AddMyBot.

AddMyBot setup guide

Prepare PDFs and FAQs for your website chatbot

Turn readable PDFs and focused FAQs into useful AddMyBot knowledge. Review extracted text, understand the 4 MB limit and index before testing or publishing.

Updated · Examples are illustrative; no customer information is used.

Choose information your business can share

Useful chatbot knowledge answers a specific customer question: opening hours, delivery coverage, a return condition or how a service works. Start with a small set of current, authoritative facts instead of a large unrelated manual.

Use content you are authorized to upload and disclose. Keep customer-specific messages, passwords, order details and confidential documents out of knowledge. Relevant text and questions are sent to configured Google AI services for embeddings and generation; private storage is not a promise that AI processing stays only in your account.

Prepare a readable file within the supported limits

  • Supported uploads are PDF, UTF-8 TXT and Markdown (.md or .markdown). Each new file must be no larger than 4,000,000 bytes, displayed as 4 MB.
  • For a PDF, check that you can select and copy the relevant text. OCR is not included: an image-only scan may provide no usable text. Complex tables, columns and diagrams need careful extraction review.
  • Use meaningful headings and short paragraphs. Remove repeated headers, footers and irrelevant material from a file you are authorized to edit.
  • A small file is not necessarily a short text source. Indexing supports at most eight chunks of up to 1,200 characters with overlap; lengthy extracted text can exceed that limit even below 4 MB.
  • If the source is too long, choose a focused excerpt or prepare separate focused sources within your existing source allowance. Do not split files to bypass a plan or provider limit.

Write specific FAQs instead of vague claims

Use Add manual text in Knowledge when a few reviewed question-and-answer pairs are clearer than a PDF. Include the condition, timeframe and exception in the same answer. The example below is synthetic and is not AddMyBot policy or a real customer’s information.

Example source title: Sample shop collection FAQ

Question: When can I collect an order?
Answer: Collection is available Monday to Friday, 10:00–17:00,
excluding public holidays, after the shop confirms the order is ready.

Question: Can I collect without confirmation?
Answer: Wait for the shop’s ready-for-collection confirmation.
This example does not promise that a chatbot can look up an order.
  • Avoid “fast delivery everywhere” without actual locations and timing.
  • Resolve conflicting facts before indexing. Two versions of a policy can lead to an unreliable answer.
  • Name the source so a reviewer can identify it. Selected source names can appear in public citations; avoid confidential names.

Review extracted text, then index the current revision

  1. Add the source in Knowledge.

    Choose Add manual text, upload a supported file or use the public-webpage import. Manual text allows up to 50,000 characters, but that input limit does not increase indexing capacity.

  2. Read the extracted text.

    Check that amounts, dates, exceptions and paragraph order survived extraction. A Processed status means extraction finished; it does not mean the source is indexed or approved.

  3. Use the source’s indexing action.

    Wait for Indexed for internal retrieval. If indexing is unavailable, read the readiness or source failure message rather than repeatedly retrying.

  4. Test before publishing.

    Use a known question in Playground and verify its cited source. Then make a separate Website approval decision for the current revision.

Using a webpage and maintaining knowledge

The webpage workflow fetches one permitted public page, lets you review extracted text and save a selected excerpt. It is not a whole-site crawl, paid scraper or scheduled refresh. Do not import internal, authenticated or login-protected pages.

When a business fact changes, update the source through the supported Knowledge workflow. Editing invalidates the previous index revision; re-index and explicitly approve the current version for public use. Recheck representative questions after the update. Do not assume the bot automatically notices changes to the original website or PDF.

Troubleshoot preparation and indexing

  • Empty or garbled text: use a readable source or rewrite the authorized facts as manual text. Do not assume an image-only PDF has been understood.
  • File rejected: check the actual byte size, extension and supported format. Compressing a PDF may reduce file bytes without shortening the text to be indexed.
  • Indexing limit reached: reduce the relevant text or use appropriately scoped sources within your plan. Saving more characters does not raise the eight-chunk limit.
  • Allowance or provider unavailable: wait for the displayed reset or operator verification. Repeated retries, deletion and new sources do not reset usage.
  • Answer missing after a source edit: verify current indexing and public approval separately. Playground can use private indexed knowledge that the website widget cannot use yet.