Six questions to ask an ad-verification vendor.

"What exactly moves when you say I only pay for real engagement?"

This is the question I have failed more of our own drafts on than any other, so I will start by showing you our own answer and the part of it we got wrong.

“You only pay when it's real” is the easiest sentence in this industry to write and one of the easiest to get subtly wrong, because paying is not one event. There is the moment your money leaves your card. There is the moment a budget is reserved. There is the moment a specific unit of that budget is actually consumed. A vendor can truthfully say “you only pay for verified engagement” while your card was charged weeks earlier.

Ours: an advertiser funds a balance first, and a campaign's budget is held up front rather than collected engagement by engagement. Every submission is graded — automatically, with an independent second check on every decision — before the advertiser is charged for it, meaning before that budget unit is consumed. Rejected and failed engagements are not charged.

Note how specific that last clause has to be to stay true. The review stands in front of the consumption of a budget unit, not in front of your card: by the time it happens, the money has already moved into the hold.

Ask a vendor to name the thing that moves, and when. If the answer is "you're only charged for valid traffic" and nobody will tell you which event "charged" refers to, you have not got an answer yet. This is worth insisting on even when — especially when — the vendor sounds confident.

"What do you actually verify, and what do you merely require?"

These get conflated constantly, including by us.

Requiring something and verifying it are different products. A reading gate that will not unlock the next step for 25 seconds requires 25 seconds of reading. That is not the same claim as measuring how long someone read, and it is not the same claim as knowing that they read it.

I hold our own copy to that line: we may say the 25-second locked read is what the product requires before the quiz unlocks. We may not say we verify or measure 25 seconds of attention on every engagement, because a gate that holds a step closed is not an instrument that measures what happened while it was closed. That distinction is invisible from outside the codebase. Which is the point.

Ask which claims are enforced by a gate and which are inferred from a signal. A vendor who cannot sort their own feature list into those two piles has not audited it.

"What's your take rate, and what else can appear on my invoice?"

Not “what does it cost”. What else can it cost.

“Pay only when” is true of a per-engagement spend and false as a description of a whole invoice, because almost every product in this category has at least one charge that runs on a clock rather than on performance. Ours does: optional Featured placement is $35 a day, charged daily while it is switched on whether or not a single person engages. A monthly campaign bills on a 30-day schedule rather than on engagement.

I will not let a Nexertise asset carry “pay only when” as its only statement about money, for exactly that reason — on its own it reads as “no other charge exists”. So, in full: engagements cost $5, $15, $30, $50 or $75 depending on the professional's tier, the minimum campaign budget is $75, the platform's take is 30% and the professional keeps the other 70%, and Featured placement is that separate $35 a day.

Ask for the clock-based charges by name. Then ask what the take rate is, and whether the number you were quoted is the take or the price.

"Who reviews the edge cases, and how long do they take?"

“Reviewed” can mean a person, a model, a rule, or a sample of any of them, and vendors rarely say which unless you ask.

Ours: every engagement — not a sample, not only the flagged ones — is graded automatically against one quality standard, with an independent second check on every decision. A professional who contests a rejection gets a separate automated re-check, and members in the EU, EEA and UK can ask for a person to review the decision. I would rather publish exactly that than a vague “our team reviews”, because a buyer can plan around a stated process in a way they cannot plan around “rigorous”.

Ask what reviews each decision, whether it is every engagement or a sample, and where a disputed decision goes next. A vendor who will not say which is usually reviewing a sample.

"What happens to a decision I disagree with?"

Ask for the list of possible outcomes, and ask whether that is the complete list.

I banned a phrase from our own copy this month because a draft offered a professional a middle outcome — a “reduced” penalty — on an appeal that the code does not have a middle outcome for. The code overturns or upholds. Offering a third thing would have been a kindness that did not exist.

Ask for the enumerated outcomes. Then ask what the appeal rate is. If every decision stands, either the detection is perfect or nobody is really reviewing.

"Which of these claims would survive someone reading your source code?"

This is the one I would actually open with, and it is the reason the other five work.

Every number in this piece is one I can point at a line of code or a founder's written decision for. That is not a virtue I am claiming over other vendors — it is a method, and you can make any vendor submit to it. We keep a register of every claim we make, sorted into what is proven, what is waiting on confirmation, and what is banned because it turned out to be false. The banned list is the useful one.

I am telling you our banned list exists because a vendor whose banned list is empty has not got one. Nobody writes thirty claims about their own product and gets all thirty right. The question is whether anyone whose job is to say no ever read them.

Ask a vendor what they have had to stop saying, and when. It is the single most informative question on this list, and almost nobody asks it.