eBay Seller Rating Levels: Audit Defects and Late Shipments With AI

On eBay's published criteria, your seller level rests on cases closed without seller resolution, your transaction defect rate and, for Top Rated, a low late shipment rate. Export your performance data weekly, strip personal details, and ask AI to group defects by cause, category and carrier. Fix the biggest cluster first, then check the rule thresholds on eBay's page.

The levels are Top Rated, Above Standard and Below Standard. Falling to the bottom one costs more than pride: eBay lists lower placement in search, possible reductions in selling limits, exclusion from its advertising tools and possible payment holds among the consequences. The useful work is finding the small number of causes behind most defects and turning them into a list you can clear in 30 minutes a week.

Follow me on Instagram@sagnikteaches

What sets the level, as eBay states it

All of this comes from eBay's English-language seller levels page; other eBay sites may use different thresholds, and eBay's help centre can change without notice, so check the page for your own site before relying on a number.

Connect on LinkedInSagnik Bhattacharya

• Metrics measured: cases closed without seller resolution, transaction defect rate, and late shipment rate.
• Minimum standards: cases closed without seller resolution of no more than 2, or 0.3% of transactions, with the higher of the two figures applying; and a transaction defect rate of no more than 2% of transactions. The page gives no late shipment threshold for staying out of Below Standard.
• Levels: Top Rated exceeds eBay's expectations, with an established sales history and compliance with other policies. Above Standard meets them. Below Standard falls below the minimum standards.
• Evaluation: on the 20th of each month. Sellers with fewer than 400 sales in the previous three months are judged over twelve months; those with 400 or more over three.
• Top Rated requirements: an account active for at least 90 days, at least 100 transactions and $1,000 in sales with buyers in the site's home market over twelve months, and compliance with the selling practices policy. Upgrades take effect on the 1st of the following month.
• Top Rated thresholds: eBay's Top Rated Seller Program page sets a transaction defect rate of 0.5% or less, with three or fewer defects from unique buyers; cases closed without seller resolution of 0.3% or less, and two or fewer; a late shipment rate of 3% or less, with five or fewer late shipments; and tracking uploaded within your stated handling time, with carrier validation on 95% of domestic transactions.
• Top Rated Plus: needs same-day or one-business-day handling and 30-day or longer free returns, and carries a 10% discount on final value fees on the page's description.

Those Top Rated figures come from the Top Rated page for eBay's main site. Other eBay sites publish their own tables with similar but not identical numbers, so if you are chasing Top Rated, copy the figures for your own site into your sheet from its page.

Subscribe on YouTube@sagnikteaches

What goes into a defect

eBay's seller centre page on defect removal names two events that add a transaction defect: you cancel a transaction because the item is out of stock, or a case closes without seller resolution, meaning eBay stepped in on a problem you had not resolved and found you responsible. An item-not-as-described return request is not a defect in itself. eBay tracks those requests in its separate service metrics, which compare your rate with similar sellers and can bring their own limits, such as higher final value fees, at a "Very High" rating; a request becomes a defect only when it escalates into a case that closes against you. Late shipment is tracked as its own rate. Write that definition at the top of your sheet so the audit uses eBay's rule, not a guess.

The same page lists defects eBay removes automatically, among them a defect caused by an eBay site issue or programme error, one tied to a buyer eBay acted against under its abusive buyer policy, a request ruled in your favour, and systemic shipping or communication delays that valid tracking confirms. Check every defect that looks outside your control against that list before you spend time fixing its cause.

Two things follow for how you read your own data:

1. One bad listing can cause many problems. A wrong item specific on a best seller produces repeat not-as-described returns for the same reason, and each one you leave unresolved is a case waiting to happen.
2. Rates move slowly for a small seller and quickly for a busy one. With the twelve-month window, one bad week stays visible for a year; with the three-month window, it falls away sooner but weighs more heavily.

