Answer engine optimization
What Client Access to AI Visibility Costs
Agencies and consultants budget for AI visibility as though every client needs a seat. Most vendors in this category stopped charging for seats, so that instinct misprices the work. The constraint that actually binds is different, and so is the reason you should not hand a client a login.
Key takeaways
- Seats are mostly free. Peec AI publishes unlimited user seats on every paid plan and Otterly includes unlimited team members on every tier.
- Profound is the exception worth knowing: its published $99 entry plan is single user and ChatGPT only, and its enterprise contracts are quote-based.
- The metered resources are brands or workspaces and prompt allowance, so a client costs you a workspace and roughly 25 to 40 prompts, not a head.
- The real objection to a client login is scope, not cost: a seat is all or nothing, and most reporting needs one result rather than the whole workspace.
- Screenshots fail for a specific reason. AI answers move, so a number a client cannot verify is a number they will eventually stop believing.
01The short answer
If you run AI visibility work for clients, stop budgeting per seat. The major self-serve vendors in this category have converged on unlimited users, and what they meter instead is the number of brands or workspaces you track and the number of prompts you measure. Your marginal cost per client is a workspace plus an allowance, which is a very different number from a per-head fee and usually a smaller one.
Then, separately, do not give the client a login anyway. Not because it costs money, but because a seat grants far more than the thing you actually want to hand over. The reporting problem is one of scope, and the right shape for it is a single result the client can open, verify, and not operate.
02Seats are mostly free now
The per-seat assumption is a holdover from the SEO tool market, where it was and often still is true. It has largely stopped being true here.
| Vendor | User seats | Notes |
|---|---|---|
| Peec AI | Unlimited on every paid plan | Also advertises free pitch workspaces, which is aimed squarely at agencies |
| Otterly | Unlimited team members on every tier | Published price ladder, so the whole bill is computable in advance |
| Profound | Single user on the published $99 entry plan | That tier also tracks ChatGPT only; larger contracts are quote-based |
| Agency-oriented tiers generally | Unlimited team members from the second tier up | Usually bundled with unlimited client workspaces and reports |
| The older pattern | Base fee for the first three users, then per seat | Still found, and still the thing that makes junior staff and client logins expensive |
Two practical consequences. First, if you priced a retainer on the assumption that each stakeholder costs money, you probably left margin on the table or padded the quote for a line item that does not exist. Second, when a vendor does still charge per seat, that is now a distinguishing feature rather than a default, and it belongs in your comparison alongside engine coverage and prompt allowance.
03What you are actually metered on
Almost every tool in this category meters two things: how many brands or workspaces you track, and how many prompts you measure across how many engines. Agency tiers typically unlock unlimited client workspaces from the second tier upward, which tells you where the vendor thinks the value line sits.
That makes the unit of agency cost a client-shaped bundle rather than a person. Published guidance on prompt set size clusters between 20 and 40 to start, so a reasonable planning figure is that one client is one workspace plus roughly 25 to 40 prompts, run on the engines you promised, at the cadence you promised. Those three variables are the quote.
The cadence variable is the one that gets underestimated, and it is not a padding exercise. AI answers are unstable enough that a single run is not evidence of anything, so a monthly deliverable built on one measurement per month is measuring noise. Whatever you promise a client about frequency is a direct multiplier on the allowance you need.
04The real problem is scope, not price
Suppose seats are genuinely free at your vendor. You still should not give the client a login, and it is worth being exact about why.
A seat is an all-or-nothing grant. It exposes the workspace: the prompt set you chose, the historical runs including the ones you would rather contextualise first, the configuration, and in a shared workspace sometimes the existence of other work. It also invites the client into an interface they will read without you sitting next to them, in a category where a single-run number genuinely can swing for reasons nobody caused.
“Roughly 40 to 60 percent of the domains cited in AI responses will be completely different just one month later, even for identical questions.”
A client who logs in on a bad day and sees a metric drop will ask a question you would have answered differently in a report. That is not an argument for hiding data. It is an argument that the artefact you hand over should be a result with a date and a scope, rather than a live console whose state you do not control.
