Supported File Formats

Synaptiq accepts four document formats for your knowledge base. Each format has different strengths, and how you structure your content within these files directly impacts the quality of answers your AI agent delivers.

Accepted Formats at a Glance

| Format | Extension | Best For | |---|---|---| | PDF | .pdf | Polished documents, product sheets, official collateral | | Microsoft Word | .docx | Internal docs, drafts, structured guides | | Plain Text | .txt | Raw content, quick additions, data exports | | Markdown | .md | Technical documentation, structured reference material |

PDF Files

PDF is the most commonly uploaded format. Synaptiq extracts text content from PDFs using an advanced parser that preserves heading hierarchy, tables, and list structures.

Best Practices for PDFs

• Use text-based PDFs, not scanned images. If you can select and copy text in the PDF, it will work well. If the PDF is essentially a photograph of a page, the AI cannot read it.
• Keep formatting clean. Simple layouts with clear headings, paragraphs, and bullet points extract more reliably than complex multi-column designs with floating text boxes.
• Tables work, but keep them straightforward. Simple data tables (pricing grids, feature comparisons) extract well. Heavily merged cells or nested tables may lose structure.
• Avoid PDFs that are primarily images. A product brochure that is 90% photographs with small text captions will yield very little usable content.

When PDFs Work Well

• Product specification sheets
• Official pricing documents
• Compliance and regulatory documents
• Published whitepapers and case studies

Microsoft Word (DOCX)

Word documents are excellent for internal documentation that may still be evolving. Synaptiq reads .docx files and preserves their heading structure, lists, and tables.

Best Practices for DOCX

• Use Word's built-in heading styles (Heading 1, Heading 2, etc.) rather than just making text bold and large. Styled headings let Synaptiq understand your document's hierarchy, which improves how it chunks and retrieves content.
• Use bullet and numbered lists for multi-point information like feature lists or step-by-step instructions. These are easier for the AI to parse than long run-on paragraphs.
• Remove tracked changes and comments before uploading. Accept or reject all changes so the document contains only the final content.
• Avoid embedded objects like Excel charts or Visio diagrams. These are not extracted.

When DOCX Works Well

• Internal product guides and playbooks
• Sales training materials
• FAQ compilations
• Process documentation

Plain Text (TXT)

Plain text files are the simplest format. What you see is exactly what the AI gets. There is no formatting to lose or misinterpret.

Best Practices for TXT

• Use blank lines to separate sections. Without visual formatting, blank lines are how the AI identifies where one topic ends and another begins.
• Write descriptive section headers on their own lines, even if you cannot make them bold. A line that reads RETURN POLICY followed by a blank line and then the policy content is parsed more effectively than a wall of unbroken text.
• Keep one topic per file when possible. A single text file covering everything from pricing to technical specs to company history makes it harder for the AI to retrieve the right section.

When TXT Works Well

• Quick knowledge additions (a new FAQ answer, a policy clarification)
• Exported content from other systems (CRM notes, support ticket summaries)
• Structured data like contact information or configuration values

Markdown (MD)

Markdown is the ideal format for technical documentation and structured reference material. Its heading syntax, lists, code blocks, and tables translate directly into the kind of structured content the AI processes most effectively.

Best Practices for Markdown

• Use heading levels consistently. # for the document title, ## for major sections, ### for subsections. This hierarchy directly maps to how Synaptiq chunks your content.
• Use fenced code blocks for any technical content like API endpoints, configuration snippets, or code examples. The AI recognizes these as distinct from prose.
• Use Markdown tables for comparison data, pricing tiers, or feature matrices. These parse cleanly.
• Use front matter if you like, but it is not required. Synaptiq reads YAML front matter but does not depend on it.

When Markdown Works Well

• API and integration documentation
• Technical reference guides
• Developer-facing product docs
• Structured FAQ pages

What to Include in Your Knowledge Base

The more relevant, specific information your AI has, the better it performs. Prioritize uploading:

• Product and feature descriptions with specifics, not marketing fluff. "Widget Pro supports up to 10,000 concurrent connections with 99.9% uptime SLA" is useful. "Widget Pro is our best-in-class solution" is not.
• Pricing and plan details including what is included at each tier, usage limits, overage costs, and billing frequency.
• FAQs that your sales and support teams already maintain. These are gold because they are already in question-answer format.
• Case studies and use cases that describe how specific customer types use your product, including measurable outcomes.
• Comparison guides that honestly position your product against alternatives, including where you have advantages and where competitors might have an edge.
• Onboarding and setup instructions so the AI can walk prospects through what getting started actually looks like.
• Terms, policies, and compliance information that customers frequently ask about during the sales process.

What to Avoid

Some content types hurt more than they help:

• Image-only PDFs. Scanned documents, brochure screenshots, or PDFs generated from image exports contain no extractable text. The AI sees nothing.
• Scanned documents without OCR processing. If you must use a scanned document, run it through an OCR tool first and upload the OCR output.
• Extremely large documents covering dozens of unrelated topics. Split these into focused, single-topic files for much better retrieval accuracy.
• Outdated content. A pricing sheet from two years ago sitting alongside the current one creates contradictory answers. Remove old versions when you upload new ones.
• Internal jargon without explanation. If a document uses acronyms or internal terms that customers would not know, add definitions. The AI will repeat jargon it does not understand, confusing prospects.
• Duplicated content. Two documents saying the same thing in different ways forces the AI to choose between them, sometimes incorrectly. Maintain one authoritative source per topic.

Formatting Tips for Better AI Comprehension

Regardless of format, these principles improve how well Synaptiq understands and uses your content:

1. Lead with the answer. State the key fact first, then provide context. "Widget Pro costs $49/month for up to 5 users" is better than three paragraphs of context followed by the price.
2. Use specific numbers. "Response time under 200ms" beats "very fast response times."
3. Write in complete sentences. Bullet fragments like "Fast - reliable - scalable" give the AI less to work with than "Widget Pro is designed for speed, with sub-200ms response times, 99.9% uptime reliability, and horizontal scaling to 100,000 concurrent users."
4. One idea per paragraph. Dense paragraphs covering multiple topics make it harder for the AI to extract the right piece of information for a specific question.
5. Be explicit about what you do NOT support. If your product does not integrate with a specific platform, say so clearly. This helps the AI give honest answers instead of guessing.