The main worked example: 480 orders, one bad month

All figures are illustrative. Say a seller of refurbished office equipment, run by a small accountancy practice as a sideline, had 480 sales in the last three months. That puts it above 400, so the three-month window applies on the page. The performance report shows 11 transactions with defects.

defect_rate = 11 / 480 = 2.29%      (the page's limit is 2%)
defects_allowed_at_480 = 480 x 0.02 = 9.6, so 9 at most
cases_allowed = higher of 2 or 0.3% x 480 (1.44) = 2
cases_seen = 3                      (over the allowed figure of 2)

Both lines are over. The seller needs to lose at least two defects (11 to 9) and one case, or add sales, for the next evaluation. Adding sales helps the ratio but not the cause; the right fix is to find what the 11 have in common. Here, eight are cancellations and three are cases closed without seller resolution, so the audit has two places to look: why stock ran out, and why three return requests ended up in front of eBay.

Now the numbers that matter for planning. If the seller sells about 160 items a month, each defect costs about 0.2 percentage points of the rate over three months. Two defects a month is 1.25% before anything else goes wrong; the headroom is smaller than it feels. Work out your own headroom: the number of defects you can afford before the next evaluation is the allowed rate times your sales, minus those you already have.

Get the data out of Seller Hub

Seller Hub's Performance area shows your level, your rates and the transactions behind them, and lets you download reports. The menu names change, so look for performance, seller level and service metrics (or similar) in your own Seller Hub, and use the download or export option on the list of transactions with defects. Then add what the export lacks:

• the item category or your own product group, matched by item number;
• the carrier and service used, from your shipping records;
• the not-as-described return requests from your returns list, since they are where cases start;
• your handling promise on the listing;
• the date you printed the label against the date the item was scanned by the carrier.

Before you paste anything into an AI tool, delete buyer names, usernames, addresses and message text from the sheet. The audit needs item numbers, dates, reasons and carriers. Keep a clean copy in your own files so that the original rows are safe.

Prompt: cluster the causes

Below is a table of my eBay defects (cancellations and cases closed without seller resolution), late shipments and not-as-described return requests from the last [period]. Columns: order date, item number, item title, category, type (cancellation / case / late shipment / return request), reason given, dispatch promise (days), actual dispatch date, carrier, case opened (Y/N), resolved by me (Y/N).
Use only this table.
1. Group the rows by type and reason, and show counts and percentage of total.
2. Within each group, look for a shared factor: the same item, the same category, the same carrier, the same day of the week, or the same handling promise. Report only patterns that appear in at least 3 rows.
3. Rank the clusters by number of rows, largest first.
4. For each cluster, give one fix that I could make this week, based only on what the data shows. If the data does not suggest a fix, say "no fix visible".
Do not guess causes that the table does not show.

Illustrative reply for the 11-defect month: five not-as-described return requests share one item (a refurbished label printer), and three of them are the cases that closed without seller resolution; six of the eight cancellations are items that sold out on another marketplace first, and the other two are single items found damaged at packing; three late dispatches, counted in the late shipment rate rather than as defects, all fall on Mondays. That is a ranked list. It is also a list built from about 20 rows, so treat any pattern with three rows as a lead to check, not proof.

What to check in what comes back

• Counts should add up to your row total. If the model reports more rows than your table holds, it has invented some.
• A "pattern" of exactly three rows, found only by splitting the data several ways, may be chance.
• Fixes the data does not support, such as "switch carrier" when only one parcel was lost, should be dropped.

Not-as-described returns: where most cases start

Start here, because an unresolved return request is how a case closed without seller resolution begins, and that is the tightest limit on the page. Open the printer listing and compare it with the items you shipped:

1. Does the title name the right model, year, size or colour?
2. Do the item specifics match the item? An AI-drafted listing may carry a wrong value, and checking item specifics against sold data catches that.
3. Do the photos show the flaws? A condition, flaws and measurements template gives every listing the same prompts.
4. Did you ship what you listed? Pick, pack and label errors count the same as listing errors to the buyer.

In the example, the printer listing said "tested, fully working" but the buyers' complaints mention missing paper trays. The fix is a three-line change to the listing (state that the tray is not included, add a photo) and a line in the packing checklist. One edit cuts the largest cluster.

Late shipments: separate the promise from the packing

eBay's help pages count a transaction as late when you sent it after the handling (dispatch) time in your listing, or when it was not delivered by the estimated delivery date; a delivery after that date is not counted late if the carrier's acceptance scan falls within your handling time. So the first carrier scan is what protects you, and two different failures hide behind the rate, with different fixes.

