Pre-Shipment Inspection Checklist: AI Drafts Your Inspector Brief

Give a third-party inspector a written brief containing your product spec, the AQL sampling level, defect classes with examples, the tests to run, the photos you need, packing checks and a pass or fail rule. AI can draft that brief from your spec sheet in minutes; you then set the numbers and have the inspector confirm them.

An inspector cannot read your mind. If the brief says "check quality", they apply their company's generic checklist, and you get a report that passes goods you would have rejected. The value of an AI draft is that it forces every vague phrase into something measurable, and it asks you about the gaps before the inspector does.

Follow me on Instagram@sagnikteaches

What the inspector is actually doing on the day

One large inspection company's description of the service is a useful baseline. A pre-shipment inspection is booked once at least 80 percent of the order is complete, with packing under way. The inspector selects samples at random using a published sampling standard (ANSI/ASQ Z1.4; ISO 2859-1 is its international counterpart), runs the checks on your checklist, which can be a standard one, a modified one or a fully custom one, and usually spends one day on site. The report arrives the same day, with findings, photos of defects, a pass or fail result and verified carton details, quantities and packing information.

Connect on LinkedInSagnik Bhattacharya

Three of those points decide how you write the brief. The inspector works from a sample, so your tolerance must be expressed as sampling numbers. The inspector works from a checklist, so anything not on it goes unchecked. And the report gives a verdict, so the brief must define what a pass is before the day, not after.

Subscribe on YouTube@sagnikteaches

Step one: collect what the AI needs to see

An AI draft is only as good as its source material. Gather, in one folder:

• The product specification: dimensions with tolerances, materials, colours with reference codes, weight, finish, labelling and any compliance marks.
• Your purchase order, with quantity per variant and per carton.
• The approved pre-production sample or a photo of it, and photos of past defects from earlier orders or customer returns.
• The packaging spec: inner pack, carton marks, barcodes, carton dimensions and maximum weight.
• Any compliance or labelling points your market requires (name them as requirements, and check them with your adviser).

The photos of past defects are worth more than the spec. A return reason such as "strap came loose within a month" turns into a specific test in the brief: a pull test on every strap seam in the sample.

Step two: set sampling and defect classes

This is the part importers most often leave to the inspector's default. Set it yourself.

Sample size from lot size

Under the standard's General Inspection Level II, which many inspectors use by default, a lot of 3,201 to 10,000 units uses code letter L and a sample of 200 units; a lot of 501 to 1,200 uses code letter J and a sample of 80. For a 200-unit sample under normal inspection, the standard's table accepts up to 10 defective units and rejects at 11 at AQL 2.5; at AQL 1.0 it accepts up to 5 and rejects at 6; at AQL 4.0 it accepts up to 14 and rejects at 15. An inspector may still run a different level or a tightened plan for a supplier with a poor record, so make the inspector confirm the plan in writing, by letter code, sample size and accept and reject numbers.

Critical, major and minor

The common convention is zero tolerance for critical defects, AQL 2.5 for major and AQL 4.0 for minor, with AQL 1.0 used for major defects on higher-risk products. The numbers matter less than the definitions, because the same flaw can be minor on a mug and critical on a baby product. Have AI propose the classification from your spec, then edit it. Here is a filled-in example for an illustrative order.

Illustration: an online clothing store orders 3,500 cotton hoodies in four colours and three sizes. The lot falls in the 3,201 to 10,000 band, so the sample is 200 units.

Notice the numbers in the examples: 5 mm, 10 mm, ten cycles. An inspector can measure those. "Print slightly off" cannot be measured, and it becomes an argument with the supplier.

Step three: the prompt

Paste your spec and defect photos' descriptions into a ChatGPT, Claude or Gemini conversation, or into a project that holds your standing quality rules. This prompt produces a draft brief and a list of questions.

You are helping an importer write a brief for a third-party pre-shipment inspector.
From the spec, purchase order and past-defect notes below, produce:

1. Product summary and reference sample details (one paragraph).
2. Sampling: lot size [N], inspection level [II unless stated], and AQL for critical (0),
   major [2.5], minor [4.0]. Leave sample size and accept/reject numbers as
   [INSPECTOR TO CONFIRM FROM THE PLAN].
3. A defect list in three classes. Every item must be measurable or testable:
   include a number, tool or method. If the spec does not give a tolerance, write
   [QUESTION FOR BUYER] instead of inventing one.
4. On-site tests: what, how many units, pass criterion.
5. Packing and carton checks: counts, marks, barcodes, weights, drop or stack test only if I specify.
6. Photo requirements, as a numbered list.
7. The verdict rule: pass, fail, or hold for buyer decision, and what the inspector must
   do if they find a defect type not on the list.
8. A final list of every question you could not answer from my material.

Spec: [paste]
Purchase order: [paste]
Past defects: [paste]

The line "if the spec does not give a tolerance, write a question" is what stops the model inventing a plausible number. In a first run on the hoodie spec, the output asked three questions: what colour tolerance (delta E or visual against a swatch), what the maximum size tolerance was, and whether the drawcord rule applied to adult sizes. Those are the questions the inspector would have asked on the morning of the visit, or worse, not asked.

