First, know exactly what an "AI search company" should do
The homeowner who used to type "AC repair near me" now opens ChatGPT and asks "who should I call to fix my AC in [city]." Gemini, Perplexity, Copilot, and Google's AI Overviews answer the same way: they name a few businesses and cite a few sources. Getting your shop into that answer is a specific discipline. It is not the same job as ranking in the blue links, and it is not the same job as showing up in the map pack.
Here is the work that actually decides whether an answer engine names you. Entity clarity: making it unambiguous who you are, what you do, and where you work, so a model does not confuse you with a shop two towns over. Schema and structured data written so a machine can lift your facts cleanly, not the generic markup a plugin drops in. Citation-worthy source pages that state something an AI can repeat: what you install, what you warranty, how fast you respond. Third-party corroboration so the sources an AI already trusts confirm your facts instead of contradicting them. And mention tracking, so you can see whether ChatGPT and the AI Overviews actually name you now.
That stack is the lane. A company that lives here can look at your site and tell you why the AI skips you. Notice what is NOT on the list: writing a hundred blog posts, buying backlinks, running ads, or optimizing your Google Business Profile. Those are real jobs, and they feed the wider picture, but none of them is what earns the citation. When you shop for an AI search company, you are hiring for the narrow stack, not the broad one.
Why this matters before you compare anyone: most agencies still sell blue-link SEO and have no answer for the citation layer, the schema, or the entity work that decides which contractor an AI names. If you cannot tell the difference, you will overpay for the old service dressed in a new word. So get concrete first. Open ChatGPT, ask who does your trade in your town, and read the answer. If your name is missing, that gap is exactly what this kind of company is supposed to close, and you should be able to say so out loud when a shop pitches you.
Make them show your current AI answer before anything else
The single fastest filter when choosing an AI search company is this: make them show you what the machines say about your shop today. It takes two minutes and it separates the real providers from the resellers instantly. A shop that does this work can pull up ChatGPT or Gemini on the call, ask "who does [your trade] in [your city]," and show you whether you are named, who is named instead, and what sources the answer cites.
If they cannot show your current state, they cannot price the fix, and they probably cannot do the work. The whole discipline starts with knowing where you stand in the answer, the same way a foreman starts by walking the job before he quotes it. A company that skips that step and jumps straight to a monthly fee is selling you a subscription, not a result.
What you are testing is not just the tool. It is how they think about the problem. Watch whether they read the actual answer engines and name what is missing (your entity is muddy, your service pages say nothing quotable, a competitor's facts are cleaner), or whether they wave at "AI visibility" in the abstract. Specifics about YOUR shop mean they looked. Generalities mean they are reciting a pitch.
A quick version you can run yourself before any sales call, so you walk in informed:
- Open ChatGPT, Gemini, and Perplexity. Ask each one who to hire for your main service in your city.
- Note whether you are named, who is named instead, and which sources get cited under the answer.
- Ask Google the same question and read the AI Overview at the top, not just the blue links below it.
- Now you know your starting point, and any provider worth hiring will show you the same picture and go a layer deeper into why.
When you have your own read, the sales call gets honest fast. You will know within minutes whether the person across the table actually understands the citation layer or is guessing at it.
The questions that separate a real GEO/AEO shop from a reseller
Most bad AI-search engagements could have been avoided with five blunt questions asked out loud before signing. Ask them on the call and watch how fast, and how specifically, the answers come.
- "Show me what an AI says about me right now, and tell me why." Covered above, and it is first for a reason. No current-state read means no real work.
- "What are the line items in the setup?" You want to hear entity work, schema, page rewrites, corroboration, and tracking as distinct pieces. "AI optimization" as one vague blob is the tell that nobody looked.
- "Which of my pages need to change, and why?" A real answer names your service and location pages and says what is missing. A reseller talks in generalities because nobody read your site.
- "How will you track whether the AIs start naming me?" Mention tracking is the deliverable that proves the work landed. If they cannot describe how they will show you the change, they are guessing.
- "Is any of this actually SEO, local, or ads in a costume?" Blue-link rankings, your map pack, and paid placements are real work, but they are not AI-citation work. If the "GEO package" is mostly those, you are paying an AI premium for old services.
One quieter test: ask what they would NOT do for you. A shop worth hiring will tell you when AI-search work is not your first priority right now (brand-new business, no reviews, conflicting entity data that has to be fixed before anything else) and point you at the honest next step. A shop that says yes to everything is selling, not advising.
Pay attention to how they handle your own facts, too. When you say "my money is in re-roofs, not repairs" or "I only want commercial HVAC," a real provider writes that down and steers which pages get made quotable and which claims get corroborated. That specificity is what gets you cited for the work you actually want more of. A reseller nods and runs the same template it runs for a dentist and a law firm.
