AI visibility is an easy service to sell and a surprisingly easy one to lose money on. The diagnostic is compelling — you show a client they are absent from answers their buyers are reading — and then the delivery turns out to be entity reconciliation, structured data and content rewriting, priced as though it were a monitoring subscription.
This is a practical guide to running it as a profitable line: what to require from a platform, how to scope the first engagement, how to price it, and the specific traps that turn this service into unbilled work.
Most platform comparisons in this category are written for in-house teams managing one brand. Agencies have a different problem: per-client cost must not scale linearly with headcount. Four criteria decide that.
Check the brand ceiling on the tier you would actually buy, not the top tier. Published August 2026 limits differ sharply — Cituna's $119 Pro tier covers 3 brands and its $399 Max covers 10; Otterly advertises unlimited brand reports across its tiers. If you onboard clients in batches, this single number sets your floor cost per client.
Every engine your platform does not check is a caveat you have to explain in a client meeting, and caveats erode authority. As of August 2026 the common gaps are Claude and Grok — Peec AI lists neither, Otterly sells Claude as an add-on, and Cognizo gates it to Enterprise. If your clients sell to technical buyers, those two engines are not optional.
This is the difference between a diagnostic that interests a prospect and one that closes them. Without it you can say a client is not cited. With it you can say they rank fourth for a question worth a known number of monthly impressions, and that the answer above it names three competitors by name. The second is a number a client can take to their board.
This is the real margin lever. If the platform outputs a dashboard, a senior strategist has to interpret it before anyone can act, and your delivery cost is set by your most expensive person's calendar. If it outputs an ordered list of specific actions, a mid-level executive works the queue.
The most common scoping error is selling a 90-day engagement against outcomes nobody can predict. Scope against work completed, with measurement as the reporting layer rather than the deliverable.
Platform cost at agency scale is close to a rounding error against a retainer. Published entry pricing runs $29 to $499 per month; even the Max tiers sit well under a single junior day rate in most markets. Anchoring your price to tool cost gives away the entire value of the work.
| Cost line | Typical monthly | Share of a mid-size retainer |
|---|---|---|
| Platform licence | $29–$499 | Small |
| Entity and structured data work | Billable hours | Meaningful |
| Content rewriting for extractability | Billable hours | Largest |
| Third-party coverage and digital PR | Billable hours or pass-through | Meaningful |
| Reporting and account management | Billable hours | Meaningful |
The substance you are billing is technical SEO and content work your team can already do. What is new is the diagnostic that justifies it and the measurement that proves it landed.
"Triple your AI citations in 90 days" is unfalsifiable at signature and indefensible at renewal. Commit to work completed and measurement published, and let the numbers be what they are. Clients who have been burned by unverifiable percentage claims elsewhere tend to find this more credible, not less.
If the client is paying for a dashboard, they will eventually notice they can buy that dashboard for $39 and cancel you. The deliverable is remediation. The dashboard is your instrument.
Reconciling business information across a client's directory footprint is genuinely time-consuming, especially for multi-location businesses, and it is nearly always underestimated because it looks like data entry. It is the single most common source of unbilled overrun in this service.
Branded prompts flatter every client. They also measure nothing commercial — a prompt containing the client's name tests whether the engine can look a name up. Report unbranded buyer-intent prompts only. It makes early reports worse and the engagement far more defensible.
The strongest framing we have found is not "AI is replacing search" — clients have heard it and discount it. It is narrower and checkable in the meeting: there is now a class of buying question your client never sees, and no analytics product will ever show it to them.
A lost ranking appears as a decline. A buying question answered by an assistant that names three competitors produces no signal anywhere — no impression, no click, no bounce. It is invisible by construction. Demonstrating that live, with the client's own questions, is a more honest pitch than any category statistic, and it survives scrutiny because the prospect watches it happen.
Send us a prospect domain and the questions their buyers ask. We will return the six-engine grid — including which competitors get named instead — so you can walk into the meeting with their data.
Request a baseline