Black Box vs Glass Box
Many AI analysis tools are black boxes. You upload an image and get a confident sounding verdict,
but you cannot see the rules underneath it. Ad Corrector is a Glass Box. We use measurable signals and fixed logic that stays consistent.
| Feature | Black Box Typical AI Opinion Tool | Glass Box Ad Corrector |
| Core engine | Probabilistic model output | Deterministic measurements and rules |
| Output | Generated commentary that can vary | Scores, signals, and fixed guidance |
| Consistency | Can shift between runs | Same input produces the same results |
| Risk | Confident sounding uncertainty | Limits are explicit and measurable |
| Goal | Pleasing feedback | Truthful constraints for visibility |
Step 1 Ad Corrector
Step 1 measures visibility risk using deterministic signals. We estimate what is likely to be harder to see, read, or process fast.
1) Speed View
A motion-visibility simulation built for fast scanning. It applies illustrative blur based on the speed and distance you select to show what detail is likely to survive when the viewer has seconds.
2) Heatmap
A visual-detail guide that compares local light-dark variation across the artwork. It helps reveal when your headline is not visually dominant. It is not eye tracking and does not predict where people will look.
3) Contrast and clarity signals
Contrast is measured as a luminance contrast ratio (the same relative luminance formula used by WCAG) on the headline and call to action. The lowest available sample is used, and supporting text is reported separately. Clarity combines edge strength in the artwork with the selected speed, distance, and word count to identify readability risk.
4) Six scored checks
Step 1 scores six checks from 0 to 100: Readability (copy length), Contrast, Clarity at speed, Colors, Composition, and Call to Action. Together they produce the overall score and letter grade.
Important: These are engineering checks designed for speed and distance reality.
They are repeatable and they do not change based on mood, prompting, or interpretation.
You can evaluate your creative against these visibility signals by running a free billboard test with Ad Corrector.
Step 2 Persuasion Engine
Step 2 interprets Step 1 findings as directional message-processing signals for the first seconds of viewing.
It does not independently measure audience response, and it does not predict recall, emotion, action, or sales.
How it works
- Uses Step 1 scores for contrast, clarity, composition, color, and copy length. Call to Action mode also uses CTA strength.
- Applies fixed weights to compute five metrics, each scored 0 to 10: Visual Salience, Message Focus, Ease of Processing, Distinctiveness Signal, and a message metric whose label depends on your campaign goal: Path to Action (Call to Action mode) or Message Clarity (Brand Awareness mode).
- Reports a composite message processing signal (0 to 100) and practical fix-first guidance.
How the five metrics combine
Each metric is calculated from Step 1 signals using fixed weights. The composite message processing signal (the Persuasion score) combines the five metrics as follows, scaled to 0 to 100: Visual Salience 22%, Path to Action or Message Clarity 24%, Message Focus 18%, Ease of Processing 18%, and Distinctiveness Signal 18%. The weights do not change between runs, so identical inputs always produce identical results.
Deterministic again: same Step 1 inputs produce the same Step 2 results. This is a scoring and interpretation layer, not a separate measurement.
How the overall score and grade are calculated
Every test produces six check scores from 0 to 100 and one overall score from 0 to 100. The overall score is converted into a letter grade. The same artwork and settings always produce the same score and grade.
The six checks and their weights
| Check | Weight | What it measures |
| Readability | 18% | Copy length, based on the entered word count. It does not measure comprehension. |
| Contrast | 16% | Luminance contrast ratio of the headline and call to action. |
| Clarity at speed | 20% | Edge strength combined with the selected speed, distance, and word count. |
| Colors | 14% | Color balance across the artwork. |
| Composition | 16% | Visual density, separation, hierarchy, and balance. |
| Call to Action | 16% | Strength of the entered call to action wording. Not scored in Brand Awareness mode. |
The overall score is the weighted average of the scored checks, rounded to the nearest whole number. In Brand Awareness mode, the Call to Action check is left out and the other five checks are averaged by their weights.
Grade scale
| Grade | Score | Plus and minus |
| A | 90 to 100 | A+ is 97 to 100. A is 93 to 96. A- is 90 to 92. |
| B | 80 to 89 | B+ is 87 to 89. B is 83 to 86. B- is 80 to 82. |
| C | 70 to 79 | C+ is 77 to 79. C is 73 to 76. C- is 70 to 72. |
| D | 60 to 69 | D+ is 67 to 69. D is 63 to 66. D- is 60 to 62. |
| F | Below 60 | No plus or minus. |
Provisional grades
If contrast cannot be measured for all of the entered headline or call to action text, the grade is marked Provisional and uses the checks that were measured. Locating the remaining text in the tool completes the grade.
Grade vs Persuasion score
The letter grade comes from the six Step 1 checks. The Persuasion score is the separate Step 2 composite. Two ads can share a Persuasion score and still earn different grades.
What you get from the Glass Box
The system is built to reduce avoidable failure before you spend money on media.
Outputs are designed to be actionable without requiring interpretation.
Measured constraints
Visibility signals, contrast risk, and fast comprehension pressure points that may explain why a message struggles at speed.
Prioritized fixes
Clear, ordered guidance focused on the few changes that move the needle most for OOH readability and first impression.
Repeatability
No randomness, prompt sensitivity, or inconsistent output between runs.
Human translation
Step 2 turns Step 1 findings into plain-language message-processing signals so you can see where the first seconds may be at risk.
What this is and is not
- Is: a deterministic pre-flight evaluation to reduce visibility risk and support clearer creative decisions.
- Is not: a prediction of recall, emotion, action, sales lift, brand lift, or campaign outcomes.
- Is not: a substitute for live market testing when you have the budget and time to run it.
Methodology questions
How does Ad Corrector assign a letter grade?
Six weighted checks (readability, contrast, clarity at speed, colors, composition, and call to action) combine into an overall score from 0 to 100. Scores of 90 and above earn an A, 80 to 89 a B, 70 to 79 a C, 60 to 69 a D, and below 60 an F. Scores ending in 7 to 9 add a plus, and scores ending in 0 to 2 add a minus. Brand Awareness mode leaves out the call to action check.
Is the grade the same as the Persuasion score?
No. The grade comes from the six Step 1 checks. The Persuasion score is a separate Step 2 composite of five message-processing metrics, so two ads can share a Persuasion score and still earn different grades.
Does Ad Corrector use AI to score creative?
No. Scoring uses deterministic measurements and fixed weights. The same artwork and settings always produce the same scores and grade.
What does a Provisional grade mean?
A Provisional grade means contrast could not be measured for all of the entered headline or call to action text. The grade uses the checks that were measured until the remaining text is located in the tool.