What AI Customer Support Actually Costs in 2026: Every Published Rate, Compared

Every major support vendor now charges something per AI resolution. The rates are public, they differ by roughly four times, and almost none of them are the whole bill.

Here is what is published as of September 2026. Verify each figure on the vendor’s own pricing page before you budget against it, because this category is changing faster than any comparison article can keep up with, including this one.

The published rates

Two things stand out. The spread is about four times between the cheapest and the dearest. And the column on the right is where most of the real money is.

The rate is not the price

This is the single most common budgeting error in this category.

A $0.99 resolution rate on a platform charging $29 per seat is not cheaper than a $1.50 rate on a platform charging nothing per seat, until you know how many seats you have and how many resolutions you run. At five agents and two thousand monthly resolutions, the seat line is $145 and the resolution line is $1,980. At twenty agents and five hundred resolutions, the ratio inverts entirely.

So the only comparison that means anything is total cost divided by issues genuinely resolved. We wrote up how to build that model, and it takes about an hour with numbers you already have.

The definitions did not converge

Every vendor adopted per-resolution pricing. None of them adopted the same definition of a resolution.

Zendesk now bills on what it calls verified resolutions, confirmed by an evaluation step rather than self-reported, which it announced alongside its Resolution Platform at Relate 2026. That is a meaningfully stricter definition than counting any conversation the customer did not escalate, and it is the direction the category should go.

Others count differently. The questions that actually determine your bill:

• Does an abandoned conversation count as resolved?
• Is a conversation handed to a human billed?
• Does a reopened conversation retract the charge?
• Is there a hard spend cap, and what happens when you hit it?
• Can you audit a sample of billed resolutions?

A vendor quoting a 70% resolution rate and one quoting 35% may be describing identical performance under different definitions. We wrote about why the definition is the contract.

Watch the overage terms

Zendesk introduced automatic overage billing in January 2026, charging teams that exceed their committed monthly resolution volume at the per-resolution rate. Gorgias publishes a separate, higher overage rate of about $1.50.

Overage is where usage pricing stops being predictable. An incident, a viral post, or a carrier disruption can multiply volume within hours, and committed-volume discounts turn into penalties exactly when you are least able to respond.

Ask what the overage rate is, whether you are notified before crossing the threshold, and whether there is a ceiling. A hard spend cap turns a variable bill into a budgeted line. Without one, the answer to “how much could this cost” is genuinely unbounded.

Where the seat fees went

One notable change: Zendesk absorbed its $50 per agent Advanced AI add-on into its Suite and Support plans rather than selling it separately. That is a simplification and it is worth reading carefully, because absorbing an add-on into a plan usually means the plan price reflects it.

Freshdesk went the other way, keeping Freddy AI Copilot as a $29 per agent add-on available only on higher tiers, alongside separately metered AI Agent sessions.

The pattern across the category is seats plus usage, with the balance between them varying by vendor. Fidiora is the outlier here in charging no seat fee at all, which is a deliberate position rather than an oversight: if you are paying for outcomes, paying again for the number of people who can log in is double billing.

How to actually compare these

1. Get your monthly resolved issue count. Not tickets closed. Issues genuinely resolved.
2. Get your seat count, including everyone who needs visibility, not just full agents.
3. For each vendor, add every line: seats, AI usage at your volume, add-ons you would actually enable, and channel fees if any.
4. Divide by resolved issues. Now you have one comparable number.
5. Model your worst month, not your average, and check what the overage terms do to it.
6. Read each resolution definition and score the rates against what they actually count.

Most teams doing this for the first time find the ranking changes once step three is complete.

What we charge and why

Since this is our site, the disclosure belongs here rather than in a footnote. Fidiora charges $0.59 per genuine resolution, with no seat fee, no add-on modules, and no channel surcharges. Abandoned chats, silent timeouts, and any conversation handed to a human are never billed. There is a hard monthly spend cap you set yourself.

The definition is published in full at how we define a resolution, and the machine-readable version is at /pricing.md so AI buying agents can read it without a sales call.

We are obviously not neutral about which model is better. The part of this post we would stand behind regardless of vendor: the rate is not the price, the definition is the contract, and you cannot compare any of these without converting them to cost per resolved issue at your own volume.

Sources: vendor pricing pages for Zendesk, Intercom, HubSpot, Salesforce Agentforce, Gorgias and Freshdesk. Rates change; check before you budget.