Add text knowledge
Paste FAQs, policies, and opening hours straight into the chatbot. No file needed, and nothing can go wrong in extraction.
Browse topics
What it is
Text knowledge is content you type or paste directly into Agentency, rather than uploading a document. Opening hours, a refund policy, a list of frequently asked questions, the three things your support team explains twenty times a week — all of it can go straight in.
It becomes a source exactly like a file does: a row on the Knowledge table, with a title, a size, and a training status. The chatbot cannot tell the difference between an answer that came from a pasted paragraph and one that came from a PDF.
Why this is often the best option
Every other source type has an extraction step that can go wrong. A PDF may have a broken text layer, a scan may be blurred, a web page may be mostly navigation. Pasted text has none of that risk — what you type is exactly what the chatbot gets.
It is also the most compact option. A page of clean text costs a couple of kilobytes of your knowledge cap. The same information inside a designed PDF, surrounded by repeated headers and legal footers, costs several times more. On Free, with 400 KB to spend, that difference decides how much you can fit.
So: if you can write the answer in a paragraph, write it here rather than hunting for a document.
When you would use it
- Short, structured facts that have no document: hours, locations, delivery times, contact routes.
- A first test source while you are setting the chatbot up and want a fast, guaranteed-clean result.
- Rewriting something that extracted badly from a file or a scan.
- Corrections. When the chatbot keeps getting one specific thing wrong, a short, unambiguous text source stating the correct fact is the most direct fix.
Where to find it
Open Dashboard → Chatbots → your chatbot → Knowledge → Add → Text.
Before you start
- Decide the title before you start typing. It appears in the chatbot's citations, so a customer may well see it.
- Have the facts settled. Editing a source later means retraining it, which costs you the wait again.
- Glance at the usage meter if your cap is close to full.
Steps
- Open Dashboard → Chatbots → your chatbot → Knowledge → Add.
- Select the Text tab.
- Give the source a clear title — "Delivery and shipping times", not "notes".
- Optionally pick a collection to group it with related sources.
- Paste or type the content. Keep one topic per source where you reasonably can.
- Watch the estimate in the drawer; it shows how much of your cap this will use.
- Leave Start training on unless you are queueing several sources to train together.
- Submit, then wait on the Knowledge table until the badge reads trained.
- Ask Test Chatbot a question in the words a customer would use, not the words you just typed.
What you will see
The Text tab is a title field and a large editor. Under the editor is a live estimate of the size this content will add, measured against your remaining capacity.
Once submitted, the drawer closes and the new row appears on the Knowledge table with a pending badge, which moves to training and then trained. Short text sources usually finish quickly.
How to write text that answers well
The chatbot searches your material for passages that match a question, then answers from what it finds. That mechanism rewards a particular style of writing.
Write the question the way a customer asks it. If people ask "how long does delivery take", a heading that says "Fulfilment SLA" will not match nearly as well as one that says "How long does delivery take?". Use your customers' vocabulary, including the informal version.
Answer immediately after the question. Do not build up to it. The passage that gets found is short, so the answer needs to sit right next to the question.
Be explicit rather than contextual. A line reading "It is 30 days" is useless on its own, because the passage may be retrieved without whatever came before it. Write "Returns are accepted within 30 days of delivery."
Keep one subject per source. Ten separate short sources beat one long document containing the same ten topics, because each source stays focused.
State the negatives too. "We do not ship to PO boxes" prevents a wrong answer just as much as a positive fact provides a right one.
Include the synonyms people actually use. If your product is a "subscription" but half your customers call it a "membership", say both somewhere in the text.
A good FAQ-shaped source looks like this:
Do you ship internationally?
Yes. We ship to the EU, the UK, and the GCC. Delivery is 5-9 working days.
We do not currently ship to the United States.
How much does shipping cost?
Standard shipping is free on orders over 300 SAR. Below that it is 25 SAR.
Express shipping is 60 SAR everywhere we deliver.
Limits and plan notes
- Pasted text counts toward the chatbot's knowledge cap like everything else: Free 400 KB, Starter 10 MB, Standard 20 MB, Pro 40 MB, Agency 60 MB. See Knowledge limits.
- A single text source has a generous ceiling of its own, far beyond anything you would sensibly type by hand. In practice the plan cap stops you first.
- Editing an existing text source and saving it retrains that source. The chatbot keeps answering from the previous version until the new training finishes, so nothing goes dark while you edit.
- Adding text requires a role that permits managing knowledge.
Common problems
Save failed with a size error.
The paste is larger than your remaining capacity. Shorten it, remove a source you no longer need, or upgrade. The estimate in the drawer is the warning you get before this happens.
Test Chatbot ignores the new FAQ.
Check the badge first — a pending row is not answerable yet. If it reads trained, the problem is almost always vocabulary: you asked using words that are not in the text. Ask again using a phrase from the source. If that works, edit the source to include the phrasing customers actually use.
The chatbot answers, but from an older source instead.
You have two sources covering the same topic and the older one matched better. Remove or update the stale one — see Remove knowledge. Contradictory sources are the most common cause of confidently wrong answers.
My formatting disappeared.
Only the words are stored. Colours, fonts, and layout are not part of knowledge, and would not help an answer anyway. Structure your content with plain headings and short paragraphs instead.
I pasted a table and the answers are muddled.
Wide tables read poorly as passages. Restate the important rows as sentences: "The Pro plan includes 40 MB of knowledge and 5 seats" beats a table cell the chatbot has to reassemble.
Common questions
Is pasted text as good as uploading the same content as a PDF?
It is usually better. There is no extraction step to fail, and it uses far less of your knowledge cap because designed documents carry repeated headers and footers.
How should I word a text source so the chatbot finds it?
Write the question the way a customer asks it, then answer immediately underneath. Keep each answer self-contained so it still makes sense on its own.
Can I edit a text source after it is trained?
Yes. Open the row, edit, and save. The chatbot keeps answering from the previous version until the updated one finishes training, so nothing goes dark.
Why does the chatbot ignore my new FAQ?
Either the row is still training, or you asked using words that never appear in the text. Test with a phrase from the source, then add the customer's phrasing to it.
Was this article helpful?
Related articles
Knowledge overview
Your chatbot only answers from the knowledge you give it. Here is what counts as knowledge and how training works.
Upload files
Add PDFs, Word files, spreadsheets, and images from your device. Ten megabytes per file, twenty files per batch.
Knowledge limits
Your cap counts extracted text, not file size. Free 400 KB through Agency 60 MB, per chatbot.
Dataset status and training
What each badge on the Knowledge table means, what the failure messages are telling you, and when to just wait.
Ready to try it on your own content?
Create a free workspace, add a document, and ask the questions your team is tired of answering.