The Questions to Ask Before You Sign an AI Search Contract

First question: what does "AI search work" actually mean in your contract?

"AI search optimization" is the newest label on the shelf, which means it is the easiest one to sell with nothing behind it. Before you sign, make them define the term in deliverables you can point at, not adjectives. When a homeowner asks ChatGPT or Perplexity "who's the best HVAC company near me," the model runs a live search, grabs a handful of pages, and writes an answer from what it found. Getting named in that answer comes down to a short list of concrete work. Ask them to walk you through each piece.

A real scope names these four levers. If an agency cannot, they are selling you a word:

• Entity clarity. Making the engines understand your company is one specific, real business: this trade, this service area, this phone number, written the same way everywhere. This is the foundation. If ChatGPT thinks you are two companies because your name appears three ways across the web, nothing downstream works.
• Schema and structured data for LLM parsing. Machine-readable facts on your pages so an answer engine can quote your services, your area, and your details cleanly. This is schema built to get cited, which is a different job than schema for a Google rich result.
• Citation-worthy source pages. Answer-first pages that directly say what a homeowner asked, so the engine can lift a sentence and attach your name to it. Vague marketing copy does not get quoted.
• Third-party corroboration. Getting your profile, reviews, and directory listings to agree with your site, because the engines cross-check your story before they name you.

Now the boundary question, because this is where padded contracts hide. AI search visibility is not the same as classic blue-link SEO, and it is not your Google Business Profile or map-pack ranking. Those are related lanes that feed AI answers, but a contract that quietly re-bills you for basic SEO under an "AI" label is charging premium for old work. Ask plainly: which of these deliverables are new AI-search work, and which overlap with SEO or Local SEO I may already be paying for? A straight shop will draw that line for you. A padded one will blur it on purpose.

How do you define a citation, and how do you track it across the engines?

This is the question that separates the real shops from the pretenders, because AI answers are hard to measure and that difficulty is where weak agencies hide. Push on it hard. If they cannot tell you exactly what counts as a win and exactly how they will show it to you, they are not measuring the thing you are paying for.

Start with the definition. A citation is your company being named, cited, or linked inside an AI answer for a query that matters to your business. That is the unit. Not "impressions," not "AI visibility score," not some vendor's invented metric with a green arrow. Ask whether they track being named in the answer text versus linked as a source, because both count and they behave differently. Then ask which engines they check: ChatGPT, Gemini, Perplexity, Copilot, and Google's AI Overviews are the ones sending contractor traffic, and a shop that only checks one is watching a sliver.

Then the mechanics. AI answers are non-deterministic: ask the same question twice and you may get named once and skipped once, especially early. That means a single screenshot proves nothing, in either direction. Real tracking is a fixed list of the exact queries that matter to your trade and area, checked on a schedule across the engines, so you can watch named-versus-skipped trend over weeks. Ask to see a sample of that tracking from a real engagement, redacted if needed. If the "report" is one lucky screenshot, walk.

One more: ask how they connect a citation to a booked job. An AI mention that never turns into a phone call is a trophy, not a result. AI-answer attribution is still imperfect, so a good shop watches the tracked citations climb alongside your calls and form fills and reads the trend together. Claiming perfect per-call AI attribution is overselling. You want the middle: measured, honest, tied to the truck.

Who owns the entity work, the pages, and the site when the contract ends?

Ask this before money changes hands, because the answer decides whether you are building an asset or renting one. AI-search work is different from a monthly ad buy in exactly this way: the entity clarity, the schema, the source pages, and the corroboration you pay to build are supposed to compound and keep producing after they are set. That only helps you if you own them.

The right answers are simple and you should hear a straight yes to each:

• You own your domain, registered in your name.
• You own the website files and the pages written for you, and you can take them if you leave.
• The schema and structured data live in your site, not locked in a vendor's platform you rent access to.
• The entity cleanup (consistent name, address, phone, and service area across the web) stays true whether or not you keep paying, because those are your real business facts, not a subscription.

Watch for two specific dodges in AI-search deals. The first is the platform trap: some agencies host you on a proprietary system and structure the deal so that when you leave, your pages and schema leave with them and your citations decay because the source pages go dark. That is not visibility you built, it is visibility you rented. The second is the "AI visibility dashboard" upsell, a monthly tool you can never stop paying for that shows you a score. Tracking is useful, but the underlying work has to belong to you, not evaporate when you cancel the dashboard.

This is one reason we hand-code static sites instead of building on WordPress or a locked platform. Your pages are plain HTML with the schema baked in, they load in under 2 seconds, and if we ever parted ways you would walk out with a clean, fast, fully owned site that keeps its structure and keeps getting read by the engines. Ask any agency the same thing and listen for whether "you keep the results" quietly means "you keep nothing you can move." Access is not ownership. Get the ownership terms in writing before you sign.

What is a realistic timeline, and what do you promise versus what you cannot control?

