The Hidden Markup Tax in AI Chatbot Pricing

If you've shopped for an AI chatbot for your website recently, you've probably seen the same pitch a dozen times: a clean, flat monthly price, no per-chat fees, simple. Then you actually try to use the thing at real volume, and the bill doesn't look flat anymore.

This isn't an accident. It's a pattern common enough across AI chatbot pricing pages that it deserves a name: the markup tax — the gap between the price on the homepage and what you actually pay once you're a real customer sending real volume through it.

Where the markup tax usually hides

It rarely shows up as a single obvious line item, which is exactly why it's easy to miss during a five-minute pricing-page skim. Instead, it's spread across a handful of predictable places:

• Credit or resolution caps. The plan looks flat until you cross a usage threshold, then you're buying add-on credits at a markup over what the underlying AI call actually costs the vendor.
• Per-agent or per-seat fees. Adding a second AI agent, or a teammate who needs dashboard access, often isn't included in the base price — even though the marginal cost to the platform of letting one more person log in is close to zero.
• De-branding and custom domain fees. Wanting your own domain instead of a shared one, or removing "powered by" branding from the widget, frequently costs extra — sometimes more per month than the base plan itself.
• Training-data or knowledge-base size gates. Some tools quietly cap how many documents, or how much total content, you can upload before pushing you onto a higher tier — a limit that has nothing to do with how many conversations you're actually having.

None of this is disclosed as a lie, exactly. It's usually sitting right there on the pricing page, or in the FAQ a few scrolls down. But "flat rate" and "simple pricing" are the headline, and the caps are the footnote. That mismatch between what's advertised and what's structural is the tax.

What this actually looks like on a real bill

Picture a small business running a modest volume of website chats — a few hundred conversations a month, nothing unusual for a local service business or a small SaaS product. On paper, the advertised plan covers this. In practice, a few things tend to happen once real customers start actually using the widget: someone on the team wants their own login, so that's an add-on. The business wants the chat widget to say "Acme Corp" instead of the vendor's name, so that's another line item. A seasonal spike in traffic pushes past the plan's built-in conversation cap, so credits get purchased at a rate well above what the AI call itself costs.

None of these individually feels dramatic. Together, they routinely turn a $39 or $49 "flat-rate AI chatbot" into a bill that's 2-3x the advertised number by the third or fourth month — which is exactly when a business has already built workflows around the tool and switching costs start to feel real.

Why this happens

AI usage is the single biggest variable cost most of these platforms carry. A truly flat, uncapped price means the platform absorbs that variability itself, which is a real financial risk if a customer's traffic spikes unexpectedly. Rather than take on that risk, most platforms build a markup into what looks like AI chatbot pricing — credits, caps, seat fees, or de-branding charges — that quietly passes the variable cost back to the customer while still marketing a flat number up front.

That's a reasonable business decision from the vendor's side. It's just not the same thing as the pricing page's headline claim, and most buyers don't find out about the gap until they're already several months into onboarding, with real customer conversations flowing through the tool.

What actually transparent pricing looks like

There's a simpler model that avoids this entirely: bring your own key (BYOK). Instead of the platform reselling AI usage as credits, you connect your own API key for an AI provider, and your usage is billed directly by that provider, at their published rate, with the platform never touching that part of the bill at all.

Under this model, the platform charges one flat fee for the software itself — the widget, the knowledge base, the dashboard, the integrations. The AI provider charges you separately for the AI usage, at cost. There's no markup to hide, because there's no resale happening in the first place — the platform simply isn't in the business of marking up tokens.

This also means the actual AI cost can be very low, sometimes close to nothing. A few AI providers now offer a free tier generous enough to run a small business's realistic chat volume at effectively $0 in AI costs, on top of whatever the platform itself charges for the software.

Why this matters more as AI gets cheaper

Underlying AI provider costs have fallen sharply over the past couple of years — a large enough drop that the actual per-conversation cost of running a well-built AI agent is now genuinely small for most small businesses. That's good news, but it also means the gap between a vendor's real cost and a marked-up "credit" price is wider than it used to be, not narrower. When the underlying cost of a resource drops but the resale price doesn't move with it, the markup — not the AI itself — becomes the majority of what a customer is actually paying for.

This is exactly why the BYOK model matters more now than it did a couple of years ago: as AI gets cheaper, a platform that bills you directly at cost passes that improvement straight through to you. A platform that resells AI usage as credits has no particular incentive to pass falling costs on — the markup is a business model, not a temporary pricing choice tied to what the AI actually costs today.

A quick checklist before you sign up for any AI chatbot

• Does the pricing page mention a cap on conversations, resolutions, or "actions" anywhere — even buried in the FAQ or a footnote?
• Is adding a second agent, a custom domain, or removing branding included, or priced as a separate add-on?
• Is the AI usage cost billed by the platform (as credits or a markup) or by the underlying AI provider directly to you?
• What happens automatically when you exceed whatever the base plan includes — do you get throttled, silently billed more, or cut off mid-conversation?

If a vendor's pricing page can't answer these plainly, without you having to dig, that's usually the answer in itself.

Try it without the markup

SikloAI runs on the BYOK model described above — connect your own Anthropic, OpenAI, Groq, or Gemini key, and your AI usage is billed directly by that provider, at cost. SikloAI charges one flat fee for the platform itself, and never touches a cent of your AI spend — no credits, no per-agent add-ons, no de-branding fee. The free 14-day trial doesn't require a card, and Groq and Gemini's free tiers make it possible to run the whole thing at genuinely $0 in AI costs while you try it.