What almost every reporting relationship actually needs is narrower than a seat and wider than a screenshot: one result, verifiable, attributable to a moment in time, that the client cannot accidentally reconfigure.
05What unlimited does and does not cover
Unlimited seats is a real and welcome change, and it is also the most load-bearing word on several pricing pages. It is worth reading carefully, because unlimited almost always attaches to the cheapest resource the vendor has.
A seat costs a vendor close to nothing. Another user reading a dashboard adds no measurement cost, because the runs already happened. Prompts are the opposite: every prompt is a real API call to a real model on every run, multiplied by every engine, multiplied by every re-check. So a pricing page offering unlimited users and a metered prompt allowance is not being generous in one place and stingy in another. It is pricing the thing that costs money and giving away the thing that does not.
That tells you where to read the fine print. Unlimited users rarely means unlimited brands. Unlimited brands rarely means unlimited prompts. Unlimited prompts, where you see it, usually comes with a smaller set of engines or a slower cadence, both of which are the same lever wearing a different label. When a plan looks generous on every axis at once, the axis you have not found yet is the cadence.
There is a harder case: the vendors who publish nothing. In our own survey of this category, several of the best-known names render no price at all, and at the top end contracts are quote-based and reportedly run from $30,000 to $100,000 a year. With those vendors, seat policy is not a policy. It is a term, and terms are negotiable in both directions. If you are an agency, that is worth using: client workspaces and seats are cheap concessions for a vendor to grant, and cheap concessions are the ones to ask for first.
06Why the screenshot habit fails
The default fallback is a screenshot in a slide. It is quick, it costs nothing, and it quietly damages the thing agencies sell, which is credibility.
A screenshot carries no provenance. The client cannot tell when it was taken, which engine produced it, whether it was one run or twenty, or whether the number would look the same today. In a stable category that is a minor weakness. In this one, where daily testing across four engines found 69 percent of a typical answer's sources changing from one day to the next, it is a structural one: your client is being asked to take an unverifiable number on trust, about a system that visibly changes its mind.
The second failure is subtler. A screenshot is a claim you made. A link the client can open is a thing they can check. The difference matters most exactly when the news is good, because an unverifiable win is the kind of thing a sceptical CFO discounts to zero at renewal.
07The four ways to show a client a result
| Approach | What the client gets | What it costs you | Where it breaks |
|---|---|---|---|
| A seat in your workspace | Everything, live | Usually nothing now | Scope: your prompts, your history, your configuration, read without context |
| A screenshot in a deck | One image | Nothing | No provenance, cannot be verified, ages badly against real volatility |
| A separate client workspace | Their own tenant | A workspace and its allowance | Duplicated setup, and you still choose between a login and a screenshot inside it |
| A shareable result link | One result, no account | Nothing beyond the plan | Only as good as its constraints: scope, expiry and revocation have to be real |
The fourth is the right shape for most reporting, and the constraints are what make it trustworthy rather than what limit it. A link with no expiry is a permanent public URL you have forgotten about. A link that carries the whole workspace has reinvented the seat. A link you cannot revoke is one you can never un-send.
08How to price a client if seats are free
Three inputs, all of which you control and none of which is headcount.
The workspace. One tracked brand per client. On most agency tiers this is unlimited above the entry plan, so the real question is which tier you need rather than what each client adds.
The allowance. Roughly 25 to 40 prompts per client if you follow the published consensus, multiplied by the engines you promised. This is the line that actually scales with client count, and the one to quote against.
The cadence. How often you re-measure, which is a direct multiplier on the allowance and the only defensible way to promise a trend rather than a snapshot. Prune quarterly on the consequence test: if losing an answer would not plausibly cost the client a deal, or you could not influence it within two quarters, the prompt is costing allowance for nothing.
Priced this way, a marginal client is a fraction of a plan rather than a new line item, which is usually a better story for your margin than the per-seat model you were braced for.
The arithmetic is worth doing once explicitly, because it changes how the quote is built. Take six clients at 30 prompts each. Under a per-seat mental model you count people: two of your own staff plus a stakeholder or two per client, and you arrive at something like fourteen seats and a number that scares you. Under the model the vendors actually use, you count 180 prompts and six workspaces, and then the only question is which tier covers 180 prompts at your promised cadence across your promised engines. Those are two completely different conversations, and only one of them is about the thing being billed.
