How to price AI visibility work for clients: audit, retainer or results-based

Here’s an illustrative version of a question that comes up constantly once an agency has run its first AI visibility audit for a client: the audit surfaced real findings, the client is convinced something needs to happen, and now the agency has to say what happens next and what it costs. Sell another audit? Move to a retainer? Tie the fee to the score actually moving? Those are three different structures with three different risk profiles, and most agencies default to whatever they already sell for SEO without asking whether AI visibility work is shaped the same way. It mostly isn’t, and getting the pricing model wrong tends to show up six months later as a client who feels oversold on a number nobody explained properly.

Why this isn’t the same pricing conversation as SEO

The reflex to reach for a flat monthly retainer is a reasonable one. Ahrefs polled 439 SEO providers and found 78.2% bill through a flat monthly retainer, with the most common band landing at $501 to $1,000 per month (20.4% of respondents) and agencies averaging $3,209 per month overall. Retainer is already the default shape of a services relationship, and it’s tempting to carry that shape over unchanged.

The problem is that a retainer’s pitch rests on a promise that keeps getting easier to prove over time, more content indexed, more links earned, rankings that move up and largely stay there once they move. AI visibility does not behave that way. A presence score can swing meaningfully between two reporting periods with nothing about the brand having changed at all, which means the same retainer pitch needs a different kind of evidence behind it, or it collapses the first time a client asks why last month’s gain disappeared.

What an audit is actually good for, once

An AI visibility audit is a point-in-time diagnostic: where the brand stands across a fixed prompt set, which sources engines are citing instead, and what’s broken in the entity data or the site itself. That’s genuinely valuable and it’s also, by design, a thing you do once and then stop doing until something material changes, a new product line, a rebrand, a competitor’s launch. Selling it as a standing subscription misrepresents what it is. The honest sales motion is to price it as a fixed-fee diagnostic that produces a prioritized list of fixes, and to treat “what do we do about the list” as a separate, second conversation.

That second conversation is where most of the actual revenue should live, because the list an audit produces is never a one-and-done task. Entity data drifts, competitors ship new content, engines change how they retrieve answers. A client who pays for an audit and then tries to execute the fix list themselves, on no schedule, with no one checking whether anything moved, is buying a report rather than a result.

The retainer: what it actually buys, and why it fits the work that follows

A retainer earns its keep here because AI visibility work is genuinely ongoing in a way a one-time audit list understates: tracking has to run on a fixed cadence to mean anything, fixes have to be prioritized against what’s actually moving, and someone has to explain the number to the client every month in a way that survives scrutiny. Our guide on reporting AI visibility to clients covers what that monthly report needs to contain, and the same logic that makes a single blended number misleading for a multi-location client applies to a single-brand retainer too: the client needs to see the baseline, the change, who’s being named instead, and what was done about it, not just a number that moved.

Price the retainer for that ongoing work, tracking, diagnosis, prioritization, monthly reporting, not for a promised score. What you’re selling is a standing capability to measure and respond, which is a real and durable thing to sell even in a month where the topline number doesn’t move.

Why results-based pricing is tempting, and why it breaks against a noisy score

Tying fees to outcomes is the model every client secretly wants and most agencies avoid, and the caution is well founded even outside AI visibility. Across the broader agency market, a survey by SoDa and Productive found project-based fees account for roughly 50% of agency revenue, retainers 44%, hourly billing 30% and value-based or performance pricing only about 10%, and separate research from the Wow Company’s BenchPress survey found that only around 17% of consultancies use value-based fees at all, largely because pricing them accurately and managing the risk at scale is hard even when the underlying outcome, a completed project, a closed deal, is unambiguous once it happens.

An AI visibility score is a much harder thing to build a bonus around than a completed project, because the score itself is noisy even when nothing real has changed. Ranqo’s analysis of 851 tracking runs across 81 brands, with no fixes shipped and no prompt set edited between consecutive checks, found overall scores still swung by up to 14.5 percentage points at the 95th percentile, and a single category-level cut swung by up to 27.3 points, nearly twice as noisy, because a narrower slice has fewer answers behind it and a single flipped answer moves it further. Our own piece on why AI visibility tools report different scores covers the same underlying fragility from a different angle: the number is the output of several compounding decisions, not a single stable measurement.

A results-based fee written against a single month’s reading is a fee written against noise, not against work. Pay a bonus because a score happened to read high in one particular week and you’re rewarding the arithmetic, not the agency. The same analysis has the fix already built in: individual prompt-engine results were actually stable, flipping only 6.4% of the time, and a mention confirmed on two consecutive runs survived to the next check 96.1% of the time, against just 54.3% for a mention seen only once. Noise mostly lives in single readings, not in results that repeat.

A structure that survives scrutiny

The workable version keeps the retainer as the base and adds a narrow bonus on top, rather than replacing the retainer with pure performance pricing. Base the retainer on the ongoing work, tracking, diagnosis, fixes, reporting, the same work it would cover with no bonus attached. Then define the bonus against mentions confirmed on two consecutive tracking runs against a fixed baseline, never against a single reading, and settle it quarterly rather than monthly so one noisy month can’t trigger a payout that a calmer month would have denied. That single change, confirmed rather than raw, turns a bonus structure from something a client will eventually catch gaming the noise into something that actually tracks whether the work did anything.

It’s also worth offering the client a choice rather than a single number. The same BenchPress research found that agencies offering three pricing options were twice as likely to see conversion rates above 60%, which matches ordinary sales experience: a menu closes better than a single take-it-or-leave-it fee, even when most clients end up choosing the middle option.

What to actually sell, in order

Start with the audit, priced as the fixed-fee diagnostic it is, not as a foot in the door for a retainer you haven’t pitched yet. Once the audit produces a fix list, pitch the retainer to execute and track it, priced for the ongoing measurement and reporting work rather than for a promised score. Only add a results-linked bonus after the retainer has run long enough to establish a real baseline, at least two confirmed reporting cycles, so the bonus is measured against something that has already survived the two-run confirmation rule rather than against whatever the first month happened to read.

If you’re running this playbook across more than one account, see how it fits agencies managing several clients at once on our agencies page, or run a free Visibility Check against a prospective client’s brand before the first pricing conversation, so the audit pitch starts from a real number instead of a guess.