how to analyze images with chatgpt: a practical guide
Learn how to analyze images with chatgpt using practical prompts for receipts, screenshots and charts. Extract useful details and check answers before use.

how to analyze images with chatgpt: a practical guide
To learn how to analyze images with chatgpt, start with one image and one specific question: attach the image, describe the result you need, and check the answer against the original. Ask for unreadable details to be marked instead of guessed. Use your first attempt for a small job, such as copying a receipt or reviewing a product photograph.
How should you prepare and upload an image?
Choose an image you can inspect yourself. For a first attempt, use your own sample receipt or a product photograph without customer information. Decide what you need before attaching it.
- Open a conversation and look for the attachment control near the message box.
- If an image or file option is available, select your prepared image.
- Check that you selected the correct version.
- Add your question and requested answer format.
- Send the request, then keep the original open for comparison.
Follow the controls actually shown in your account. The sources here explain the API, the connection software uses to send requests to a model. They do not establish ChatGPT's current button labels, subscription access or upload limits.
Choose a straightforward file
I suggest starting with JPG or PNG. OpenAI lists PNG, JPEG, WEBP and non-animated GIF for API image input; that does not establish the supported formats in your ChatGPT account.
If the attachment fails, read the displayed error before changing anything. Try one replacement file and keep the same request. Avoid changing several things at once when you are trying to locate the problem.
For a document, include its heading and relevant column labels. For a product, decide whether you need the whole item or a specific detail. Keep an untouched copy so you can return to the original during review.
What should you ask ChatGPT to do with the image?
Ask for a useful deliverable: a transcription, a comparison table or a list of proposed changes. A prompt is simply the instruction you give the tool. Write it as a short brief you could hand to a colleague.
Vision means processing image content. OpenAI documents reading visible text and answering questions about objects, shapes and colors. Its vision guide describes these capabilities.
| Your material | A focused request | Your next check |
|---|---|---|
| Receipt | Copy the printed entries | Match every amount |
| Storefront photograph | Review sign readability | Inspect the proposed changes yourself |
| Advertising screenshot | List visible claims | Confirm each claim is supportable |
| Sales chart | Describe the displayed movement | Check dates and units |
| Product image | List visible features | Add verified specifications separately |
Use four parts in the request
Name the material, explain the purpose, specify the output and set a rule for missing information. Start with the following template, then replace the bracketed text.
This is [describe the image]. I need [specific result] for [business purpose]. Return [table, list or exact transcription]. Mark anything you cannot determine as unknown. Separate what is visible from your suggestions.
For example, ask for a poster's wording before asking for a stronger offer. Once you have checked the wording, use the advertising copy prompts to develop alternatives. Keep the verified offer details in your next request.
As an editorial rule, I recommend separating extraction from advice. First approve what the image says. Then decide what you want to change.
How do you turn a receipt into a checked table?
Copy the printed information before requesting calculations. Keep the printed total and any calculated total in separate fields. Inspect disagreements before transferring the result into your records.
A hypothetical expense example
Imagine a freelance photographer organizing a purchase receipt after a job. The immediate task is to prepare a draft expense entry. Start with the following request rather than asking for a complete financial assessment.
Read this receipt and produce a table with item description, quantity, unit price and printed line amount. Leave missing fields blank. Copy the printed total separately. Preserve the displayed currency and mark unclear characters with [unreadable]. Do not classify the purchase for tax purposes.
Check the merchant name, date and individual entries against the photograph. Review any discount or extra charge as a separate line. If an entry is uncertain, keep it out of the final record until you resolve it.
Ask for a targeted correction
The second item's amount does not match what I can read. Revisit only that row. Show the uncertain part without filling in a likely value. Leave the other rows unchanged.
After checking the transcription, ask for the format you need for your spreadsheet. State the column order and how you want missing values handled. You can develop the next step with these spreadsheet prompt templates.
