PARTLY

As of 13 August 2026, AI can only partly count your stock from photos.

This still needs a person who signs their name to it.

Can you do it?

5 minutesto a draft.

30 minutesto something you’d act on.

Cost, all in£0

Skill neededchat-fluent

Who has to check ityou

What the alternative costsNo priced stock-counting alternative is listed in the supplied tool data.

If this goes wrong: you record the wrong quantity and make a purchasing, fulfilment or availability decision from it.

What to actually do

  1. Hand it to a person

    The route this page recommends

    A person who owns the outcome does this end to end, worth it when the failure is dear.

  2. Use a tool built for this

    Second choice
  3. Do it yourself

    The distant third

    A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

    How to actually do it

    1. Open your stock list or inventory system and copy the product names, SKUs and pack sizes into a separate note.
    2. Photograph each shelf, bin or pallet from a clear angle with labels facing the camera, using separate photos for different product groups and avoiding duplicate views.
    3. Paste the product list into a chatbot, attach the photos, and send the counting prompt.
    4. Copy the returned table into your stock sheet and mark every low-confidence count, duplicate-photo warning and unidentified product.
    5. Physically recount the marked items and a sample of the other items, then compare those counts with the table before updating your inventory system or placing an order.

    Prompt

    Count the stock items visible in the attached photos. Treat each photo as evidence of the same physical stock area, so do not add duplicate views of the same item. Use the product list below to distinguish similar items. Return a table with product name, counted quantity, photo references, confidence as high, medium or low, and a short reason for every low-confidence count. Do not count items that are hidden or only partly visible unless the quantity is clear. Do not guess missing quantities. Flag overlapping items, reflections, blurred labels, duplicate photos and any product you cannot identify. At the end, list the items that need a physical recount. Product list: [paste product names, SKUs and pack sizes here].

    Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.

What it gets wrong

What caps this at PARTLY: verification cost, context depth and stakes of error.

How we scored this

Five axes, each scored nought to two by hand: ten means AI carries the task cleanly, and the thresholds that turn a total into YES, PARTLY or NO are published in the methodology. Each axis name links to its definition.

AxisScore (0–2)
Output2
Inputs2
Verification1
Liability1
Effort delta1
Total7 / 10

FAQ

Can AI count stock from a photo?
Yes, it can count visible items in a clear photo and put the result into a table. It cannot reliably count hidden, overlapping or ambiguous items, so recount flagged results before using them.
How accurate is AI stock counting from photos?
There is no fixed accuracy for every stockroom or photo set. Clear, separated items with readable labels are easier to count than mixed, stacked or partly hidden stock, and the result still needs a physical check.
Can I use ChatGPT to count my inventory?
You can use a vision-enabled chat tool to make a first count from attached photos. Give it your product list and ask it to flag uncertainty, then check the count against the stock in front of you.
What photos do I need for AI stock counting?
Take clear, well-lit photos showing the full shelf, bin or pallet, with labels and product faces visible. Use different views for different stock groups and avoid attaching duplicate photos of the same items.

Nearby answers

Assessed by gpt-5.6-luna (gpt-5.6-luna) on 2026-08-13, second-checked by an independent model. Wrong somewhere? Email [email protected] and it gets re-checked.

The newsletter

AI news, new answers and product picks, straight to your inbox.