Reading what comes back

Illustrative extract from the AI's draft, with the edits the buyer made:

Step four: photo requirements that make the report usable

Photos are your only way to see the goods, so ask for specific ones. Without a list, you get twelve pictures of cartons and one blurred close-up. A reasonable list, to adapt:

1. The lot as presented: the full stack of cartons, with the factory's carton marks readable.
2. Each randomly selected carton, closed, with its number, then opened, with the contents in place.
3. One unit of each variant laid flat next to the approved sample, same frame, same light.
4. Every defect found: a wide shot showing where it is, a close shot with a ruler or coin for scale, and the unit's identifying mark.
5. Labels and tags in focus, front and back, including barcodes scanned and the scan result.
6. Each measurement taken, with the tape visible in the frame.
7. The weighing scale showing a full carton's gross weight, and the carton's measured dimensions.
8. The inspector's tools and the sample units, with the box sealed at the end, if you want to keep a reference set.

Ask for originals, unedited, in a shared folder, as well as embedded in the report. If a dispute starts later, the originals with timestamps are what you hold up.

Step five: write the pass or fail rule before anyone goes on site

A written rule saves a bad conversation later. Here is a version to adapt:

• Pass: no critical defects; major defects at or under the accept number; minor defects at or under the accept number; carton counts and marks match the purchase order; all required tests passed.
• Fail: any critical defect, or major or minor defects at or over the reject number, or a failed required test.
• Hold for buyer decision: the inspector finds a defect type not in the brief, a colour or measurement that sits on the tolerance line, or fewer units finished than the brief states. The inspector must call or message before closing the report.

Back it with your supplier agreement. A short line in the purchase order, "goods are subject to third-party inspection against the attached brief, and shipment is released only on written approval from the buyer", gives the brief weight. Have a solicitor or adviser phrase it for the contract you actually use.

A filled-in checklist for the hoodie order

The final brief is shorter than people expect, because each line is a check. An extract from the illustrative hoodie order, in the form an inspector can tick:

• Quantity check: 3,500 units across four colours and three sizes, matched to the purchase order split. Count finished and packed units; report the percentage packed.
• Random selection: 200 units drawn across all colours and sizes in proportion, from at least 8 cartons chosen by the inspector, not the factory.
• Appearance: no holes, stains, open seams, broken stitching, or misprints over 5 mm, checked on all 200.
• Measurements: 5 units per size, three points of measure, plus or minus 1.5 cm.
• Function: drawcord length on child sizes against the safe length on the spec; ten-cycle fastener test on 10 units.
• Labels: content, care and size labels present, legible and correct on all 200; barcode scan on 10 cartons.
• Metal check: running a metal detector over the sample, if the factory has one on site, with the result in the report.
• Packing: 12 units per carton, polybag closed, carton marks as per sheet, gross weight within 0.5 kg of the sheet.

Two briefs that failed on the day (illustrative)

The ambiguous word. A specialist food importer asked an inspector to check that "labels are correct" on 4,000 jars of olive paste. The inspector checked the label text against the artwork, which was right, and passed. The brief never said to check the best-before date format or that it should be legible at the point of sale. Half the jars had smudged dates. The fix was to turn "correct" into a list: ingredients in order, weight statement, date present, date legible, lot code present, barcode scans. When AI drafts the list for you, ask it to split every vague adjective into its parts.

The factory chooses the cartons. A furniture maker sourcing metal bed frames let the supplier pick which cartons to open. The goods inspected were the best-packed ones. The brief now says the inspector chooses by random method, writes down the carton numbers in advance on the report, and opens cartons from the top, middle and bottom of the stack.

Checking the finished brief before it goes out

1. Read it as the inspector would. Is there any line you cannot measure with a tape, a scale, a gauge or by comparison to the sample?
2. Count the questions the AI left open. Each one needs your answer or an explicit "inspector's standard".
3. Send it to the inspector two days ahead and ask them to reply with their sampling plan: lot size, code letter, sample size, accept and reject numbers. If the numbers differ from your expectations, ask why.
4. Send a copy to the factory. Factories that see the brief beforehand tend to fix what is on it, which is the point.
5. After the report, compare it with the brief line by line. Anything the report calls "not checked" is a gap to put right next time.

Deciding what to do with a failed lot

A fail is a business decision before it is a quality one, and it helps to have the options priced before the report lands. Illustrative sum for the hoodie order: the 3,500 units cost $11 each, so the lot is $38,500. The inspector finds 14 major defects in the 200 sample against an accept number of 10, so the lot fails. Extrapolating, 7 percent of units may carry a major defect, which is about 245 hoodies. Your options, with illustrative figures:

The cost of returns is the number to hold in your head. If a defective hoodie costs you a refund, return postage and a lost customer, one defective unit can cost far more than its $11 purchase price, which is why sorting before shipment is usually cheaper than sorting after.