Timeline is the fastest way to spot a liar in this lane, because AI citation runs on its own clock and no honest shop can shortcut it. The engines have to re-crawl your updated pages, understand your entity, and cross-check your story against the rest of the web before they will put your name in an answer. That takes months, and it happens in a rough order that a real provider can describe.

Here are honest ranges for an established contractor who already ranks decently in Google, so you can catch a pitch that is too good:

• Your own brand name ("tell me about [Company] in [city]") should resolve correctly within weeks once the entity and schema work lands. If it does not, that is the first thing to fix.
• Specific, low-competition service questions ("how much does a panel upgrade cost," "do you repair Rheem heat pumps") can start earning citations in roughly 2 to 6 weeks.
• The competitive head queries ("best roofer in [metro]," "who installs mini-splits near me") are a 4 to 9 month job, longer in a big metro where national brands are already the default answer.

Now the red flag, and it is a hard one. If an agency guarantees a ChatGPT #1, guarantees a specific AI citation, or promises the competitive answer in 30 days, leave. Nobody controls how these engines assemble an answer. They are non-deterministic and they change without notice. A shop that promises a guaranteed spot is either lying or does not understand the channel it is selling. What an honest provider commits to is the work (the entity, the schema, the pages, the corroboration, the tracking) and gives you a realistic range on the outcome, the same way you bid a job honestly: here is the range, here is what makes it faster or slower, here is what I can commit to versus what the algorithm decides.

Ask specifically: what makes my case faster or slower, and did you actually look? A clean existing entity in a mid-size market moves faster than a company the engines have never heard of in a top-20 metro. If they give you the same timeline no matter what your web presence looks like, they have not looked at it.

What happens when the numbers are behind, and how often do we talk?

Every honest engagement hits a stretch where the citations are not moving as fast as hoped. What matters is what the contract and the shop do about it. This is the question that catches agencies who plan to bill you on autopilot and blame the black box when you ask why your name still is not in the answers.

The middle of an AI-search engagement is the hard part: past the entity and schema work, before your name shows up in answers consistently. A weak provider disappears here, hides behind "AI is unpredictable," and keeps invoicing. A real one watches the leading indicators and tells you the truth about them in plain language. Ask them to name the signals they will show you before you get named in answers:

• Brand-query accuracy. Does each engine get your trade, service area, and details right when asked about your company? This resolves first and is the earliest honest sign the work is landing.
• Recrawl and index of your source pages. Have the engines re-read your rewritten, marked-up pages yet? Nothing downstream can change until they have.
• Corroboration alignment. Do your site, profile, reviews, and directories tell the same story, character for character? When they line up, citation follows; when they drift, it stalls.

Then ask the two questions that separate a service from a subscription. First: what happens in month three if the tracked citations are flat? A real shop has an answer about diagnosing which lever is stuck (entity, source pages, corroboration, or the organic starting point underneath it) and adjusting. A pretender has a shrug and an invoice. Second: how often do I get a real human on a call to explain the report? A dashboard emailed on the first of the month with no context is a data dump, not reporting. You want someone who will walk you through what moved, what stalled, and what they are doing about it, in language you would use with a customer.

One honest caveat runs through all of this: the contractor trades are seasonal. A furnace page that becomes citable in July does not really get tested until the first cold snap, so judge the trend over a quarter and a season, not a bad week. A good shop factors your season into the read instead of taking credit for weather or blaming it for a stall.

The AI-search red flags that should end the conversation

Some answers are disqualifying on their own, and you do not need to be a search expert to catch them. You just need to know what they sound like. Here is the short list of things that should end the call before you sign anything.

1. Guaranteed ChatGPT #1 or "AI citation in 30 days." Nobody controls how the engines build an answer. A guarantee is a sales trick or proof they do not understand the channel.
2. "AI is a black box, you just have to trust us." AI visibility is measurable enough to manage: brand-query accuracy, tracked citations across engines, corroboration alignment. A shop that will not measure it is a shop that cannot deliver it.
3. They invented a metric. An "AI visibility score" with no tie to a real query and a real buyer is the new version of domain authority: a vanity number designed to look like progress.
4. They cannot explain the mechanics. If they cannot tell you why an engine cites one contractor over another (entity, schema, source pages, corroboration), they are reselling a buzzword, not doing the work.
5. They re-bill basic SEO as "AI optimization." Ask what is genuinely new AI-search work versus classic SEO or Local SEO you may already pay for. A blurred answer is a padded invoice.
6. You cannot own or move the work. If leaving means your pages, schema, and citations evaporate, you are renting visibility, not building it. Non-negotiable.
7. The proof is one screenshot. AI answers are non-deterministic. One lucky screenshot of ChatGPT naming you is not tracking, it is a coin flip they got to keep.

One more, and it is the tell that runs under all the others: watch how they answer hard questions. A shop that does the work will happily get specific, admit what it cannot control, and tell you when you are not a good fit. A shop selling a buzzword smooths over every concern with a promise. You already vet subs this way. You check the license, ask for the last three jobs, watch how they talk about the work. Hire an AI-search shop the same way, and most of the bad ones filter themselves out before the first invoice.