What AI Visibility Costs — and What Changes the Price

The price of AI visibility work is set almost entirely by two things: how much off-site reputation you have to build from scratch, and how many buyer prompts you need monitored — not by how much content anyone writes for you. A business with an established review profile and a few directory listings is buying a different, cheaper job than a business starting at zero, even if the deliverables list looks identical.

That distinction is worth understanding before you take a single sales call, because it is the part most quotes obscure. Google’s own keyword planner data puts the cost per click for “ai visibility audit” at roughly $62, and for “ai seo agency” at about $29 — advertisers bidding those amounts are competing for buyers who have no idea what the underlying work costs. This article is our attempt to make the cost structure legible, using our own numbers where we have them and declining to invent the ones we don’t.

What actually drives the price of AI visibility work?

Three variables move the number: your starting off-site position, the number of prompts and engines under monitoring, and whether the engagement includes producing content or only directing it. Everything else on a typical proposal is a rounding error by comparison.

Starting position dominates. If your Google Business Profile has reviews, your business is listed in the city directories that publish agency and vendor rankings, and a few independent roundups already mention you, then the work is largely on-site: sharpen the pages, structure the answers, monitor the results. If you are at zero on all three, the same engagement has to fund months of outreach and review generation before the on-site work has anything to amplify. We know this one from the inside — our own profile still sits at zero reviews, which has kept us out of an entire answer format regardless of how good the pages are.

Prompt coverage is the second driver, and it scales linearly in a way content does not. Monitoring twenty buyer questions across ChatGPT, Perplexity, Claude and Google AI Overviews is roughly four times the work of monitoring five, because each prompt has to be run on a schedule, logged, and read against competitors. This is genuinely mechanical work, which is why tool-only options exist and are much cheaper than services.

The third variable is scope of production. An engagement that writes and ships pages costs meaningfully more than one that audits, prioritises, and hands your team a queue. Neither is wrong. Buying production when you have a capable in-house writer is the most common way businesses overpay in this category.

Should you buy a tool, a service, or nothing yet?

Buy a tool if you have someone who will act on a weekly dashboard; buy a service when you know the gap and cannot close it yourself; buy nothing yet if you have never sampled your own prompts. The third case is more common than the industry admits.

The sampling test costs an afternoon. Write down the five questions a real buyer would type when looking for what you sell, run each one through ChatGPT with web search on, and record which businesses get named and which sources the model cites. If you are already named in three of five, your AI visibility problem is smaller than a retainer’s worth. If you are absent from all five and the model is quoting a directory you are not in, you have learned the actual shape of the work — and it is probably reputation work, not content work.

We have run exactly this test on ourselves every week since June 2026. Across fourteen weekly samples of the same Provo buyer question, our business was named five times and absent nine, and the sources ChatGPT cited rotated between independent roundups, city directories, and agencies’ own service pages with no stable pattern. Fourteen weeks of that log cost us a few minutes a week and told us more about where to spend than any vendor pitch could have.

What you’re buying	Fits when	What it will not do
Monitoring tool	Someone in-house will read and act on it weekly	Get you into the sources the engine reads
Audit, one-off	You need the diagnosis before committing budget	Change anything on its own
Ongoing service	The gap is off-site and you lack the time or relationships	Guarantee a mention in any assistant

How do you tell a rational quote from an invented one?

A rational quote names the prompts it will monitor, the engines it covers, and what it will do when the answer is “your problem is reviews” — an invented one leads with a promised outcome. The tell is whether the proposal describes work or results.

The strongest single filter is to ask what happens if the audit concludes the fix is off-site. An agency selling a content retainer has an incentive to find a content problem. Ask directly: if the finding is that we need reviews and directory listings rather than pages, does this engagement change? A straight answer is worth more than a case study. When we ran our buyer prompt on 7 September 2026, ChatGPT volunteered a version of the same advice unprompted, telling the buyer to ask each finalist to show actual before-and-after AI citation results rather than Google rankings. That is good advice and it is worth using.

Be equally sceptical of guarantees in the other direction. No agency controls what an assistant says, because the model is synthesising from sources it selects at query time. The Verge (2026) documented how quickly the AI SEO category filled with vendors selling certainty about a system nobody controls, and Business Insider (2026) reported on small businesses whose traffic moved sharply on AI Overview changes they had no say in. Anyone promising placement is either misunderstanding the mechanism or counting on you to.

There is a floor below which the work cannot be done honestly, and it is set by the manual sampling. If a proposal covers a dozen prompts across four engines with weekly reporting at a price that implies an hour a month, the sampling is not happening — the deliverable is a template.

What does the price actually buy you in results?

It buys eligibility and evidence, not placement — and the timelines are uneven enough that any single-number promise is a guess. Structural on-page work shows up in weeks; review and reputation work shows up in months; genuine authority is slowest of all.

The evidence on what page-level work can do is reasonably firm. Aggarwal et al. (Princeton, 2023), the paper that named generative engine optimization, found that adding citations, quotations and statistics to a page lifted its visibility inside generated answers by up to 40%. That is a real effect and it is cheap to capture. It is also bounded: it improves how a page performs once the model reaches it, and does nothing about whether the model reaches it.

The off-site half has a harder floor and a bigger payoff. Trustpilot research reported by TechRadar Pro (2026) found businesses with 80 or more reviews were referenced in over 75% of relevant AI answers, against roughly 1% for those with none. Our own weekly samples suggest the eligibility threshold for the business-panel format is far lower than 80 — we have logged panels containing firms with a single review — but the direction is not in dispute. Reviews are the cheapest lever with the largest gate behind it, and no amount of retainer substitutes for asking your customers.

Traffic is the wrong success metric throughout, which changes what you should be paying for. SparkToro’s 2026 analysis found fewer than one in three Google searches still sends a click anywhere. A business can be named constantly inside answers while its referral numbers stay flat. If a proposal’s success criteria are sessions, it is measuring the wrong channel.

What we’d actually do

We would spend nothing until we had sampled our own prompts, then buy the narrowest thing that fixes the specific gap the sample revealed. In practice that means: run the five buyer questions yourself this week; if you are absent and the model is citing directories, spend on getting listed and on review velocity before you spend on content; if you are named but described wrongly, the fix is one well-written service page and it is cheap.

We publish our own numbers on the pricing page rather than quoting ranges here, and we would rather tell you on a call that your problem is reviews than sell you a content retainer that will not touch it. If you want the diagnosis without the commitment, book a call and we will run your prompts live while you watch. The methodology behind the weekly log is in what ChatGPT recommends when you ask for an SEO agency, and the monitoring side of the work is described on our AI citation tracking page.