# We Tested Checked It Against Two Real ABAC Panel Rulings. It Matched Both in Under 90 Seconds. > We tested Checked It against two real 2026 ABAC breach rulings. It matched the Panel on both, cited the same provisions, and scored 92 and 82. Source: https://www.hiaitus.ai/posts/checkedit-vs-abac-panel-rulings _Published 2026-08-05_ Can an AI compliance check match a real ABAC Adjudication Panel finding? There is only one honest way to answer that. Take rulings the Panel has already published, run the same creative through the engine cold, and compare. So we did. Two upheld determinations, both from 2026. Both matched. **The short version** - Two published ABAC Adjudication Panel determinations, both upheld as in breach. - Checked It matched both verdicts. Kendor Distillery scored 92 out of 100. Neu Lemon Crush and Vodka scored 82. - Both checks returned in under 90 seconds. - We fed the engine screenshots taken from the published rulings, not the original creative, to make the test harder than real conditions. - Both determinations turned on the same provision of the ABAC Responsible Alcohol Marketing Code: Part 3(c)(iv). ### First, who actually decides **ABAC is the authority here**. The Adjudication Panel makes the final call, and human judgement leads. Nothing in this test suggests otherwise, and any tool that claims to replace a regulator's determination is selling you something you **should not buy**. Checked It exists to get you to that point sooner. It is a pre-screen, run before creative goes out, so that a breach is caught while it is still cheap to fix rather than after a complaint has been lodged and adjudicated. ### What did we test, and how? Two ABAC determinations from 2026, both upheld. For each one we took the creative as it appears in the published ruling, which means screenshots, some of them low resolution, rather than the original production files. That was deliberate. If the engine performs on degraded stills, it performs on the real thing. Every number below is an accuracy score. It measures how closely the Checked It assessment matched the Panel's published finding. A score of 100 would mean an exact match on three things at once: the breach identified, the provision cited, and the reasoning behind it. Getting the verdict right is not enough on its own. ### Ruling one: Kendor Distillery, ABAC 26/26 A social media post comparing drinking the product to a therapy session. Cheaper than therapy, and by implication, better. The Panel upheld it. In its determination, the Panel found that the post "creates a direct comparison between drinking the product and a therapy session, the clear import is that the product is cheaper and arguably more effective than therapy," and that a reasonable person would understand it as suggesting alcohol offers a therapeutic or mental health benefit, in breach of Part 3(c)(iv). Checked It flagged the same breach and cited the same provision, Part 3(c)(iv). It also called out the "we're not saying we replace therapy" disclaimer as ineffective, which is the part most first-pass reviews let through, because a disclaimer reads like a fix even when it does not undo the impression the creative has already created. Score: 92 out of 100. ### Ruling two: Neu Lemon Crush and Vodka, ABAC 98/26 A social media video from Deltaforce Drinks positioning the product inside a Pilates session. The Panel upheld it, finding that a reasonable person would likely understand that the video positions the product as offering a health benefit. Same provision again: Part 3(c)(iv). Checked It returned a Failed Check on the same creative, working from grainy, low-resolution stills. Score: 82 out of 100. The verdict matched. The eighteen points sit in how closely the reasoning tracked the Panel's, which is exactly what you would expect when the input is a degraded screenshot rather than the ad itself. ### Why did both rulings turn on the same provision? Because Part 3(c)(iv) is where modern alcohol creative keeps landing. The provision says an alcohol marketing communication must not suggest that consumption offers any therapeutic or health benefit. Read that against how alcohol brands currently talk to people. Wellness framing. Self-care. The drink as the reward at the end of a hard week, or the thing that goes with the workout rather than against it. Neither ad set out to make a health claim. One made a joke about therapy. The other put a drink next to a Pilates mat. Both landed in the same place, because the test is not what the brand intended. It is what a reasonable person would understand the communication to be suggesting. That is a hard test to self-assess against, because the people closest to the work are the least able to see it fresh. It is also the kind of test a clause-level pre-screen is well suited to, since the machine has no investment in the idea. ### What the scores do and do not mean They do not mean the tool is 92 per cent right about alcohol advertising. They mean that on these two determinations, working from degraded inputs, the assessment tracked the Panel's published reasoning closely. Two rulings is a small sample. We will keep publishing these as we run them, including the ones where the score is worse, because a benchmark you only publish when it flatters you is marketing, not evidence. What the test does establish is that the failure mode people worry about, an AI confidently clearing creative a regulator would go on to uphold, did not occur here. Both were caught. Both were caught before the complaint would have existed. ### The point of pre-screening _It is not replacing legal review. Your legal and compliance team keeps the judgement and the sign-off._ It is catching the breach before there is a complaint to answer. Before the media spend. Before the screenshots end up in a published determination with your brand name attached to them. ### Frequently asked questions **Can AI check alcohol advertising against the ABAC Code?** Yes, as a pre-screen. Checked It assesses creative against the ABAC Responsible Alcohol Marketing Code and returns findings tied to specific provisions. In this test it matched two upheld Adjudication Panel determinations, citing Part 3(c)(iv) correctly. It does not replace the Panel, which makes the final call. **What is ABAC Part 3(c)(iv)?** It is the provision of the ABAC Responsible Alcohol Marketing Code stating that alcohol marketing must not suggest consumption offers any therapeutic or health benefit. Both determinations in this test were upheld under it, one for a therapy comparison and one for positioning a drink within a Pilates session. **Does an AI compliance check replace legal review?** No. It is a pre-flight pass with a human in the loop. The engine does the repetitive first-pass triage and cites the provision behind each finding. Your legal and compliance team keeps the judgement and the final sign-off, and spends its hours on the calls that actually need them. **How long does an ABAC compliance check take?** Both checks in this test returned in under 90 seconds, working from screenshots rather than original creative files. That is fast enough to run at brief and storyboard stage, which is where a finding is cheap to act on, rather than after production spend is committed. **Why use screenshots instead of the original ads?** To make the test harder than real conditions. The published determinations only include stills, some of them low resolution. If the engine identifies the breach from degraded images, it will do at least as well on the creative files a marketing team would actually upload. **What happens if ABAC upholds a complaint against your ad?** The ad comes down, and the determination is published with the advertiser and product named. The direct cost is the withdrawal, the reshoot and the wasted media spend. The indirect cost is a permanent public record of the breach, searchable by anyone including your next media partner. Five free checks are available if you want to test it against your own creative. Details at [checkedit.ai](https://www.checkedit.ai), or email [stusheridan@hiaitus.ai](mailto:stusheridan@hiaitus.ai). Rulings referenced: ABAC Adjudication Panel Determinations 26/26 and 98/26, both published 2026.