If a buyer cannot estimate the bill, your AI pricing page is broken

Your AI pricing page has one job. Let a buyer estimate what they will pay next month, in units they already count.

Most AI pricing pages fail that job. They list credits, tokens, or "AI actions," then a "Contact sales" button on the tier anyone serious would pick. The buyer does the math, cannot finish it, and leaves.

If the buyer cannot estimate the bill, you did not publish pricing. You published a demo request.

Why is AI pricing harder to read than SaaS pricing?

Because the unit moved from people to work, and most pages never translated it.

Seat pricing was easy to read. Ten reps, ten seats, one number. AI products bill on usage because your costs are usage. Fair. But the unit on the page is yours, not theirs.

A buyer does not know how many credits a month their team burns. They know how many tickets they close, accounts they research, or calls they run. If your page does not bridge those two, the buyer has to guess. Buyers who guess on cost guess high, and guessing high kills deals.

What happens when model costs keep falling?

Buyers notice, and your page has to explain why your price did or did not move.

TechCrunch reported on Sep 22 that OpenAI priced its new GPT-6 API tier at roughly half of GPT-5.6 API pricing. On Sep 28, Anthropic pitched Sonnet 5.5 as cheaper and lighter on token burn. Your buyers read the same headlines.

So a credit price with no story reads like you kept the savings. Say what the price pays for. Support, accuracy work, guardrails, the workflow around the model. If the page only shows credits, the buyer compares you to the raw API bill, and you lose that comparison every time.

What does a pricing page that lets buyers estimate look like?

Four parts. Most AI pages ship one.

1. A unit the buyer already counts. Translate credits into work. "One credit is about one researched account" beats "1,000 credits per month." Pick the job your product actually does and price in that job.
2. A worked example bill. Show a real-looking team: five people, 400 researched accounts a month, and the dollar total they would pay. One example per tier. Let the buyer find themselves in it.
3. What happens at the limit. Overage rate, hard cap, or auto-upgrade. Say it plainly. The fear of a surprise bill is what sends buyers to "Contact sales," and most of them never click it.
4. A public price on the tier people actually buy. Hide the enterprise tier if you must. Do not hide the one a 20-person company needs.

This is the same filter as the grunt test. What do I get, what does it cost, why trust you. If a stranger cannot answer the cost question in one breath, the page fails.

Why does this matter more than the homepage?

Because the pricing page is where the in-market buyers show up.

Ehrenberg-Bass and the LinkedIn B2B Institute put about 95% of B2B buyers out of market at any given time. The 5% who are buying visit pricing. That page gets the smallest and most valuable audience you have.

If you are running ads to a page that cannot answer "how much," you are paying for the traffic and then refusing to quote. That is grow-stage spend before the page can close. The spend is not the problem. The page is.

Should you price on outcomes instead?

If you can define the outcome, yes. If you cannot, fix the unit first.

TechCrunch reported that Ema prices on task outcomes rather than seats or tokens, alongside about 180% net dollar retention. I wrote about the buyer side of that in price the GTM outcome, not the AI seat. The seller side is simpler. Outcome pricing only works when you and the buyer agree on what counts. A resolved ticket. A booked meeting. A completed workflow.

Most early AI products cannot define that yet. A clear unit and an honest example bill beat a vague outcome promise.

What should I change this week?

Open your pricing page and time yourself as a buyer. Pick a team size and a workload. Estimate the bill.

If it takes more than a minute, rewrite the unit into work the buyer counts. Add one worked example per tier. Write one sentence on what happens at the limit. Put a real number on the tier your best customers are on today.

Then watch your CAC. Fewer calls that exist only to quote a price. Faster yes from buyers already in the market.

Price it so a stranger can do the math. Or keep paying sales to do it for them.

Start Signals, Convert, Grow

FAQ

What should an AI pricing page show?

A unit the buyer already counts, one worked example bill per tier, what happens at the usage limit, and a public price on the tier most customers buy.

Are credits a bad pricing unit?

Credits are fine on the invoice. They fail on the page when there is no translation into work, such as accounts researched or tickets resolved.

Should I hide pricing and force a sales call?

Hide the enterprise tier if the deal is truly custom. Hiding the tier a small team needs turns in-market buyers into people who just leave.

Do falling model costs mean I should cut my price?

Not always. It means you should say what your price pays for beyond the model, like accuracy work, guardrails, and support. Otherwise buyers compare you to the raw API bill.

Is outcome pricing better than usage pricing?

Only when you and the buyer agree on what counts as the outcome. Until then, a clear unit and an honest example bill work better.