For handwritten notes or a photographed form, use the same review discipline. Request an exact copy first and editing second. Preserve names and reference numbers until someone has checked them.
If this is client work, agree on the review responsibility before delivery. Use the freelancer prompt examples to write a clear handover note explaining what has been checked and what still needs attention.
How should you review a chart or product photograph?
For a chart, establish what is being measured before requesting an explanation. For a photograph, establish what is visible before requesting sales language. Make the first answer a record you can verify.
A chart request for a marketing meeting
Identify the chart title, date range, axis labels and units. List the visible changes without inventing exact values. Separate observations from possible explanations. For each explanation, say what additional information we would need to check it.
Hypothetical example: a small marketing team uploads a campaign chart. Ask the team to confirm whether the vertical axis shows revenue, orders or enquiries. Before discussing performance, have someone verify that the dates match the campaign period.
Where the underlying spreadsheet is available, use it for exact calculations. Keep the screenshot for discussing presentation and visible patterns. Put the supporting numbers beside any conclusion you plan to share.
For the surrounding review, use the social media audit guide. Choose one decision for the meeting, such as which campaign needs closer investigation, rather than requesting every possible recommendation.
A product photograph request
Review this product photograph for an online listing. List visible features, then suggest changes to framing and background. Do not infer material, dimensions, authenticity or durability. Put any information the seller must supply in a separate checklist.
Hypothetical example: an online shop is preparing a bag listing. Have the owner supply the actual dimensions and material from the product record. Use those confirmed details when drafting the description.
When preparing the final image for the website, continue with the image SEO guide. Keep the visual review and the upload checklist as separate tasks you can sign off.
How do you decide whether an answer is ready to use?
Approve the answer only after checking the details that matter to your next action. For a receipt, check amounts. For an advertisement, check claims and offer terms.
Small text, graph interpretation and inaccurate descriptions are documented limitations. OpenAI lists the relevant cautions.
- Compare names, dates and amounts with the original.
- Look for details added without evidence.
- Keep unclear fields visibly unresolved.
- Separate copied values from calculated values.
- Label suggestions as suggestions.
- Name the person who approves the final output.
Run a small test this week
Choose an ordinary receipt, a clear chart and a product photograph. Write a different question for each. Record what you requested, what needed correction and whether the final answer was useful.
For your next attempt, change the part that caused difficulty. If the table structure was wrong, specify the columns. If the task was too broad, narrow the question.
Do not make custom model training a requirement for this exercise. Vision fine-tuning means additional training using image examples, as described in OpenAI's separate technical guide. Begin by testing your material and instructions.
Before making the process routine, write a short handover rule: who prepares the image, who checks the answer and where the approved result goes. Keep the original alongside your checked output. Revisit the process when the input or intended use changes.
Common questions
How do you upload an image to ChatGPT?
Look for the attachment control, select your image if that option is available, and add your request. Follow the options shown in your account.
Can ChatGPT read text inside an image?
Yes. Request an exact transcription and check it against the original. Reading visible text is a documented capability.
What image formats does ChatGPT accept?
Try JPG or PNG first. The cited API format list does not establish which formats your ChatGPT account accepts; check the upload response.
What can ChatGPT do once it sees an image?
Start by requesting visible text or details. After checking those, ask for a comparison or proposed improvements tied to your business purpose.
Sources
- OpenAI: Images and vision. Consult for API image capabilities and limitations.
- OpenAI: Vision fine-tuning. Consult to distinguish additional training from an ordinary image request.
Follow me on Instagram
Short notes, practical examples and daily digital strategy ideas.
I'm Anar Rustamli - a strategist, entrepreneur, and AI adoption leader working at the edge of growth, technology, and human thinking. Since 2016, my work has focused on helping businesses evolve in a rapidly changing digital landscape. I design growth systems, AI-powered workflows, and strategic frameworks that align performance with purpose. I believe real growth happens when strategy, data, and human insight work together - and my mission is to help businesses adopt AI in a way that strengthens both their results and their identity.

