You will be able to predict when an uploaded file will be read well and when it will be misread or skipped.
Hui Min is about to rent a room in a condo in Toa Payoh. The landlord's agent sent the tenancy agreement as a phone photo of each printed page, eleven images in all, and asked her to sign by Friday. She uploaded them to an assistant and asked what the notice period was. It told her two months. When she read the page herself, the clause said one month, but the "1" had been printed faintly and sat right next to a coffee stain.
Nothing about her question was wrong. The problem happened before the model read a word, in the step where her photos were turned into something it could read at all. This module is about getting accurate answers from files, and it starts with that step.
Most assistants now accept PDFs, word processing documents, spreadsheets and images, and many accept several at once. The model itself, as AI fundamentals lesson 3.1, Your words become tokens before the model sees them, explained, works on tokens. So every file has to be turned into text or tokens before the model can use it.
How that happens depends on the file. A digital PDF or a Word document already contains its text, so the text can be pulled out directly, and this usually works well. A spreadsheet is converted into rows and values, and sometimes the assistant runs code to read it, which is good for numbers but can lose formatting and notes. An image, including a photo of a printed page or a scanned PDF, has no text inside it at all. The text has to be recognised from the picture, either by a separate text-recognition step or by a model that reads images directly, as AI fundamentals lesson 7.1, How AI makes and reads images, described.
You usually cannot see which of these happened. The assistant just answers, and the answer reads the same whether the conversion was perfect or full of errors.
Some material is much more likely to come out wrong.
Scanned and photographed pages are the biggest risk. Faint print, shadows, skewed angles, stains and low resolution all cause misreadings, and a misread digit in a figure or a date looks just as confident as a correct one. Hui Min's "1" became a "2" this way.
Complex tables come next. Merged cells, tables that run across two pages and columns without clear borders can lose their structure in conversion, so a value ends up attached to the wrong row. Charts are harder still, because the numbers are not written anywhere and have to be estimated from the shapes. Handwriting, including signatures, initials and notes in the margin, is often misread or ignored.
A useful rule: the further a document is from clean digital text, the more you should check.
The other thing that can go wrong is quieter. A long document may not fit in the context window, the limit described in AI fundamentals lesson 5.1, The context window is the model's only working memory. Products deal with that in different ways. Some cut the file off at a certain length. Some summarise parts of it first. Some split it into pieces and search for the passages that look relevant to your question, the retrieval approach from AI fundamentals lesson 5.3, Retrieval: how an assistant looks things up.
Each approach means some parts of the file may never actually be read when the model answers you. A question about a clause on page 40 of a 60-page contract can get an answer based on pages 1 to 10 and a guess. Even when the whole file fits, AI fundamentals lesson 5.2 explained that detail in the middle of long material is the most likely to be missed.
How each product handles files is different and changes often: which file types it accepts, how large a file can be, how many files per chat, and whether long files are read in full or searched. These details are on each product's help pages, and they are worth reading for the assistant you use most, because they decide which of your documents will be handled well.
Two habits help whatever the product. First, use the cleanest version of a document you can get. Ask for the digital PDF instead of a photo, or export a spreadsheet properly rather than taking a screenshot. Hui Min asked the agent for the original file, and he emailed it within the hour. Second, for long documents, upload only the part you need when you can, or ask about one section at a time.
You will see the difference between clean and messy files most clearly by testing both on the same questions. Find one clean digital PDF and one document you only have as a scan or a photo, and choose three questions whose answers you can check yourself in each.
Upload one clean digital PDF and one scanned or photographed document, ask the same three questions of each, and compare the accuracy.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).