulebook. What exists instead is measurement of what actually gets cited.

  • An audit of more than 200 local service citations from January to May 2026 split them Google Business Profile 42 percent, directory aggregators 28 percent, the company's own website 17 percent, local press 8 percent and review platforms 5 percent (beancount.io, 2026).
  • An analysis of 28 million AI responses counted Yelp cited 512,680 times in the fourth quarter of 2025, about 3.4 times the Better Business Bureau at 149,710, with Angi at 145,633 and Thumbtack at 56,004 (Foundation Marketing and AirOps, via ppc.land).

Most of what gets said about you is not written by you. Your website is a minority shareholder in your reputation.

What actually moves the needle

Nothing here is new, which is the point. The work that gets you cited by ChatGPT is the work that gets you found on Google.

Your Google Business Profile, treated as the main asset. Every service named the way a customer would say it, real job photos instead of stock, reviews asked for on the day of the job and replied to. It was the single largest source in that citation audit, at 42 percent (beancount.io, 2026).

The same name, address and phone in every listing, to the character. Yelp, Angi, Thumbtack, Nextdoor, the Better Business Bureau. Nobody outside those companies has published how an assistant picks between two conflicting versions of you. Given the error rate above, fewer versions of you means fewer ways to be wrong.

Your own prices, published. The biggest miss in the trades. With no number from you, the answer gets built out of the national guides: gutter cleaning at HomeAdvisor's $119 to $234 averaging $168, This Old House at $191 to $529 averaging $360, Fixr at $200 to $400 (homeadvisor.com and thisoldhouse.com, 2026; fixr.com, 2025). A window cleaning answer generated on 2026-09-04 landed on $150 to $302 a visit averaging $221, drawing on angi.com, homeguide.com and thisoldhouse.com. Your customer turns up holding a number written for the whole country. A dated table with your town and your real price is the thing those guides cannot offer.

Answers written as answers. Nobody has published how an assistant decides what to lift out of a page. What is safe to say is that "gutters need cleaning once or twice a year, and two to four times under heavy tree cover" stands on its own, and a paragraph that warms up for five sentences does not. Pages cited by AI answers also run 25.7 percent fresher on average than pages that are not (Ahrefs, via Search Engine Journal). A pattern, not a proven cause, but dating your pages costs nothing.

Reviews, in volume and recent. The first two numbers here are about people, not machines: 47 percent of consumers skip a business with fewer than 20 reviews, and 31 percent will not consider one under 4.5 stars (BrightLocal, 2026). The machine number is thinner: businesses ChatGPT named in one 2026 review averaged 4.3 stars (cited.so, 2026).

About llms.txt. It is a small text file telling AI crawlers what they may read. We put one on every build because it costs nothing, and we will not sell it as the door into AI answers: no assistant maker has published anything saying its product reads it.

How to check, and why the check lies

Ask the way a homeowner would, in a fresh chat, signed out where you can: best gutter cleaning company in your town, who cleans gutters there, what does it cost. An assistant you have been chatting with all morning hands your own company back, which means nothing. Treat any single answer as one data point, because the same question asked twice can come back with different names. Ask those three once a month and watch the list, not the day.

Two things deserve a reaction. If it names you with a wrong phone number or the wrong service, go fix the listing it read, because that error is reaching callers right now. If it names only Angi, Yelp and Thumbtack, that is not a fault in the assistant. It is your market telling you the directories out-published everybody local.

What nobody knows yet

The ranking factors are not published. OpenAI has not said how a local shortlist gets built. Google's guidance for its own AI answers says only that there is no special markup and no requirement beyond an indexed page and a current Business Profile (developers.google.com), which is not a ranked list of anything. Every "AI search ranking factor" article you have read is inference from watching outputs. Some of it is probably right. None of it is confirmed.

Most local operators are nowhere. A first quarter 2026 study of more than 320 prompts across five engines found 78 percent of independent local operators had effectively no share of the citations, while Yelp, Angi and Thumbtack absorbed them (5W, 2026). If your name never appears, you are in the large majority.

The size of the prize is unmeasured. Because those calls mostly arrive untracked, nobody can say what this channel is worth. Anyone quoting you a return figure on AI search work made it up.

The one thing you can buy

The list of names is not for sale, and nobody has published a way to buy a place in it. The space underneath it is a different story.

From OpenAI's own documentation: ads began running below ChatGPT answers in 2026, they show to logged-in adult users on the Free and Go tiers in the US, and Plus, Pro, Business, Enterprise and Education subscribers see none (openai.com, 2026-02-09). The self-serve ads manager opened in beta on 2026-05-05 (openai.com, 2026). You buy per click or per thousand views in a relevance-weighted second-price auction, with a suggested opening bid of $3 to $5 a click and a $25 minimum daily budget, and every account passes a manual review before it runs (help.openai.com, 2026). Targeting is by state, television market or ZIP code, so you show only where you work, and local services is a permitted category under the ad policy updated 2026-08-31 (openai.com and help.openai.com, 2026).

The channel is real: OpenAI reported a $1 billion annualized revenue run rate and availability in more than 40 countries on 2026-08-31 (openai.com). What does not exist is evidence that it books jobs for contractors. No operator has published a cost per booked job from it, and this page will not invent one. Run it after Google is already paying for itself, on a fixed budget, in your ZIP codes only, and count booked jobs rather than clicks. The full record is on our ChatGPT ads for contractors page.

Frequently asked questions

Does ChatGPT recommend specific contractors? It names them. It does not evaluate them. The names come from listings, reviews, directories and published pages, so what looks like a recommendation is a citation of what others wrote about you.

Why does it name Angi and Yelp instead of local companies? Because the directories publish more, and more consistently, about every trade in every town. In that 28 million response analysis, Yelp was cited more than any rival (Foundation Marketing and AirOps, via ppc.land). Being listed there accurately is part of the answer, not a defeat.

Is it worth doing anything about it yet? Yes, because it is the same work. A complete Google Business Profile, matching listings, recent reviews and pages that state your prices plainly are what win the local map results, which showed on 93 percent of "near me" hire searches in one study (Whitespark, May 2025).

Your next step

If your name does not come back when you ask, the fix is not an AI trick. It is a site that publishes your prices and answers questions in plain sentences, with a Google Business Profile and listings that agree with it. Ours goes live 7 days from handover, $2,499 to own outright or $297 a month to rent. Our flagship client, Murphys Turf, runs at 200 or more leads a month on the full program, the site plus the ads we manage on top of it.

We take one company per trade per service area, first come first served, while you are a client. Book a strategy call and we will check what the assistants say about your company, in your towns.

Related reading: