AI Card Grading, Explained: How It Works and When to Trust It

How AI card grading works

You photograph the front and back of a card. The model then evaluates the same four categories human graders at PSA, BGS, CGC, SGC and TAG score:

• Centering: border ratios left to right and top to bottom on both sides. For reference, CGC's published scale allows centering of roughly 55/45 on the front and 75/25 on the back for a Gem Mint 10. Binder AI also includes a manual centering tool if you want to place the border lines yourself, and you can check any card's ratios with the free card centering calculator.
• Corners: sharpness, whitening and dings across all eight corners.
• Edges: chipping and whitening along every border.
• Surface: scratches, print lines, holo wear, clouding, indentations and staining.

The output is an overall grade estimate with a subgrade for each category and written condition notes. You get the reasoning, not just a number: "slight whitening on bottom-left rear corner," "front centering approximately 60/40." For a breakdown of what each category means to a grader, see card grading criteria.

Computer vision pre-grading in plain language

An AI card scanner does not "look" at a card the way you do. It turns the photo into measurements and patterns, then compares those patterns with what it learned from many examples. In practice, a card grading app does roughly four things in order:

1. Find the card. The model locates the card's outline in the photo and straightens it, so a slightly tilted shot can still be measured.
2. Measure the geometry. With the card outline and the inner print border located, the borders on each side can be measured and turned into a centering ratio like 55/45.
3. Inspect regions. Corners, edges and the open surface areas are examined for the visual signatures of wear: bright white specks where the top layer of a corner has lifted, broken lines along an edge, or streaks that break up the gloss.
4. Translate to a scale. The findings are mapped onto the grading scale you picked. Binder AI estimates on the PSA, BGS or TAG scale, and each scale treats the same flaw a little differently, which is why a card can read as a 9 on one and a 9.5 on another.

Binder AI also identifies the card from the same scan, so the grade estimate sits next to eBay sold prices, price history and a graded value table for that card. That pairing is what turns a condition estimate into a decision.

What a photo can and cannot show

A camera is a powerful tool, but it is not a loupe and a tilting lamp in a grading room. Knowing the limits keeps your expectations honest.

What photos capture well

• Centering: this is pure geometry, and a straight, flat photo measures it well.
• Corner and edge whitening: white flecks against a dark border, especially on dark card backs, stand out clearly.
• Obvious creases, dings and stains: anything you can see at arm's length, the camera can usually see too.

What photos can miss

• Fine surface scratches: many hairline scratches only appear when light rakes across the card at a low angle. A single flat photo can hide them completely.
• Glare: a bright reflection washes out the exact area you need to inspect, and can also look like a defect when there is none.
• Holo and foil cards: holographic patterns create busy, shifting reflections that make scratches, clouding and print lines harder to separate from the design.
• Dents and indentations: a shallow dent without a color change may be invisible from straight above.
• Alterations: trimmed edges, recolored corners and pressed-out creases are designed to fool the eye. A photo-based estimate is not an authentication check, and professional graders use tools and handling that a phone photo cannot replace.
• Sleeves and toploaders: plastic adds its own scratches and reflections, which can be read as card damage.

How to take photos that give AI a fair look

1. Take the card out of its sleevePlastic adds glare and scratches that are not on the card. Handle the card by its edges with clean, dry hands.
2. Use soft, even lightDaylight near a window or a diffused lamp works well. Avoid direct overhead bulbs that throw a hot spot onto the surface.
3. Shoot straight on, on a contrasting backgroundHold the phone parallel to the card so the borders are not distorted. A dark mat behind a light-bordered card (or the reverse) makes the edges easy to find.
4. Fill the frame and check focusGet close enough that the card fills most of the photo, then confirm the corners are sharp before you submit the scan.
5. Check holos at an angle yourselfAfter the scan, tilt holo and foil cards under a light and look for scratches the flat photo may have missed. The full checklist is in how to pre-grade cards.

Is AI card grading accurate?

The honest framing matters: AI grading is a screening tool, not a certification. No AI grade adds slab value to your card. What it does is help you avoid paying a grading fee to learn a card is an 8, and flag the 9 and 10 candidates that are worth a closer look.