It also changes which client is expensive. Under per-seat thinking, the expensive client is the one with a big stakeholder list. Under allowance thinking, the expensive client is the one in a broad category with many distinct buying questions, or the one who wants weekly rather than monthly reporting. The second framing is the accurate one, and it is also the one that tells you where to push back during scoping: a client asking for weekly numbers is asking for four times the allowance, and that is a scope conversation rather than a cost you should silently absorb.
09What to check before you commit
A short due-diligence list, in the order these things tend to bite.
Read the seat policy on the pricing page today. Not from a roundup. Where no price is published at all, which is common here, treat every limit as a negotiation.
Count workspaces, not users. Find the tier where client workspaces stop being metered and price from there.
Divide the allowance by your client count before you believe a prompt number. An allowance that is generous for one brand can be thin across six.
Ask what a recipient sees. If the only answers are a full seat or an export, your reporting will default to screenshots whatever you intend.
Check that the trend survives. You are selling change over time, so history you cannot reconstruct later is the one thing you cannot buy back.
Finally, whatever mechanism you land on, the artefact itself should carry enough context to be checkable. A client-facing result that is worth trusting states which question was asked, which engines answered it, when the run happened, and how many runs sit behind the number. A result without a date is a claim; a result without a run count invites the reader to treat one sample as a trend. Those four fields cost nothing to include and they are the difference between a number a client can act on and a number they have to take your word for.
Show one result, not the whole workspace
Frequently asked questions
Do AI visibility tools charge per seat?+
Mostly not any more, and the assumption that they do is the most expensive misconception in this category. Peec AI publishes unlimited user seats on every paid plan, and Otterly includes unlimited team members on every tier. Profound is the notable exception at the entry level, where its published $99 a month plan is single user and tracks ChatGPT only. What almost everyone charges for instead is brands or workspaces and the prompt allowance, so the cost of serving another client is a workspace, not a login.
What does it actually cost an agency to add a client?+
Price the workspace and the prompts, not the people. In practice that means one tracked brand plus enough prompt allowance to measure that client's buying questions across the engines you promised. Because prompt allowances are the metered resource almost everywhere, an agency's marginal cost per client is close to a fixed fraction of the plan rather than a per-head fee, which is why per-seat anxiety usually misprices the deal in the wrong direction.
Should I give a client a login to my AI visibility tool?+
Think about scope before price. A login is all or nothing: it exposes the workspace, the historical runs, the prompts you chose, and often your other configuration, and it invites the client into an interface they will interpret without you. Most reporting relationships need a client to see one result, not to operate the tool. If your only options are a full seat or a screenshot, you are choosing between too much access and too little evidence.
Are screenshots good enough for client reporting?+
They are the most common answer and the weakest one. A screenshot has no provenance, cannot be checked, and ages badly in a category where the underlying answer genuinely changes: one study of 530,875 citations found 40 to 60 percent of cited domains differ a month later for the same question. A client who cannot verify a number will eventually stop trusting the number, and the volatility of AI answers means they are right to be careful.
What is a shareable result link?+
A link to one measured result that works without an account, so a client or a stakeholder can see the finding without being provisioned into the tool. The useful versions are deliberately narrow: they carry a summary rather than the whole workspace, they expire, and they can be revoked. Linkeddit's version shows the summary only, never the underlying answer text or the cited sources, expires 30 days after creation, is never extended by being viewed, and its page is not indexed by search engines.
How many prompts does one client need?+
Published guidance clusters between 20 and 40 to start, and the pruning rule matters more than the starting number: if losing an answer would not plausibly cost the client a deal, or you could not influence that answer within two quarters, the prompt does not earn a slot. For agency pricing the practical consequence is that a client is roughly 25 to 40 prompts of allowance, which is the unit to quote against rather than a headcount.
Related guides
Sources: vendor seat and workspace policies as summarised by Semrush, Sight and Peec AI; volatility figures from the GetMentions AI 530,875-citation study and SE Ranking. Vendor pricing moves quickly; verify every figure on the vendor's own pricing page before you quote it.