Red flags: the pitches that mean you should walk
Because "AI search" is new and hot, it attracts more costume-wearing than almost any service in this business. A lot of agencies are slapping "AI" on the same retainer they sold last year. Treat each of these as a reason to end the call.
| What they say | Why it's a red flag |
|---|---|
| "We'll get you into ChatGPT, guaranteed." | Nobody controls what a model returns. Real work makes you the clearest, best-corroborated answer, which raises your odds over months. A guarantee means they do not understand the layer or they are lying. |
| "Flat $X a month, sign here." (before looking at your site) | The price is set by how clean your data is and how contested your market is. A national rate quoted sight unseen is a guess dressed as a price. |
| "It's the same as SEO, we've always done this." | It is not. Entity clarity, LLM-parsable schema, and citation-worthy pages are different work than blue-link ranking. This answer means you are buying old SEO with a new label. |
| "We'll submit you to a hundred AI directories." | There is no such thing. You do not get cited by spamming submissions. You get cited by being the clearest, most-corroborated source. Bulk submission is the AI-era version of a link scheme. |
| Twelve-month lock-in with a steep exit fee. | Much of this work is one-time setup that holds. A long contract with a big cancellation penalty means the paper is doing the work the deliverables should be doing. |
Two structural red flags matter more than any single sentence. First, the vagueness trap: if the provider cannot show your current AI answer, cannot itemize the setup, and cannot name a single page that needs to change, there is no work behind the fee. Second, the platform trap: if the schema, entity data, and pages they build only live inside their proprietary system and vanish when you leave, you built citation equity on rented land. Your AI mentions will fade with your subscription. Real work goes onto your own site, in your own account, and keeps working after the invoices stop.
What a real AI-search audit looks like (and why to demand one)
The best way to choose is to make a shop prove itself before you commit: get an audit. A company that knows this work can look at your site, your listings, and the live answer engines and tell you specifically what is holding you back. You should have that audit in 1-3 business days, and it should read in plain English, not in SEO jargon meant to impress another SEO.
A real AI-search audit covers a few concrete things. It shows what ChatGPT, Gemini, Perplexity, and Google's AI Overviews say about your trade in your town right now, and whether you are named. It checks your entity: whether your name, service list, and area served agree across your site, your Google profile, and the directories, because a model will not trust facts that contradict each other. It checks whether your schema is built so a machine can actually parse and quote your facts. It reads your service and location pages and flags which ones say nothing an AI could repeat. And it looks at corroboration: whether the sources AI already trusts confirm your claims or ignore them.
What you are really testing with the audit is how they think, not just the findings. Do they speak in specifics about YOUR shop, or in a template that could be pasted onto anyone's business? Do the recommendations connect to getting named in an answer a homeowner reads, or just to abstract scores? A good audit reads like a foreman walking your job site: this entity here is muddy, that is why the AI names your competitor, here is the fix and roughly what it takes. That clarity is exactly what the whole engagement will feel like. If the audit is fog, the work will be fog.
One caution: an audit is not a scare tactic. There is a version built only to alarm you, a long red list of "critical AI errors" run off a tool and printed to look urgent, that never connects a single finding to getting cited. The tell is that it prioritizes nothing. A real audit says here are the two or three things keeping the AIs from naming you, here is why, and here is roughly what it takes to fix them. Fear-selling by report is just the digital version of the tech who finds a cracked heat exchanger on every furnace he inspects. And a genuine audit is useful even if you never hire that shop, because you keep the findings. That alone tells you whether they are trying to earn you or trap you.
Trade focus, ownership, and the fine print before you sign
Two contractors can get two fair quotes that are hundreds of dollars apart, because the price is set by the condition of your data and the heat of your market, not by a rate card. That is normal. What is not negotiable is the three parts of the paper that matter more than the price.
Trade focus. A generalist agency that does dentists on Monday, restaurants on Tuesday, and you on Wednesday will write pages that do not know your trade, and an AI will not cite pages that read like nobody. A shop in one lane (home-service contractors) already knows that an emergency plumbing searcher and a remodel shopper convert differently, that a re-roof and a repair are different animals, and that your buyer asks the machine "who should I call" from a hot house at 9pm. Since 2008 that has been our one lane: home-service contractors, twenty trades, no restaurants, no law firms. Ask any shop how many contractors they work with and how many industries total. Depth in your lane is what makes the pages quotable.
Ownership. Get it in writing that you own the domain, the website, the content, the schema, and your Google account. This sounds obvious and it is violated constantly. If the entity data and citation-grade pages live on the agency's locked platform, your AI mentions walk out the door when you leave. Ownership is the difference between building equity and paying rent.
Reporting and timeline, in writing. The deliverable that proves this work is mention tracking: are the answer engines naming you more than they did, and on which queries. Be suspicious of reports built on "impressions" or vanity scores that never tie to being cited. On timeline, an honest shop sets the clock up front: entity and schema fixes can land in weeks, but earning consistent citations in a real market takes months as the engines re-crawl and re-trust your data. Anyone who promises instant inclusion is setting up a fight in month two. Read the exit terms too: you want to leave and take everything, the site, the schema, the pages, the account. If leaving is expensive or you leave empty-handed, that is the reseller model, and it is the one thing this whole guide is built to help you avoid.
Get those three right, trade focus, ownership, and honest reporting, and you have filtered out most of the bad actors before a single page gets written.