Ask AI to compute the gap between the handling promise and the first carrier scan, by weekday and by carrier, from your own rows. A weekday pattern is easy for it to find and for you to verify. Be careful with time zones: if your export uses one zone and the carrier's scan another, a parcel can look a day late or early. Check three rows by hand.

The cheapest improvement is often a promise that matches what you can do. If you promise next-day dispatch but really ship in two days, the fix is to say two days. The Top Rated Plus benefits on the page need same-day or one-day handling, which is a reason to hold that promise only if you can keep it every week.

Cancellations and stock

Cancelling an order because the item is out of stock adds a defect on eBay's own definition, and it often happens to multi-channel sellers whose stock runs out on another site first. Look at your cancellations by item and ask whether they come from one product line. In the example, six of the eight cancellations came from items also listed elsewhere, so the fix is a stock buffer for eBay or a sync, not an apology message. The multichannel listing tutorial covers keeping stock and rules in step.

Cases closed without resolution

A case that escalates to eBay and closes without you resolving it counts twice: once as a defect and once against its own limit, which is tiny: no more than 2, or 0.3% of transactions, whichever is higher. For a seller with 480 sales that is two cases. This is a hard line, so the audit has one extra task: for every open case, look at it today. Reply, offer the return or refund that fits, and close it yourself before eBay decides. A quick, polite reply costs less than a case.

AI can help you draft replies from your notes, but give it only the facts of the order, and the same rule as in appeals applies: it adds nothing you did not give it. The removed-listing tutorial has a prompt that enforces that.

A second business: slow sellers and the twelve-month window

Say a small architecture practice sells surplus drawing equipment and model-making supplies on eBay, about 60 sales in the last three months. Under the page's rule that puts it in the twelve-month window, which holds perhaps 240 sales. Two defects in one rough month, say one out-of-stock cancellation and one return that escalated and closed against the seller, are 0.8% of 240, fine on their own. But a third and fourth a few months later take it to 1.7%, close to the 2% line, while each month on its own looked quiet.

For a slow seller, the audit is about the whole year. Paste the full twelve months, group by cause, and ask a different question: "which two fixes would have prevented most of these?" With only a handful of rows the answer is often a stock buffer, a packaging change or a clearer size line. Because one defect weighs roughly 0.4 points here, against 0.2 for the busier seller, you cannot afford to wait for a pattern of three. Act on the first sign, and record it in the log so the next occurrence is recognised.

Before and after: one listing fix, measured

Back to the 480-sale seller. Before the fix, the label printer sold 40 units in three months and drew 5 not-as-described return requests, three of which became cases closed without seller resolution. After the listing change (tray not included, extra photo, packing checklist), the next 40 units drew 1 return request, which the seller resolved within a day, and no cases. Add a stock buffer that stops six of the eight cancellations, and defects fall from 11 to 2 (two cancellations, no cases), about 0.4% of 480 transactions, well under the 2% line, while the case count falls from 3 to 0. This is illustrative arithmetic, but the structure is real: a high-volume item with a high return rate, plus one stock habit, can account for most of the total.

The same arithmetic tells you which fix to do first. Rank clusters by how many defects they could remove, not by how annoying they are. A cluster of five tied to one listing outranks a cluster of five spread across five.

The weekly fix list: a 30-minute routine

1. Export (5 minutes). Download the defect and late shipment rows since your last audit.
2. Clean (5 minutes). Remove personal fields, add category and carrier.
3. Cluster (5 minutes). Run the prompt, check the counts add up.
4. Verify (5 minutes). Open three listings from the top cluster and confirm the cause by eye.
5. Fix (8 minutes). Edit the listings or the process the cluster points to.
6. Log (2 minutes). One line: date, cluster, fix, who owns it.

The log is the point. After a month you can see which fixes stuck, which clusters returned, and how many defects you have left in the headroom before the next evaluation on the 20th. The routine does not need to be perfect; it needs to be weekly.

What your sheet should show at a glance

A single page with five numbers. Transactions in the evaluation window. Defects so far, and the allowed number at your sales volume. Cases closed without seller resolution against the allowed figure. A late shipment rate you track yourself against the Top Rated cap of 3%, since the seller levels page gives no late shipment line for staying out of Below Standard. And the date of the next evaluation. Illustrative layout: "Window: 3 months, 480 sales. Defects 11, allowed 9. Cases 3, allowed 2. Late shipments 7 of 480 (1.5%). Next evaluation: the 20th."

