The contract that makes AI-visibility work payable on results.

Why this contract could not exist before

AI answers vary run to run. A vendor can always claim a win that is pure chance. A client can always dispute it. That is why performance deals in this category have stayed rare, informal, and dispute-prone: nobody had an instrument that separates a real change from the model's own noise. Nadelio's bounded score, the same one behind every free audit on this site, is that instrument. This agreement is what it makes possible.

The three parties

The clauses

1. Fixed question setThe brand selects up to 25 buyer-intent questions at signature, chosen for business value, not ease of winning. The set cannot change during the measurement window. Why Pay-for-performance SEO died mostly from agencies chasing whatever was cheapest to move, not from measurement disputes. Fixing the set at signature, by the paying party, closes that door before it opens.
2. Control groupA set of untreated competitor brands is measured on the same schedule, in parallel. A shared swing across the whole group (a model update, a seasonal shift) is attributed to the model, not the campaign. Why AI models get updated without notice. A score can move for everyone at once. Without a control group, that looks exactly like a win or a loss it is not.
3. 95% confidence thresholdA gain is recognized only when the post-campaign interval and the pre-campaign interval no longer overlap: before 44±9, after 61±5 clears; before 44±9, after 48±11 does not. Why This is the same bound published on every Nadelio measurement. Nobody pays on a coin flip, and nobody can dispute a rule accepted in writing before the campaign started.
4. Model-version lockIf the assistant being measured ships a material update mid-window, the window pauses and restarts. A dependency the agreement points at, not one either party has to argue about after the fact. Why The clearest way this contract could become unarbitrable is a model update landing mid-campaign. Locking the window removes the argument before it starts.
5. Source qualityCitations traced to link farms, paid placements not disclosed as such, or sources unrelated to the fixed question set do not count toward the gain. Why The other cause of death for old performance-SEO deals was gaming the metric with junk sources. This clause exists so the instrument cannot be fed noise dressed up as signal.
6. Settlement window60 to 90 days, set at signature. At the end of the window, Nadelio publishes the verdict: cleared or not, with the full measurement trail, timestamped. Why Open-ended windows invite renegotiation. A fixed date makes the verdict a fact, not a discussion.
7. Flat settlement feeNadelio is paid 300 to 500€/month per brand measured, for the act of measuring and settling, regardless of whether the gain clears. Never a percentage of the bonus, the retainer, or the outcome. Why A referee paid on the verdict is not a referee. This is the one clause that cannot be negotiated away without the whole agreement losing its point.

Where this stands today

This is version 1 of an open template, published so any brand or agency can read exactly what they would be signing before a conversation starts. The first agreements are being signed now, in one sector, in France. Every settlement, cleared or not, becomes part of a public record: the point of an arbitrator is a track record nobody can fake retroactively.

Not legal advice. Intended as the commercial skeleton of the deal; your counsel adapts the language to your jurisdiction and adds what your situation requires.