What you are not charged for

Technical information first: a one-time report on AEOSearch is charged for the answers it delivers and never for the calls that died. When a call to ChatGPT, Claude, Gemini, Perplexity or Grok fails, the run keeps re-asking it for thirty minutes, no more often than every five, because an outage is not fixed by hammering it. What still has no answer when that window closes is refunded pro rata, weighted by what each unit costs, so a dead Gemini call hands back less than a dead GPT call, using the same weighting the quote used to charge for them. You don't ask. The refund is written against the payment the checkout recorded, and if Stripe refuses it the row is marked owed and we are alerted to issue it by hand. On a monthly report there is no single payment to refund against, so the same tally reaches us as a credit owed on the period, and we apply it by hand.

We built it that way because the alternative, which we have seen sold, is worse than a lost call: a failed call printed as a zero and handed to you as your invisibility. We wrote up our own failures last month, 160 of them, most of them ours. The refund is the billing half of that promise. A failure is disclosed on the report as a failure, excluded from every metric, and taken off the bill.

Every line says what it is

Open the configurator and the rail lists the bill before you have paid it: one line per engine, with the model version named beside it and its price for the questions you chose; a $2 line called "Questions and diagnosis", which is the question generation and the gap diagnosis, the fixed work that doesn't scale with the engines; a $50 line per competitor report you ticked; and the 90-day action plan at $150 or, on the monthly, marked included. The Stripe line item says the same thing in words: "Custom AI visibility report: 20 questions on GPT-5.6 Sol, Claude Sonnet 5, Gemini 3.6 Flash", with the competitor reports and the plan appended if you took them. The model is on the invoice because the model is on every stored answer, and a bill that named less than the report does would be the odd one out.

Rounding, in advance: a quoted price rounds up to the next $5, never down, and no report sells under $25. That's a positioning decision rather than a cost one, and it lives in the same file both the page and the server read, so the number can't be one thing on the button and another at checkout. A plain rule you can predict is worth more to us than a cleverer one you'd have to trust.

What you're also not paying for

A model we've never run. Every rung on the page is labelled measured cost or list price: measured is the top of what a grounded answer cost us on our own account, list price is the vendor's published rate scaled by the token profile of its measured sibling, with a margin over it, for a rung we've run but not yet probed. A rung our adapter has never answered on at all is shown as not yet available and can't be bought, so nothing sells on a model we've only read about, and the label tells you which kind of number you're paying. A competitor report doesn't re-ask an engine: it reads the answers already stored for your run with the other name as the subject, which is why it is a flat $50 whatever the report above it cost. And the plan's own model call is paid by its $150 line, not folded into the per-question rate, so a buyer who skips the plan is not quietly paying for it. The refund, for the same reason, is measured against the report alone: competitor reports and the plan are not engine calls, so they are never the thing that failed, and a settlement never reads them back.

What this doesn't tell you

The bill is honest about what ran. It is not a promise about what the engines will say. A refund covers a call that never answered; it does not cover an answer you didn't like, and the engines don't name you more because you paid for five of them. The per-question rate is built from the top of the cost band we've measured on each model, so a run where a model searched less than usual is a run where we kept a little more, and one where it searched more is one where we kept a little less: the price is fixed before the model decides how much work to do, and the model does not consult us. And a pro rata refund is proportional to cost, not to importance: the one question you most needed answered, if it is the one that died, hands back the same as any other on that engine.

The incentives, stated: we profit from every line in this post. A bill people trust is a bill people pay again, the monthly report costs the same as the single one because the re-run costs us the same, and the transparency we're describing is also a sales argument, which is why we've tried to put every claim here beside the rule that makes it true rather than beside an adjective. The rules are in the code that prices the page and settles the run. If we have described them wrongly, the receipt will show it, and we'd rather be caught by a receipt than believed on a paragraph.

What you can do with it

Build the report you actually want, read every line before you pay, and know that any call that dies comes off the bill on its own. The engines start within a minute of your approving the questions; the report opens on your dashboard and is emailed the moment it lands, with the failure banner on it if there's anything to disclose and nothing on it if there isn't, which is a measurement too. Our methodology is public, and so, now, is our bill.

If you'd like the fix list turned into work, we do it at our agency rates and the next report grades us on the same questions. We look forward to growing with you, and our agents are at your beck and call.