AI features
You can help candidates find jobs and help employers review applicants with AI. The plugin works fully without it. Add a provider key under Career Board > Settings > AI Settings and the tools below turn on.
- AI Chat Search - candidates search jobs in plain language.
- Recommended jobs - matches for each candidate, based on their resume.
- Applicant ranking - a 0-100 fit score and a one-line reason per applicant.
- Applicant summary - a one or two sentence summary of each candidate.
- Cover letter writer - candidates draft a letter in the apply panel.
- Description writer - “Generate with AI” on the post-a-job form.
Step-by-step guides for candidates and employers, and AI troubleshooting, are in the WP Career Board (Free) docs under AI features: Overview, Setup and providers, Candidate AI features, Employer AI features, and Troubleshooting. This page covers the Pro side.
Where each feature lives
Section titled “Where each feature lives”| Feature | Surface |
|---|---|
| Job embeddings | Created when a job is created, if an embedding provider is set. Index existing jobs (AI Settings tab) backfills the rest. |
| AI Chat Search | Block wcb/ai-chat-search, shortcode [wcbp_ai_chat_search], POST /wcb/v1/ai/match |
| Recommended jobs | POST /wcb/v1/ai/match and GET /wcb/v1/candidates/{id}/matches |
| Applicant ranking | Employer dashboard and GET /wcb/v1/ai/ranked-applications/{job_id} |
| Cover letter | “Generate” in the apply panel and POST /wcb/v1/jobs/{job_id}/ai-cover-letter |
| Description writer | Post-a-job form and POST /wcb/v1/jobs/ai-description |
Providers are chosen per task
Section titled “Providers are chosen per task”Choose one provider for analysis and one for matching. Each has its own key and model.
| Task | Providers | Notes |
|---|---|---|
| Analysis and ranking (ranking, summaries, cover letters, descriptions, chat) | Anthropic Claude, OpenAI, Ollama | Claude defaults to Sonnet |
| Matching (embeddings) | OpenAI, Ollama | Claude has no embeddings, so it is not offered here |
| Provider | Fields | Defaults |
|---|---|---|
| OpenAI | API key, model (GPT-4o mini or GPT-4o), embedding model (3-small or 3-large) | GPT-4o mini, text-embedding-3-small |
| Anthropic Claude | API key, model (Haiku, Sonnet or Opus) | Sonnet |
| Ollama (self-hosted) | URL, model, embedding model. No key. | http://localhost:11434, llama3, nomic-embed-text |
Fastest setup: one OpenAI key, with OpenAI chosen for both Analysis and Matching. Ollama keeps all data on your server but needs server CPU or GPU. Developers can add a provider: see AI providers and endpoints.
Applicant ranking
Section titled “Applicant ranking”- What the model reads: the job title and description, the cover letter, the screening answers from the Field Builder, and the resume the candidate applied with (their profile resume if they chose none). Guests are scored on what they sent. An uploaded PDF CV is not read.
- Auto-score applicants on submit (off by default): each new application is scored in the background about 30 seconds after it arrives, using one model call.
- Rank on demand: the ranking button returns applicants already scored at once, lists the rest as Not scored, and queues them. Scoring runs in batches of 25 in the background. Newly scored applicants appear the next time the dashboard loads. Only the newest 500 applications of a job are ranked per pass.
- Not scored: if the provider is down or replies with nothing usable, the applicant shows Not scored, not 0%. Nothing is saved and scoring is retried up to three times, after 5, 10 and 20 minutes.
- Cost: scores, reasons and summaries are saved on each application, so reopening the dashboard never bills the model again.
Telling applicants
Section titled “Telling applicants”Career Board > Settings > AI Settings > Tell applicants shows applicants “Employers on this site may use AI to help review and rank applications.” in the job page apply panel. It shows only while an analysis provider is set up. It is on for new installs and off for sites upgraded from before 1.8.0, where an admin notice lists it as a recommended safer default.
Cover letters
Section titled “Cover letters”The cover letter writer uses the signed-in candidate’s resume and the job’s title and description (first 2,000 characters). It works only for published jobs, so a draft or pending job never leaks into a letter. Candidates need permission to apply for jobs.
Limits
Section titled “Limits”- Each signed-in user can make 30 AI requests per hour across the AI routes. Only requests the user is allowed to make count; refused or signed-out requests use nobody’s budget. After 30 the route returns 429.
- Matching is computed in PHP over the newest 2,000 jobs with embeddings. Developers can change that with the
wcbp_ai_match_scan_capfilter. - No resume file parser. Resume text for matching comes from the candidate’s resume fields.
- No “Test connection” button: check a provider by generating a description on the post-a-job form.
- No WP-CLI AI commands.