Choosing the inspection level, and when to tighten it

General Inspection Level II is the usual default. Level I uses smaller samples and suits a supplier you have used many times with clean reports. Level III uses larger samples and suits a first order, a product with a safety angle, or a supplier with a history of problems. The cost of a larger sample is more inspector time on site, so check how the inspection company prices extra man-days.

A sensible pattern for a new supplier: Level III on the first order, Level II on the second, and Level I only after three clean reports in a row. Ask the AI to keep this record, in a table with one row per order: supplier, lot size, level, AQL, result, defects found, action taken. When the pattern changes, for example a major defect that had never appeared, that row tells you to step the level up again. The same table is also the evidence if you need to claim against the supplier.

An example from a furniture maker's component supply

Illustration: a furniture maker imports 1,800 powder-coated steel table legs a quarter. The first order was inspected at Level III with AQL 1.0 for major defects because coating failure shows up months later. The AI-drafted brief added two checks the maker had not thought of: a cross-hatch adhesion test on 5 legs per colour, and a salt-water soak on 3 legs left overnight where the inspector's schedule allowed, with the photos at the start and end. The first report failed on adhesion. The factory changed its curing time, the second order passed at Level II, and the maker never saw the field complaints that the first batch would have produced.

Using each report to improve the next brief

After each inspection, paste the report's defect list into the same conversation and ask the AI to update the defect classes and the question list. Over three or four orders you build a defect library for each supplier, which is more useful than any generic template. If the same defect keeps appearing, move it from minor to major in the brief, or add an in-line check during production as well as the final one. Tracking supplier prices, lead times and risk shows how to log this per supplier, and AI quality control for small manufacturers is the version for your own production line. If you are still choosing between factories, comparing supplier quotes side by side comes before any of this.

Confirm the AQL figures, the sampling plan and what counts as a pass with the inspection company before you rely on this tutorial's numbers. They come from the standard's normal-inspection tables, and the inspector's own plan is what governs the day.

Questions importers ask before booking an inspector

When should I book the pre-shipment inspection?

Book it for when production is nearly finished and packing has started, so the inspector can pull cartons of finished goods. One large inspection company states 80 percent complete as the minimum. Book a few days ahead and agree the date with the factory so the goods are ready, not half-packed.

Is AQL 2.5 right for everything?

No. It is a common default for major defects, with 4.0 often used for minor ones and zero tolerance for critical ones, but the right level depends on what a defect costs you. Safety-related items need stricter handling than decorative ones. Agree the numbers with the inspector, who will know what is usual for your product type.

Can the inspector change my checklist on the day?

Only if you allow it. State in the brief that the inspector must report anything outside the checklist but classify it only with your approval, and must contact you before passing or failing a lot on a point the brief does not cover. This avoids a verdict you cannot defend to the supplier.

What if the inspection fails?

Decide before the visit: sort and rework, then re-inspect; accept with a price reduction; or reject the lot. Put that choice in the brief and in your supplier agreement. Without it, you negotiate from a failed report while the shipping slot runs out.

Further reads

• How Much Does AI Visual Inspection Cost a Small Manufacturer? — What camera-based inspection costs on your own line.
• AI Invoice Processing: Stop Typing Supplier Bills by Hand — Match the supplier invoice to what was inspected.
• Supply Chain AI Software: What a Small Importer Needs First — Where inspections sit in an importer's wider toolkit.
• How to Automate Purchase Orders and Supplier Emails With AI — Keep purchase orders and supplier emails consistent.
• Best AI for Ecommerce Operations: Orders, Stock, Returns, Suppliers — See how defects show up later as returns.
• AI Dropshipping Product Finder: Check Demand and Supplier Risk First — Check supplier risk before placing a first order.
• AI for Customs Brokers: Draft Client Updates and Entry Checks — Draft shipment status emails from your file notes and run a consistency pass on invoices and packing lists, with the licensed broker still signing the entry.
• Alibaba Sourcing With AI: Vet Suppliers and Plan a Sample Order — A supplier scorecard, a red-flag check for messages and a sample-order plan, with the document checks left to a person.
• Kickstarter Campaign Page: Draft the Story and Reward Tiers With AI — Draft the story and video script with AI, then price each reward tier so fees and shipping don't eat the margin.
• Landed Cost Calculator in Excel: Test Tariff Changes With AI — One worked importer example: build the sheet, then test a duty rise and see the price you need to hold your margin.
• Can ChatGPT Find the Right HS Code? How to Check Before You File — How to use ChatGPT for a first-draft HS code, a ten-product test you can run on your own goods, three worked products and the checks before filing.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: QIMA's pre-shipment inspection page (booking at 80 percent complete, ANSI/ASQ Z1.4 sampling, checklist types, one day on site, same-day report); QIMA's AQL page (code letter L, sample of 200, accept 10 and reject 11 at AQL 2.5; AQL 1.0 for higher-risk categories, 0 for critical defects); a published apparel AQL chart (Level II bands 501 to 1,200 = J, 80 units, and 3,201 to 10,000 = L, 200 units; accept and reject points at AQL 1.0 and 4.0); an inspection firm's guide to Levels I, II and III. Confirm accept and reject numbers against the inspector's own plan.