PDF, DOCX or web page? Choosing the right source for an AI assistant
The same information may live on a web page, in a PDF and in Word, but these are not equivalent sources for visitors or an AI assistant. Use this practical format guide.
A municipal website stores a timetable in a PDF, application guidance in Word and a current notice on a web page. A company adds a price list, product catalogue and internal glossary. These sources differ in freshness, structure and their ability to give an AI assistant a verifiable public link.
Format affects whether a visitor can find the information, open it on a phone, read it with assistive technology and verify an AI answer against the original source.
A public answer should lead to a public source. A supporting document may explain meaning, but it should not create a result that a visitor cannot open.
When a web page is the best source
HTML is the first choice for information that changes, receives frequent searches or helps people complete a task. Opening hours, contacts, service prices, returns procedures, application requirements, deadlines and common questions belong here.
GOV.UK guidance on accessible documents recommends publishing in HTML wherever possible. It notes that PDFs can make content harder to find, use and maintain, and often work poorly with assistive technology. HTML also adapts to mobile screens, text zoom and browser preferences.
A web page gives an assistant a clear public URL. Visitors can read the full context, while the organisation maintains one current version instead of synchronising duplicate content.
When PDF still has a clear purpose
PDF suits content that needs fixed layout, page numbering, an approved version or reliable printing. Examples include annual reports, signed decisions, print forms, technical sheets and archived records.
A PDF extension does not guarantee usable text. A scan may contain images only. W3C PDF techniques cover OCR for scanned documents, tagged headings, correct reading order, image alternatives, bookmarks, document language and marked-up tables. Missing structure affects both AI extraction and people using screen readers.
Google Search Central confirms that Google can index PDFs and other text document formats. Indexability alone does not make every file a good landing page. With changing information, visitors may find an older copy when several similar versions remain public.
DOCX and TXT as supporting material
Word and plain text are useful for internal guidance, terminology lists and detailed explanations that should not appear as standalone public results. They can help an assistant distinguish an old product name, departmental abbreviation or technical term.
They should not be the sole source for a public price, deadline, authority or availability claim. Visitors need a public page where they can verify information that affects their next action.
Four questions before adding a source
1. Will a visitor act on this information?
If people need to call, apply, pay, arrive at a given time or buy a product, give the information a current public page. A document can remain as an attachment.
2. How often does it change?
Prices, stock, dates and contacts need one owner and one primary source. A controlled XML feed may suit a large catalogue because availability does not have to be copied into editorial pages.
3. Does the fixed document need to remain?
An approved, signed or archival version is a clear reason to retain PDF. Add a short HTML page with a summary, effective date, content owner and link to the file.
4. Can software read the document correctly?
Check whether text can be selected, whether scans have OCR, whether headings form a logical outline and whether tables make sense without their visual layout. In DOCX, use real heading styles instead of enlarged bold text.
How Informio Search separates public results from supporting context
Informio Search creates result cards from explicitly selected public web and XML sources, so each result leads to a real public URL. Selected manual text, PDF, DOCX and TXT sources can provide supporting knowledge that clarifies terminology and context. A supporting document does not become a public result card or create a new public fact.
A manufacturer may keep product pages on the website and model-family terminology in an internal document. The glossary helps interpret an older name, while the result remains the current product page. A municipality can use internal guidance to explain an administrative term, while the resident receives a link to the valid public procedure or form.
A practical source clean-up
- List the pages, PDF, DOCX, TXT and data feeds used by people or the assistant.
- Give each source an owner, validity date and purpose: public answer, supporting context or archive.
- Move frequently searched and changing information into HTML or a controlled data source.
- Keep PDF where its fixed or approved form matters. Add OCR, headings, language, alternatives and a meaningful filename.
- Remove obsolete public copies or mark them clearly as archives.
- After indexing, test real questions and verify that each answer leads to the right public source.
Municipal example
An approved bylaw may remain a PDF. The practical procedure, contact, deadline and form belong on an HTML page. Internal guidance can explain terms, but it should not replace the public bylaw or procedure.
Company and ecommerce example
A technical data sheet may remain a downloadable PDF. Product name, price, availability, key specifications and returns information belong on a product or service page. An internal model glossary can improve query interpretation while results lead to the current catalogue.
The shortest decision rule
Put information on the web when people need to find, cite and regularly update it. Keep an accessible PDF when an approved or print-ready form matters. Use documents for controlled supporting context, and base public claims on public pages.