How to Choose a GEO Consultant for a Home Service Business: 10 Questions to Ask Before You Pay
If you own an HVAC, plumbing, roofing or appliance repair company, more of your customers now ask ChatGPT or Google's AI answer "who should I call" instead of scrolling a list of links. Getting your business named in that answer is called generative engine optimization (GEO), and in 2026 nearly every agency has added it to the menu. The hard part for an owner is telling apart someone who can show that the work brought in booked jobs from someone who sends a monthly chart that goes up.
This guide gives you ten questions to ask before you pay anyone for GEO, what a straight answer sounds like, and the warning signs. Under each question I put my own answer, so you can see what a specific answer looks like and judge me by the same test.
Why does it matter who you pick?
AI answers are easy to flatter. The headline number most tools report (how often your business appears in AI answers) comes from the tool asking AI assistants its own list of questions, so changing the list changes the number. A provider can show you growth that has nothing to do with your phone ringing.
There is also a quieter problem. On 5 November 2026 I asked ChatGPT, as an HVAC owner would, to recommend five independent consultants who could get a business recommended in ChatGPT and Google AI answers. It named five people. What they had in common was not a big website: one of the sites had fewer than 600 words. What they had in common was that their page titles said, in plain words, what the owner had asked for: a consultant, for contractors, for ChatGPT and AI Overviews. AI assistants pick the business whose own pages answer the question in the customer's words. Your consultant should understand that about your site, and ideally about their own.
Will you do the work yourself, or hand it to someone else?
Many agencies sell GEO and pass the work to a junior or a contractor you never speak to. That is not always bad, but you should know who is touching your Google Business Profile and your website.
A good answer: a name, and who you can talk to when something breaks.
My answer: I do the work myself, with no agency and no account managers in between. I am one person, based in Europe and working US hours, so I am the wrong choice if you need a team managed rather than a system built.
Can you show me what ChatGPT says about my business today?
Before you pay, the provider should be able to show you a baseline: what ChatGPT, Google AI Overviews and similar assistants say today when someone asks for your trade in your city, and whether your business is named. Without a baseline, nobody can tell you later whether anything changed.
A good answer: a written snapshot with the exact questions asked, the date and the assistant used.
My answer: that baseline is part of my $149 local visibility report on the pricing page: who ChatGPT and Perplexity recommend instead of you, and which pages they cite. The AI visibility audit, a fixed $199, adds your server log, which shows which AI assistants actually opened your pages. Nothing gets installed for either.
Which tool produced this number?
There are four different things people call AI visibility, measured by four different tools, and three of them can rise while revenue does not move. A provider who cannot say which tool produced a figure is giving you a number you cannot act on.
A good answer: the name of the tool, the date of the reading and what the tool cannot see.
My answer: crawler access comes from your server log; live answer fetches come from the same log, split from background crawling by the name each program reports; answer presence comes from a paid tool that samples AI answers; and revenue comes from a first-touch mark carried into your CRM. The full breakdown of what AI visibility means explains which tool is blind to what.
What would failure look like in your report?
This is the most revealing question on the list. Anyone can produce a chart that goes up. Someone who measures honestly can tell you in advance what a bad month would look like.
A good answer: a specific number that would stop growing or fall, and what they would do about it.
My answer: failure shows up as the count of your pages that AI cites going flat. On one client, AI answers citing the site rose from 63 to about 3,000 across six AI platforms over 15 weeks. During the first six weeks the number of cited pages went 13, 14, 17, 22 and then stopped, which meant the answers were all landing on the same small set of pages. If the answer count rises while the cited pages stay flat, I report that as a problem, not as a win.
How will I know which calls and jobs came from it?
For a service business the customer usually phones, and the job gets typed into the CRM by a dispatcher with no source attached. If nobody connects the AI answer to the call and the call to the job, you will never know what you paid for.
A good answer: a plan for tracking calls and forms by channel and joining them to jobs in your CRM.
My answer: this is where I start. On the main case, 45% of jobs in the CRM had no source at all; after tracking was joined to CRM jobs and invoice amounts, 11,724 past jobs were traced back to what produced them. The method is on the call tracking and lead attribution page.
Can you name a number you had to correct?
A provider with no corrections has either never measured anything or does not tell you when they are wrong. The second is worse.
A good answer: a real example and what they changed.
My answer: I had a strong finding that AI crawlers never requested a site's llms.txt file across two sites and 6,676 requests, and I threw it out, because that file was served in a way my logger could not see. I also once published one metric as two different numbers because I quoted two time windows without naming either, and I went back and fixed the text. Both are written up on this site.
What can your measurement not see?
Every measurement has a blind spot, and a provider who lists none is selling.
A good answer: at least one honest limit.
My answer: a page fetch is not a citation. My log proves an assistant opened your page, not that it quoted it, which is why I also pay for a sampling tool. The program names in a log are claims, not proof, so I call those counts claimed. And a few sites in one industry is a probe, not a study. The raw numbers are public on the AI crawler data page.
How would you know if the tracking broke?
The expensive failures in this work do not show an error. A tracking mark that stopped working three weeks ago still produces a confident report.
A good answer: they watch for numbers that suddenly change shape, such as a field that is never empty starting to be empty.
My answer: I found exactly this in a live AI voice agent that was booking jobs onto technicians who were marked off, because a calendar check saw only 600 of 7,801 events and returned a valid looking answer every time. How I watch for it is in LLM observability: monitor outputs.
What do I keep if we stop working together?
If the answer involves a dashboard only they can log into, you are buying a dependency.
A good answer: everything lives in your accounts, with written notes on how it works.
My answer: everything runs on your own accounts and infrastructure. The tracking stays, the log stays, and I write the handover notes for whoever comes next. The public code I can show is an anonymised version of my call analysis pipeline, call-audit-demo, because the working version runs against client accounts.
What does it cost, and what is the smallest first step?
GEO prices in 2026 range from under $100 a month for a tracking tool to five figures for an enterprise program, because those are different products sold under one name. A provider should publish a price or at least a starting price, and offer a small first step before a monthly contract.
A good answer: a written price and a first step you can judge on its own.
My answer: my prices are on the pricing page. A free written diagnostic costs nothing, the local visibility report is $149, the AI visibility audit is $199, GEO work starts from $2,000 as a project, and the monthly system is $1,500 a month, month to month.
What are the warning signs?
Be careful when a provider promises a position in ChatGPT, because nobody controls what an assistant says. Be careful when every number in the report comes from one tool's dashboard with no date and no method. Be careful when nobody asks how your calls are tracked, because for a service business that is where the money is decided. And be careful when the contract is long and the first step is large.
Who am I the wrong choice for?
I am one person, so if you want a team with a project manager and weekly meetings, an agency suits you better. I work in writing and asynchronously, and I would rather send a document than hold a call. My main measured case is a residential appliance repair company in the US, with smaller cases outside home services: a music education site, an event company in Spain and a cleaning company in Texas. More about how I work is on the about page.
What should you do with these questions?
Send them by email to anyone you are about to pay for GEO or AI visibility, including your current agency, and compare the answers in writing. Specific answers name a tool, a date, a person and a limit. Vague answers or a dashboard only they can log into are the warning signs. If you want to put the same questions to me, the free diagnostic is a good start, and the terms for longer work are on work with me. If you are looking at automation as well, real numbers from that work are on AI automation.
Related: