Client
AI Training
The per-client knowledge base every AI-written asset is built from.
AI Training is shown under Client. It is a per-client knowledge base, and every AI-written asset for that client is built from one training record. That makes it the biggest single lever on content quality.
What the model knows with no training
With an empty record, the model works from the business name and website alone. Generation runs as usual and the output reads generic, which is why a filled record is the biggest lever on content quality.
What one record holds
| Field | What it does |
|---|---|
| Bot name | Required. Labels the record in the left panel. |
| Business name | Required. Used verbatim in generated copy. |
| Business URL | The site the content is written for. |
| Services | Free text, shown back as tags. Feeds service pages and article topics. |
| Service areas | Drives geo and local content. |
| Bot tone | Professional, Friendly, Casual or Authoritative. Defaults to Professional. |
| Description | Long-form context about the business. The highest-value field after services. |
| Knowledge base documents | Uploaded source material about the business. |
A client can hold several records, switched from the left panel. The detail view summarizes what the training powers: blog content, Google Business posts, SEO recommendations and Jack AI chat replies.
Documents
- PDF, DOC, DOCX, TXT and CSV, up to 10MB each, several at once.
- Stored in the client’s own folder and reachable only by people with access to that client.
- Editing a record keeps existing files and appends new ones.
The app asks for material in roughly this order of value: brand guidelines, tone of voice and taglines, sales collateral, positioning notes, services and pricing, then proof and FAQs.
How the training reaches the output
At generation time the app loads the training row and prepends a business context block to the prompt: description, services, service areas and tone. That is layered with the client record (name, phone, address, website) plus inventory, existing article titles, categories and topical map data where relevant.
Features that read it
- Blog articles, auto-published daily posts and bulk content.
- Geo articles and press releases.
- Google Business posts and the daily GBP queue.
- Page rewrites and topical maps.
- Keyword research, keyword ideas and content gap analysis.
- Jack AI chat.
Where uploaded files are actually read
| Job | Files used | Characters per file |
|---|---|---|
| Manual blog article | First 5 | 2,000 |
| Bulk generation | First 3 | 1,500 |
| Daily blog job | First 3 | 1,500 |
| Press releases, AI receptionist instructions | File names only | Contents not read |
Automatic filling
- The setup wizard writes a stub row (business name, website, industry as services, city as service area), flagged as onboarding context rather than training.
- A scheduled job runs every 30 minutes across five clients per run. It scrapes the homepage plus tracked keywords, then creates a missing record or repairs a thin one.
- That repair overwrites tone and description, keeps names and URLs, merges services and service areas without duplicates, never touches uploaded files, and skips recently updated records.
Launchpad’s "Train the AI on Your Brand" step stays open until a record holds real training: a tone set, at least one file, or a record that was not created by the wizard.
How files are used
Files are read as plain text. TXT and CSV contribute their content directly. Each file contributes its opening characters, and a handful of files are included per generation run. Each generation run works from a single training record for the client, so a second record acts as a separate profile. Features such as rankings grid keywords, prospect keywords, SEO goals and visibility prompts read the live website directly.
Highest return for the least work
Fill description, services, service areas and tone, then upload brand guidelines as a text file.