ChatGPT Ads for Local Service Businesses in 2026
Reading about it is step zero. Check whether your service area is still open: about a minute of questions, then you pick a time. ChatGPT ads for contractors.
Yes, you can buy an ad that sits under a ChatGPT answer. In the United States, today, through a self-serve account, pointed at your own ZIP codes, starting at the $25 daily minimum OpenAI publishes (help.openai.com, 2026).
What you cannot buy is a straight answer about whether it books jobs. Nobody has published one.
This post is the record and nothing else: what OpenAI says about its own product, in plain language, with the date and the page it came from. Where the record runs out, it stops and says so.
Table of contents
- The short version
- What the ad actually is
- Who sees it, and who never will
- What it costs to get in
- Where you can point it
- Whether your trade is allowed to run it
- What the record does not tell you
- Where this sits next to Google
- How to run it as a test you can read in four weeks
- The free half: getting cited without paying
- Questions people ask
The short version
Everything in this list is on OpenAI's own pages. None of it is an estimate.
- Ads show below ChatGPT's answer, for logged-in adults on the Free and Go plans in the US. Paying subscribers on Plus, Pro, Business, Enterprise and Education see no ads at all (openai.com, 2026-02-09).
- The self-serve ads manager opened in beta on 2026-05-05, with cost per click bidding, a tracking tag for your website and a way to send confirmed bookings back to OpenAI (openai.com, 2026).
- You buy by the click or by the thousand views, in a relevance-weighted second-price auction, with a suggested opening bid of $3 to $5 a click and a minimum daily budget of $25 (help.openai.com, 2026).
- Every account goes through a manual review by a person before a single ad runs (help.openai.com, 2026).
- You can aim it by state, by designated market area, or by ZIP code (help.openai.com, 2026).
- Local services are a permitted category. OpenAI's ad policy, updated 2026-08-31, limits ads primarily to consumer verticals including lifestyle and household goods, local services, and travel and experiences (openai.com ad policies, 2026-08-31).
That is the buyable product. If somebody sells you a detail that is not on that list, ask which OpenAI page it came from.
What the ad actually is
One format: a labelled unit underneath the answer, carrying your business name, one line of copy, an image and a link (openai.com, 2026).
It sits in its own labelled slot below the reply, the way a sponsored result sits above the map on Google. It is not the answer.
That is a useful place for a contractor to be, because the answer above it was written for everybody. It cannot price a two-story house on your street and it cannot book Thursday morning. Your ad, and the page behind it, can.
What OpenAI has not published: how often an ad appears, or which questions trigger one. So nobody can honestly say how often a homeowner in your county would see yours.
Who sees it, and who never will
Logged-in adults on the Free and Go plans in the United States, and nobody else (openai.com, 2026-02-09).
In plain terms: the homeowner on the free plan asking what a fence costs per foot can see your ad. The one paying for the top tier never will, whatever you spend.
Is that most of your market or a sliver? Nobody has published a breakdown of US homeowners by ChatGPT plan, so any percentage you are quoted was invented. That hole is worth naming rather than filling with a guess.
What it costs to get in
Three published numbers, and no others.
- Minimum daily budget: $25 (help.openai.com, 2026). Left running every day, that floor comes to $750 in a 30-day month.
- Suggested opening bid: $3 to $5 a click (help.openai.com, 2026). Suggested, not fixed. It is where OpenAI tells new advertisers to start.
- Pricing model: cost per click or cost per thousand views, in a relevance-weighted second-price auction (help.openai.com, 2026).
Second-price auction carries its usual meaning: you do not pay your full bid, you pay roughly what it took to beat the next advertiser. Relevance-weighted means the fit between your ad and the question counts alongside the money. How OpenAI scores that fit, it has not published, so nobody outside the company can tell you what actually makes one ad win over another.
One cost nobody lists is the wait. A person reviews your account before anything runs (help.openai.com, 2026), so file before a season starts, not during it.
Where you can point it
By state, by designated market area, or by ZIP code (help.openai.com, 2026). A designated market area is the television market a town sits in, which for most contractors is far too wide.
Use ZIP codes. The ones your truck drives to, and nothing past them. A lead outside your route is not a win, it is a call you have to talk your way out of. This is the setting most likely to burn the first $25 a day: the wider you aim, the faster the budget clears and the worse the calls get.
Whether your trade is allowed to run it
Yes. OpenAI's ad policy, updated 2026-08-31, says ads are primarily limited to consumer verticals such as lifestyle and household goods, local services, and travel and experiences (openai.com ad policies, 2026-08-31). Local services is where pressure washing, lawn care, fencing, gutters, turf, concrete, tree work and irrigation sit.
Categories outside that list can be refused at review, so check anything you sell alongside the core service before you count on it.
What the record does not tell you
There is no published cost per booked job from ChatGPT ads. Not for pressure washing, not for lawn care, not for any trade. The self-serve product opened in beta on 2026-05-05 (openai.com, 2026), which is not long enough for anyone to hold a full year of seasonal data, and the operators running it are not publishing their numbers.
Also unpublished: how a click from an assistant compares to a click from a search result once it reaches your phone.
