How marketing agencies package AI visibility as a service, and what the measurement layer has to carry

Agencies have been asked for this since buyers started getting shortlists out of assistants instead of search results. The service is real, the delivery is unfamiliar, and the pricing is unsettled, which is an opportunity for the agencies that get the measurement right and a liability for the ones that sell it as another ranking report.

A caution first. No defensible median monthly US fee for a clearly defined AI visibility service exists in anything we have measured. Figures circulate. We have not collected them, so none appear on this page.

What the service actually is

It has two halves and they are usually sold as one thing, which is where the trouble starts.

Measurement. A fixed panel of buyer questions, run on named engines on a schedule, with every raw answer and its cited sources kept. This half is recurring, cheap to run once built, and the only part that can report a result.

Work. Changes to the client's own pages, technical fixes so the pages can be retrieved, and the slow business of getting the client described accurately by other people. This half is where the hours go and where the outcome actually comes from.

Measurement without work produces a monthly report nobody can act on. Work without measurement produces an invoice nobody can defend. Priced together as one retainer, the first client who asks what changed gets an answer about activity rather than outcome.

The deliverables worth committing to in writing

• The question panel, written out and dated, with branded and unbranded questions in separate lists and the client's brand name absent from the unbranded set.
• Named engines and surfaces, with Google AI Overviews counted separately from Gemini, because they are different surfaces.
• A monthly report whose every rate carries its denominator, and which reports brand naming and owned-domain citation separately rather than blended.
• Raw answers and cited URLs available to the client, not just a dashboard summary.
• Coverage reporting, meaning successful observations over scheduled observations, so a quiet month is distinguishable from a failed collection.
• A tolerance band agreed in advance, below which a monthly move is not reported as a change.

Three ways agencies price it, and what each one survives

A measurement retainer

A fixed monthly fee for the panel, the runs and the report. Easy to deliver, easy to defend, and it caps your exposure because you are not promising an outcome. The weakness is that the client can eventually buy the tool directly, so the retainer has to carry judgement and not only numbers.

A project with a measurement tail

A one-off audit and remediation sprint, then a small ongoing fee for the panel. This matches the work honestly, because most of the fixable technical and content problems are found once. It is the model that produces the fewest arguments in month four.

Outcome or performance pricing

Attractive to clients and the most dangerous of the three, because the outcome depends on things nobody controls. Model updates, competitors publishing, and whether a question calls for a company name at all. In our own run, 184 of 635 readings named no company whatsoever. Billing against movement in a number that can be flat for structural reasons is a way to lose money for doing good work.

The multi-brand limit that catches agencies out

Check, before you sell ten clients onto one subscription, how many brands the plan genuinely covers, and get the answer from the published page rather than a sales call. Our own pricing page is a cautionary example and we will use ourselves as it. It describes the Agency plan as covering 10 client brands, while its own comparison table answers No to more than one business in all eight columns, including Agency. Both of those render today, and an agency planning a client roster should plan against the restrictive reading and ask us to confirm in writing.

Also check white labelling specifically rather than assuming it. Our pricing page, 119,203 bytes, renders no white label at all, tested by word boundary across the whole document, along with no SSO, no SAML, no SOC 2, no ISO and no DPA. A loose search for sso on that page matches crossOrigin and subprocessors rather than the feature, which is how this gets described wrongly in both directions. Read on 22 October 2026, tryprofound.com/pricing mentions a self-serve Agency Growth plan with 400 credits per month per client workspace and publishes no price for it, and publishes SSO, SAML and SOC 2 compliance with up to 9 answer engines tracked on Enterprise. For an agency whose clients run security reviews, that is a real advantage over us and worth saying plainly.

What the client report has to show, with an example

Three fractions and a limit.

Naming rate. Readings in which the client was named, over readings in which any company was named, with the date and the engines. Citation rate, as its own line, never added to the first. And downstream traffic with its attribution limit stated, because Google's documentation on AI features, read on 22 October 2026 at developers.google.com/search/docs/appearance/ai-features, says AI features traffic is included in the overall search traffic in Search Console under the Web search type and describes no way to separate AI Overviews or AI Mode traffic from the rest.

Our own measurement, offered as the calibration most agencies never get. On 22 October 2026 we scored 635 readings on ChatGPT and Gemini from 324 US questions in 19 batches, 550 of them with live search. There were 1,053 namings across a roster of 24 vendors. Our own brand was named 0 times and our domain appeared in 0 of the 19,793 citation URLs across 3,307 hosts. Half of all those citations came from just 51 hosts.

And the result an agency should quote to any client expecting volume to fix this. Between 20 and 22 October 2026 we went from 134 recommendation-shaped pages live to 641, and the naming count moved from 0 to 0. Publishing more pages of the usual shape, on its own, did not move the measurement.

Where we fit, and where we do not

We are not an agency and we do not sell advertising, SEO services, content production or outreach. What we sell is the measurement layer underneath all of it, whether ChatGPT, Gemini, Google AI Overviews, Perplexity and Claude name a business when a buyer asks. An agency can use that to show whether its own work is landing, which is a different purchase from the work itself.

Our published US prices are SEO Start 149, Growth 299, Scale 599, SEO Scale 699, Inbound 1099, Outbound 1299, Agency 1299 and Everything 1799 dollars a month, read on 22 October 2026. SEO Start reads 0 questions, 0 credits and 0 engines, so nobody should quote AI visibility as starting at 149 dollars, least of all in a client proposal. Growth at 299 is the cheapest plan that reads all five surfaces, and a Claude check costs 3 credits rather than 1, which changes a cost-per-observation model if Claude is in the panel. Every plan starts with the same free scan on the client's own domain, which needs no account.

Who should not sell this service

An agency that cannot keep raw answers and show them to a client should not sell it, because the whole report rests on evidence the client can read. An agency unwilling to publish a tolerance band should not sell it, because the first noisy month will cost the relationship. And an agency whose clients sit in categories where assistants name nobody should test twenty questions by hand before building a product on it. That test costs an afternoon and occasionally saves a year.

Questions agencies ask about selling AI visibility

Can I white label an AI visibility tool

Ask for it in writing per vendor. Our own pricing page renders no white label, so with us today the honest answer is no.

What should I charge

We have no measured median to give you. Price the measurement as a recurring fee you can deliver profitably in a quiet month, and price the work separately against hours.

How many clients fit on one subscription

Fewer than the marketing suggests, usually. Our own page says 10 brands in one place and No to more than one business in another, and an agency should plan against the stricter line until a vendor confirms otherwise.

Is this just SEO with a new name

The technical half overlaps and the decisive half does not. Being retrievable is familiar work. Being described the same way by several independent sources is what decides whether a name appears, and that is not a ranking problem.