AI Context & Style

How to think about it

Techne currently supports context conditioning, not model fine-tuning. You guide output with prompts, workspace context, and style preferences that get injected into generation requests.

Configure these in Settings → AI and in Studio when creating a post.

Train Your AI (Workspace Context)

In Settings → AI, the Train Your AI card shows whether the active workspace is currently trained, what it's trained on, and how many characters of context are saved. Once trained, every blog, metadata, social, and image generation in that workspace uses this context automatically.

• Status badge: Trained (with character count) or Not trained yet.
• Saving to: the active workspace name is always shown above the Save button — important when managing multiple brands or client accounts.
• Three sources: a website URL (via copy-paste prompt), up to 5 PDFs, or free-text notes.
• Only workspace owners and admins can train, retrain, or clear context.
• You can view, edit, or clear the saved context any time.

Agencies and multi-brand accounts: training is per workspace. Switch to the right workspace before training so context for one client never leaks into another's content.

Train from a website URL (no PDFs needed)

If you do not have PDFs handy, Techne ships a copy-paste prompt that turns any public website into a Company Context using ChatGPT, Gemini, or Claude. You will find it in Settings → AI → Company Context, just above the PDF uploader.

How it works:

1. Enter your website URL (for example, https://techne.blog/).
2. Click Copy prompt and open ChatGPT, Gemini, or Claude using the quick-launch buttons.
3. Paste the prompt and let the LLM read your website.
4. Copy the LLM's response and paste it back into the Paste the LLM's response box.
5. Click Use this response, then Synthesize Company Context to save it to your workspace.

The prompt that gets sent to the LLM asks for a structured JSON profile so we can auto-fill writing style and image style at the same time:

You are a brand & business analyst. I will give you a single website URL. Visit the homepage and obvious sub-pages (About, Services/Products, Pricing, Blog, FAQ, Contact) and produce a structured brand profile another AI can use to write on-brand content.

Website: https://techne.blog/

Return ONLY a single JSON object — no prose, no markdown, no code fences, no commentary. Use this exact shape:

{
  "business_context": "500–1500 words of flowing prose (no headings, no bullets)…",
  "writing_style": "4–10 short rules in second person, separated by newlines.",
  "featured_image_style": "1–3 sentences describing on-brand imagery.",
  "suggested_keywords": ["5–10 short SEO phrases"],
  "suggested_emoji_use": "none | sparing | frequent",
  "suggested_emdash_use": "avoid | ok | love"
}

When you paste this back into Techne, the review screen lets you keep or skip each suggested field. Plain prose still works as a fallback — it's saved as Company Context only.

Tips:

• Works best when the site has a clear About page and product/service pages.
• Use a model with web browsing enabled (ChatGPT with browsing, Gemini, Claude with web search) so the LLM can actually visit the URL.
• You can edit the saved context any time in Settings → AI.
• Privacy: Techne does not fetch the URL itself. Only the third-party LLM you choose sees the URL you paste.

Fine-tune Style Preferences

Below the training card, the Fine-tune section gives you direct control over style fields. Training can prefill some of these (with your approval), but they always remain editable:

• Writing Style Guide — tone and voice rules. Auto-fillable from training.
• Featured Image Style — default visual direction. Auto-fillable from training.
• Allow em dashes — toggle. Never changed automatically by training.
• Allow emojis — toggle. Never changed automatically by training.
• Target Keywords — your SEO strategy. Training can suggest keywords for you to add, but never overwrites your list.
• Default Backlink URLs — fully user-controlled.

Per-Post Overrides in Studio

In Studio, you can override workspace defaults for a single post:

• Toggle Use workspace defaults off to set one-off writing style, em dash, and emoji rules.
• Add per-post PDF context uploads (up to 3 PDFs, 10MB each).
• Set an optional image prompt for featured image generation.

Featured Image Customization

Image generation combines your topic, tone, business context, and article body. Style direction is resolved in this order:

• Per-post image prompt/style hint (when provided).
• Workspace Featured image style from Settings → AI.
• Automatic fallback style and palette selection when no style is provided.

CTA Guidance

If you want specific calls-to-action, include them in your business context and style guide (for example: preferred CTA phrases, offer framing, tone constraints, or where CTAs should appear).

The blog generator uses business context to incorporate relevant CTAs naturally, rather than enforcing a rigid CTA template.

Best Practices

• Keep your company context specific: audience, offer, differentiators, and CTA preferences.
• Use per-post overrides for exceptions, not your default workflow.
• Upload source PDFs when factual accuracy matters (research, product docs, internal strategy).
• Review generated content in Content before publishing to external channels.