On size, OpenAI reported a $1 billion annualized revenue run rate and availability in more than 40 countries on 2026-08-31. The channel is real and it is growing. That says nothing about whether it books fence jobs in one county.
So: the product is real and buyable, the performance is unknown. Any agency quoting you a cost per lead for ChatGPT ads in your trade is making it up, and you should treat that number like a roof quote from someone who never went up the ladder.
Where this sits next to Google
Behind it. Google is the first dollar. This is the experiment you layer on once the first dollar pays for itself.
Google has numbers and this does not. In Google's window cleaning category, where gutter work also sits, about 13.58% of ad clicks turn into a lead at a $66.69 cost per lead (localiq.com 2025; clicksgeek.com 2026), and Local Services Ads gutter leads run $25 to $60, paid per lead, behind Google's screening (thevalleymarketinggroup.com 2026). You can plan a season around figures like that. Not around a channel with no published figures at all.
Scale matters too. In a study of 540 local searches, an AI answer appeared on 15% of local-intent queries while the map results appeared on 93% (whitespark.ca, May 2025). The assistant is where a lot of the research now happens. The map is still where most of the hiring happens.
How to run it as a test you can read in four weeks
If you try it, run it like a test and not like a bet.
- Get the account in for review early. A person reviews it before anything runs (help.openai.com, 2026), so file the month before your season, not the week it starts.
- Fixed window, fixed budget. Four to six weeks in season, at or near the $25 daily floor (help.openai.com, 2026). Set the number first and do not raise it mid-test.
- Land the click on a page that prices the job for that town. The answer above your ad was written for everybody. If your page cannot beat it with a real local number and a booking button, the click is wasted.
- Count booked jobs, not clicks. The account takes a tracking tag for your website and confirmed bookings sent back to OpenAI (openai.com, 2026), so you can measure the thing that pays you instead of the thing that is easy to count.
- Report it weekly beside the Google numbers, and keep it only if it pays. A test you will not switch off is not a test.
Worst case, a few hundred dollars buys you something nobody in your trade knows yet. An open-ended budget handed over because AI is the future buys you nothing.
The free half: getting cited without paying
The other half of this is being cited by ChatGPT with nobody paying anything, and for most contractors that is the better first move, because it carries no daily budget.
Most citations are not your website. In the same whitespark study, 60% of AI citations went to third-party publishers and 40% went to business sites (whitespark.ca, May 2025). Your Google Business Profile, your reviews and your directory listings do more of this work than your homepage does.
The names that come back are often national. In a 330-response sample of ChatGPT Search and Google AI Mode answers for the best window cleaning service, Fish Window Cleaning was named in 21% of answers, Angi in 20%, Window Genie in 19% and Yelp in 15% (parse.gl, 2026-08-25). Chains and directories, mostly.
There is no secret file to upload. Google's own guidance says there are no extra requirements and no special markup needed to appear in AI Overviews or AI Mode. What it names instead is an indexed page a machine can read, and a current Google Business Profile.
Nobody outside these companies can see how one local company gets picked out of an answer and another does not. What the people studying it keep pointing back at is dull and cheap: honest pages that state the real local price, a Google Business Profile that matches them, reviews after every job, and the same name, address and phone everywhere a machine can find them. None of that is new advice. It is the same work that got companies found on Google, which is probably why so few people are trying to sell it to you.
Questions people ask
Can a local service business advertise on ChatGPT right now?
In the United States, yes. The self-serve ads manager opened in beta on 2026-05-05 (openai.com, 2026). A person reviews your account, then your ad can appear below answers for Free and Go plan users in the ZIP codes you choose (help.openai.com, 2026).
What does a ChatGPT ad cost?
Per click or per thousand views, in a relevance-weighted second-price auction. OpenAI suggests a $3 to $5 opening bid and sets a $25 minimum daily budget (help.openai.com, 2026). Those two figures are all it publishes, so any fixed cost per click quoted for your trade is a guess.
Do ChatGPT ads work for contractors?
Nobody knows, and this page will not pretend otherwise. No contractor has published a cost per booked job from the channel. That is exactly why it belongs in a small timed test with a number set in advance, not in your main budget.
Is this the same as being cited by ChatGPT?
No. The ad is a paid unit below the answer. Being named inside the answer is unpaid, and it comes from your profile, your reviews, your listings and pages worth quoting. Our AI search page for turf cleaning shows that work.
Your next step
The full product breakdown lives on ChatGPT Ads for Contractors, from the same sources as this page.
What the test looks like in your trade: pressure washing, lawn care, gutter cleaning, fence installation, window cleaning, turf cleaning, roof cleaning and tree services.
If you have no website worth sending a click to, start there instead. Every paid click lands on a page, and if that page cannot price the job for that town and take the booking, the channel was never the problem. We build sites that go live 7 days from handover, $2,499 to own outright or $297 a month to rent. Our flagship client, Murphys Turf, runs 200 or more leads a month, none of it from ChatGPT ads.
We work with one company per trade per service area, first come first served, while you are a client. Book a free 45-minute strategy call and we will map your fastest path to booked work, whether ChatGPT ads belong in it or not. You keep the plan either way.
Related reading: