The trade press now tells dealers to audit for AI. That is the right instruction aimed at the wrong moment. An audit is a comparison between what happened and what should have happened, and it only works if someone wrote down what happened while it was happening.
Most dealers cannot do that comparison today. The AI running their advertising, whether it sits inside an agency or a vendor platform, leaves behind a dashboard. A dashboard shows a state. A ledger shows a sequence of decisions, each with an author, a reason and a time. Those are different objects, and only one of them survives an audit.
What Is Actually Being Said About Auditing AI?
CBT News reported on October 26 that David Spisak is telling dealers they must audit for AI, and that he is launching a new show on the outlet. The coverage frames the audit around visibility: whether a store can be found when shoppers ask machines instead of search boxes.
"Shoppers are using AI tools to research vehicles before they reach a dealership website, and many stores are not set up to be found." CBT News | #1 Source for Automotive News & Dealership Intelligence
That is a real audit target, and it is the outside-in half of the problem. How a store appears to AI answer engines is something you can inspect from the street. Our own view of that half is in our piece on AI as the test before the drive.
The inside-out half gets less airtime. Dealers also have AI working for them: writing ad copy, moving budget, pausing campaigns, composing offers. When a GM is told to audit that AI, the first question is not what to look at. It is what exists to look at.
Why Can't You Audit a Dashboard?
A dashboard is built to answer "how are we doing right now." It overwrites itself. Yesterday's budget split is gone the moment today's loads, and the reason the split changed was never a field in the first place.
An audit asks a different set of questions. Who changed this? Was it a person, a script or a model? What did they know? What rule applied? Was the change inside the dealer's approved limits? None of those answers live in a performance view.
The agency version of this is the monthly review. A strategist assembles a deck from memory, exports and whatever the account manager remembers. The deck is a story told after the fact by the party being reviewed. It is not evidence. Nobody would accept that standard from a dealership's own controller.
The vendor version is the black box. The platform says its model optimizes budget, and the dealer sees outcomes but never the decisions. When an offer prints wrong or spend lands in a channel the dealer never wanted, the dealer asks why and gets a support reply that reconstructs a plausible account. Plausible is not recorded.
What Does an Audit-Grade Record Have to Contain?
Four things, written at the moment of the decision, not assembled later.
First, the actor. A named person, an AI acting on behalf of a named person, or the system itself. Those three must never blur together, because accountability changes with each.
Second, the reason, in the words used at the time. A reason reconstructed in a quarterly meeting is a rationalization.
Third, the time, in the store's own clock, so that a decision can be lined up against an OEM program change, a price move or a compliance event.
Fourth, tamper evidence. A log that anyone can edit proves nothing. The record has to be chained so that an edited or removed entry is detectable.
Take away any one of the four and the audit degrades into an interview. The dealer asks the vendor what happened and believes the answer or does not.
Why Does the Agency Review Fail This Test?
It fails on the first element. An agency's work is done by people across several tools, and the decisions are scattered across ad platform change histories, email threads and chat messages. Nothing joins them into one record the dealer can read.
It also fails on incentives. The party writing the review is the party under review. A monthly cadence compounds the problem: a bad change on the 3rd gets described on the 30th, after the money is spent.
None of this requires bad faith. It is structural. A retainer buys labor and a narrative about the labor. It does not buy a ledger. A dealer who wants to know who approved a budget move or why an ad ran a payment without its terms has to ask a person to remember.
Our argument in the hand-off piece applies here too: the record of the hand-off is the product, not the automation around it. The same is true of an ad account. The automation is table stakes. The record is what lets a dealer trust it.
What Does an Audit Trail Look Like When It Is Built Into the Platform?
Every dealer-impacting decision AEGIS makes is hash-chained through AXIOM, and the dealer can read that trail.✓ Oct 27 The ledger names who acted: a person, AEGIS acting for a person, or the system.✓ Oct 27 AEGIS's own work is recorded as AEGIS with the run that did it and, when a dealer asked for it, the verified person it acted for, so the Vault reads "AEGIS (for <first name>)".✓ Oct 27 Schedulers read "System," and a person's own click or authenticated email carries their name.
That distinction is the whole point. When a budget moves, the dealer can see whether a human did it, whether the AI did it on a human's written instruction, or whether a scheduled process did it. Each has a different owner.
Asked who changed something and why, AEGIS reads the store's own decision ledger and reports the actor, the time in store time, the reason recorded at the time word for word, and whether the chain is intact.✓ Oct 27 When no reason was recorded, it says so.✓ Oct 27 A system that admits a gap is more credible than one that never has any.
Dealer controls leave the same kind of trace: every lock or unlock of a budget allocation is hash-chained in the decision audit trail.
Dealer controls leave the same kind of trace: every lock or unlock of a budget allocation is hash-chained in the decision audit trail. When a verified person on the dealer's account writes the approved monthly figure to AEGIS, the change is recorded on the dealer's decision ledger under that person's name, with their own words as the reason. The dealer's authority and the AI's actions sit in one sequence.
The dealer's authority and the AI's actions sit in one sequence.Does the Record Matter Before Spend or Only After?
Both, and the "before" half is the one agencies cannot offer. An audit trail that only documents mistakes is a postmortem. A trail that sits behind a gate prevents some of them.
AXIOM runs a three-stage compliance triad, strategist, composer, verifier, over every ad copy and landing-page assertion before spend is approved.✓ Oct 27 Dealers can also write their own house rules in plain English, such as "never quote a monthly payment," and those rules are read by every copy composer and reviewer before anything ships.
A record only helps if the AI's own account of its work can be tested against it. A statement the AI makes about what it did, whether to a dealer or in its own notes, is worth only as much as the evidence behind it. That is why the ledger and the pre-spend checks above belong together: one records what was done, the other keeps unsupported statements from standing in for it.
That matters for audits because the AI's own statements about what it did are checked against what it did.Price and offer pages show the same logic from the outside. We made the case in the hidden-fees piece that the agency which audits ads and never opens the landing page is auditing the wrong surface. A ledger plus a pre-spend gate is how you stop auditing the wrong surface.
The AUTONOMi Approach: A Ledger Before a Dashboard
AXIOM is the policy engine that gates every AEGIS action, built on four pillars: exposure, action, auditable and authority.✓ Oct 27 The auditable pillar is the ledger described above. The authority pillar keeps the chain of command legible: strict tier descent runs from the orchestrator to domain agents to sub-agents to tools.✓ Oct 27
AXIOM also holds spend ceilings per dealer per day, and the monthly approved-spend lock pauses the account when counted spend reaches the approved monthly.✓ Oct 27 The delete action is always blocked by default.✓ Oct 27 Limits are written into the platform, not promised in a deck.
The ad accounts, analytics properties and tag containers are dealer-owned, and AEGIS works through delegated access the dealer can revoke. The hash-chained trail lives on the AUTONOMi side but records actions taken, not the underlying customer records those actions touched.✓ Oct 27 The dealer keeps the accounts and can read the record of what was done to them.
This is not a claim that nothing ever goes wrong. It is a claim about what you can do when it does. You can read who acted, why, and when, and you can tell whether the chain is intact. That is the minimum an audit needs.
What Should a Dealer Ask Before Signing the Next Renewal?
The audit instruction is going to spread, and it will reach ad spend. Boards, lenders and OEM field staff will ask GMs the same question: show us what your AI did and why. The dealers who can answer in minutes will look disciplined. The dealers who have to request a deck from the agency will look exposed.
Put four questions to every agency and vendor that touches your accounts. Can I read a record of every change, with the actor named? Does the record carry the reason written at the time? Can I tell whether the entry has been edited since? And does a compliance check run before money moves, not in the next monthly review?
If the answers are a dashboard login and a promise to prepare a report, you have a reporting arrangement, not an audit trail. The loser here is the monthly review and the black box, because both ask for trust where a ledger offers evidence. If you want to see what a decision record looks like on your own accounts, sign up and read the ledger for yourself.
Source: CBT News | #1 Source for Automotive News & Dealership Intelligence