Binder AI provides an 87% accurate prediction, based on Binder AI's internal comparison of app estimates with returned professional grades. You can read more in PSA vs AI grading. In practice, accuracy depends heavily on photo quality and on the card itself. Clean, glare-free, straight-on photos of an unsleeved card give the model the clearest view. Blurry, shadowed shots of a holo through a sleeve do not.

How to use an AI grade estimate

• Use it to rule cards out. The most reliable use of an estimate is spotting cards that will not reach the grade you need. If centering measures 65/35 or a corner shows clear whitening, a 10 is unlikely no matter which company you pick.
• Treat high estimates as "inspect further." An estimated 9.5 or 10 means "worth a hands-on look," not "guaranteed gem." Inspect it under angled light before you pay to submit.
• Run the money math. Compare the graded value at the likely grade with the raw price plus fees and shipping. The grading ROI calculator does this for you, and is my card worth grading? explains the logic.
• Re-scan after you clean up your photo setup. If an estimate surprises you, retake the photos with better light before you trust or dismiss it.

How TAG uses imaging in official grading

Imaging is not just a pre-grading idea. TAG, one of the five major grading companies, builds its official process around it. TAG says its cards are graded "using TAG's machine-learning software in conjunction with our patented image capture technology," and that the data is then "analyzed and confirmed by the Verification team." TAG says it prefers the term "machine-learning" over "AI," and that all defects its system identifies "are thoroughly reviewed by human grading experts before Grade Finalization."

TAG's service pages list high-resolution stereoscopic card imaging and a Digital Imaging and Grading (DIG) report with notable defects, corner, edge and surface analysis, and centering metrics. Its scale page explains that the 1000-point TAG Score maps onto a 1 to 10 grade, with Gem Mint 10 covering scores of 950 to 989 and Pristine 10 covering 990 to 1000.

The difference from phone-based AI grading is the hardware and the handling. TAG captures the card in a controlled imaging setup, with specialists verifying the results and the card physically in hand. A phone app works from whatever photo you take. Both use computer vision, but only one ends in an authenticated slab. For more on the company, see PSA vs BGS vs CGC vs SGC vs TAG.

Photo pre-grading vs. human pre-grading

Some collectors pay an experienced person (a card shop, a friend with a loupe, or a paid pre-grading service) to look over cards before submission. Here is how the two approaches compare.

The two work well together: let the AI card grader triage the stack, then give your own careful hands-on inspection, or a trusted human's, to the short list that survives.

AI grading vs. professional grading

How Binder AI works, step by step

1. Download the appBinder AI is free to download on iOS and Android, and also works on the web. AI grading is included with a subscription.
2. Pick your grading scaleChoose the PSA, BGS or TAG scale for the estimate, depending on where you are most likely to submit.
3. Scan the front and backFollow the photo tips above. Binder AI identifies the card from the scan.
4. Read the grade and subgradesYou get an overall estimate plus centering, corners, edges and surface subgrades with written notes. Use the manual centering tool if you want to set the borders yourself.
5. Check the valueReview eBay sold prices, price history and the graded value table to see what the card is worth raw and at each grade.
6. Decide and organizeSubmit the strong candidates, keep the rest raw, and file everything into binders organized by game or sport.

For a plain look at the app's strengths and limits from the team that makes it, read our Binder AI review, or go straight to the AI grading app.

What collectors actually use it for

• Pre-submission screening: grade everything with AI, then submit only the strongest candidates.
• Buying raw cards: scan before you pay "near mint" prices at shows, then re-check on delivery for online purchases.
• Collection triage: work through a binder or a storage box and let the estimates tell you which cards deserve sleeves, slabs or the sell pile.
• Learning to grade: reading the written notes on many cards trains your own eye. Pair it with the card centering guide to learn what graders look for.

Sources, checked October 2026: TAG: steps of the grading process, TAG: does TAG use AI to grade cards, TAG grading scale, TAG Base grading service, CGC grading scale.