To have AI explain a document in plain English, give it the text (paste it, attach the PDF, or take a photo of the page) and ask three things: what the document says in simple words, what it wants you to do, and by when. Adding “at a sixth-grade reading level” or “as if I’ve never seen one of these before” makes the explanation much clearer.
The documents people most want explained are also the most personal: medical bills, insurance letters, tax notices, leases, benefit forms. Here’s how to get a clear explanation without sending those pages to a company’s servers, and where AI explanations go wrong.
Which documents can AI explain well? #
AI language models are good at rewording formal, jargon-heavy writing into everyday language. They’re less reliable at telling you what the law requires or what you should decide.
| Document | What AI does well | What to double-check |
|---|---|---|
| Lease or rental agreement | Plain summary, who pays for what, notice periods | Local tenant law, which can override the lease |
| Insurance letter or explanation of benefits | What was covered, what you owe, why a claim was denied | Appeal deadlines and amounts against the original |
| Medical bill | Codes and line items in plain words | Whether charges are correct; call the billing office |
| Tax or government notice | What the letter is about and what it asks for | Deadlines and amounts; the agency’s own website |
| Employment contract | Notice periods, non-compete and IP clauses in plain words | Enforceability; a lawyer for anything unusual |
| School or benefits form | What each section is asking for | Eligibility rules on the official site |
| Terms of service | What you’re agreeing to, especially data and cancellation | Any clause you plan to rely on |
| Medication leaflet | Side effects and instructions in simpler words | Anything about dosing with your pharmacist |
The pattern: use AI to understand the document, then check anything you’ll act on against the document itself or the organization that sent it.
How to get a document explained, step by step #
- Get the text into the chat. Paste it, attach a PDF, or photograph the page. A PDF with real text is the most reliable. A photo works for a one-page letter. For long scanned documents, see the note on scans below.
- Ask for a plain summary first. “Explain this letter in plain English in five sentences.”
- Ask for actions and deadlines separately. “List everything this asks me to do, with any dates or deadlines. Quote the exact wording for each deadline.”
- Ask about the words you don’t know. “What do ‘deductible’ and ‘coinsurance’ mean here, using the numbers in this letter?”
- Ask what to ask. “What questions should I ask the insurance company when I call?”
- Check the quotes. Find each quoted deadline and amount in the original. If you can’t find it, the AI may have made it up.
Prompts that work #
Plain summary:
Explain this document in plain English at a sixth-grade reading level. Start with one sentence saying what kind of document it is and why I received it.
Action list:
What does this letter want me to do? List each action, the deadline, and what happens if I don’t. Quote the sentence each item comes from.
Money:
Explain every amount in this bill. What am I being asked to pay, and why? Put it in a table.
Jargon:
Make a short glossary of every technical or legal term in this document, explained in everyday words.
Before a phone call:
I need to call about this letter. Give me five questions to ask and what information I should have ready.
In another language:
Explain this document in simple Spanish. Keep names, dates and amounts exactly as they appear.
The “quote the sentence” instruction matters more than any other. It forces the model to point at the source, which makes mistakes much easier to catch.
Why keep these documents off cloud chatbots? #
A single insurance letter can hold your full name, address, member ID, date of birth, diagnosis codes and the name of your doctor. A tax notice has your tax ID number. When you upload that to a cloud chatbot, it’s stored on the company’s servers under its privacy policy and, depending on your settings, may be reviewed by staff or used to train models.
You can black out identifiers before uploading, and for many people that’s enough. The simpler option is an AI that runs on your phone, so the document never leaves it.
Personal LLM does this on iPhone and Android. You attach a PDF, text or Markdown file to a chat; the text is extracted and searched on the phone, and the model answers from the matching passages and tells you which ones it used. For a paper letter, you can snap a photo instead: every model in the catalog can read images once you add its image-support file. There’s no account and no server, and once a model is downloaded the whole thing works in airplane mode. The app can also read any answer aloud with on-device text-to-speech, which helps if reading long explanations is tiring.
Our step-by-step guide to chatting with a PDF offline covers file types, long documents and what the on-device search misses.
Where AI explanations go wrong #
It invents or misreads numbers. Amounts, dates and account numbers are exactly where small models slip. Always check them against the page.
It misses the one clause that matters. On a phone, a long PDF is searched for the passages most related to your question, and only those go to the model. If you ask “what does this say about pets?” and the clause calls them “animals,” it can be missed. Ask the question a second way, or use a broader question like “list every obligation the tenant has.”
It explains the document, not your rights. A lease might say you owe a cleaning fee; your state’s law might say otherwise. AI can tell you what the paper says. It can’t reliably tell you what’s enforceable where you live.
It sounds certain. Explanations come out fluent and confident whether they’re right or not. Our guide to why AI makes things up explains why.
Scanned documents. A scanned PDF is a picture of text, with no text layer to search. Document chat can’t read it. Photograph the pages and ask about them with a vision model, one page at a time, or use your phone’s built-in text recognition to copy the text out first.
For health documents specifically, see whether it’s safe to ask AI health questions. The short version: use AI to understand the words and prepare questions, and leave decisions to your doctor or pharmacist.
Plain-language explanations as an accessibility tool #
Plain English isn’t only for tricky contracts. For people with dyslexia, a cognitive disability, low vision or limited English, dense official writing can be a real barrier, and asking an AI to restate it simply is a practical workaround. A few prompts that help:
- “Rewrite this in short sentences, one idea per sentence.”
- “Use bullet points and bold the actions I need to take.”
- “Explain this in Arabic, then give me the English terms I’ll need on the phone.”
- “Read this back to me as a numbered checklist.”
Combined with read-aloud, that turns a two-page letter into a short spoken checklist. Our overview of offline accessibility features covers what your phone itself can do, such as screen readers and live text.
Frequently asked questions #
Can AI explain a legal document? #
It can explain what a legal document says in plain words, define terms and point out clauses worth asking about. It can’t tell you whether a clause is enforceable where you live or what you should do. For anything with real stakes, confirm with a lawyer or the organization involved.
Is it safe to upload my medical bill to ChatGPT? #
It’s stored on OpenAI’s servers like any other chat, so black out your name, member ID and account numbers first if you use it. An on-device AI app avoids the upload entirely, since the document stays on your phone.
Can AI read a photo of a letter? #
Yes, if the app supports images. Take a clear, flat, well-lit photo of each page. For long documents, a PDF with a text layer works better, because photos are handled one image at a time.
How do I get AI to explain something more simply? #
Ask for a reading level (“sixth grade”), a format (“short bullet points”) and a limit (“in five sentences”). If it’s still too complex, reply “simpler” or “explain it like I’m new to this.” Asking it to use an everyday comparison also helps.