Add one formula: defects allowed minus defects so far. If the number is negative, you are over and the audit is urgent. If it is below three, treat every order as risky and slow down on the items in your top cluster until the next evaluation. This arithmetic is simple enough to do in a spreadsheet without AI; use AI for the clustering and for the summary, not for the thresholds.

What Below Standard can cost you

eBay's page lists these consequences, applied immediately after evaluation or from the 1st of the following month, or after more than two consecutive months below standard: lower placement in Best Match search, a possible cut in selling limits, being blocked from eBay's advertising tools, possible payment holds until tracking shows the item is on its way, and losing the ability to deduct from a refund for used or damaged returns. It does not map each to a time, so read the page's current wording. Some eBay sites also charge extra fees to sellers who stay Below Standard, so check your own site's policy page.

The advertising block matters if you rely on promoted listings: you cannot test ad rates against a margin floor if the tools are closed to you. Another reason to spend 30 minutes a week on the sheet.

Mistakes that make the audit useless

• Clustering without a window. A mixed export from several months hides a recent change. Always audit the same window eBay uses.
• Trusting AI percentages. Recalculate the main rate yourself and compare it with Seller Hub's figure.
• Fixing the symptom. Replying faster to cases is good, but if five come from one listing, edit the listing.
• Pasting personal data. If a row has a buyer's name or address, remove it before you paste.
• Chasing the level instead of the cause. The level is a result. A small number of causes make most of it.

For help turning raw exports into tidy tables before the AI step, cleaning messy data goes through the common formatting traps, and the same habits keep a weekly sheet reliable. If a pattern in your exports does not fit any of the clusters above, a short call to look at your own data usually shows what is happening faster than guessing from a template.

Questions about seller levels and defect audits

How often does eBay update my seller level?

On eBay's seller levels page, evaluation happens on the 20th of each month, and a Top Rated upgrade takes effect on the 1st of the following month. Other eBay sites can use different thresholds, so read the page for the site you sell on.

Which sellers are judged over twelve months and which over three?

On that page, sellers with fewer than 400 sales in the previous three months are evaluated over twelve months, and sellers with 400 or more are evaluated over three. A small seller therefore carries an old bad month for much longer than a busy one.

Should I give an AI tool my Seller Hub export?

Give it order and defect rows with the buyer names, addresses and messages removed. You need item, category, dates, reason codes and carrier, not personal details. Keep the original export in your own files.

Can AI predict my next level?

It can calculate your current rates from your own rows and show how many more defects tip you over a line. It cannot see eBay's exact internal counts, so compare its figures with the numbers in Seller Hub's performance page before acting.

Further reads

• eBay Bulk Listing for Sports Cards: An AI Error Checklist — A category where wrong specifics create returns.
• eBay Bulk Listing Tool: Magical Bulk Listing vs Third-Party Apps — Listing tools change where errors enter.
• eBay HTML Description Generator: Mobile-Friendly Templates With AI — A consistent description layout cuts misreadings.
• DeepSeek or ChatGPT for eBay Sellers: Which Writes Better Listings? — Which chat model handles spreadsheet analysis best.
• How to Write eBay Listings With AI That Sell Faster — Clearer listings mean fewer disputes.
• Amazon A-to-z Guarantee Claims: Reply With AI-Drafted Evidence — A seller-side routine for A-to-z claims: the evidence pack by claim type, an AI-drafted reply, and the order defect rate effect.
• Can You Use AI Photos on eBay? What the Picture Policy Says — eBay's picture policy has no AI-specific rule, but six clauses still decide what an AI edit may do. Each one, and one edit put through them.
• eBay Store Automation With AI: What Is Allowed and What Isn't — Which eBay store jobs AI can safely draft, where the bot and agent rules bite, and how API-based tools differ from scrapers.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: eBay seller levels and performance standards page (export.ebay.com); eBay seller centre pages on the Top Rated Seller Program and on defect removal; eBay help pages on seller performance and service metrics for its other sites (late shipment definition, service metrics).