You store the bands. Add a custom post type for roles, with fields for level, region, base low, base high, equity range, and last update date. SleekAI reads those fields per conversation and uses them as the anchor for any counter suggestion. If a band is missing, the system prompt instructs the bot to flag the gap rather than guess. Coaches update the bands quarterly without redeploying anything.

Edge cases are exactly where you should tighten the prompt. The bot is configured to defer to a human coach above certain thresholds, such as offers over $300k total comp, equity in a Series C or later private company, or international relocations with tax complications. For those it produces a short summary of the offer, suggests two or three questions to ask, and points the user to your booking page.

Yes. After the conversation the bot can produce a one-paragraph counteroffer email referencing the specific levers discussed, formatted plain enough to paste into Gmail. The reply includes a placeholder for the recruiter's name and the role title. The user reviews, edits the tone if needed, and sends. The conversation log keeps both the suggested email and any edits the user made.

Yes, with one nuance to be honest about. Conversation rows are stored in your WordPress database in wp_sleek_ai_chat_log. The model call still goes out to whichever provider your API key belongs to, OpenAI or Anthropic or Google. Those providers have their own retention policies, often 30 days for API traffic on enterprise plans. No third-party SaaS chatbot vendor sits in between, which is the data path most coaching sites worry about.

Yes. Set a turn limit in the widget config, for example 6 messages per session, after which the bot suggests booking a real coaching call. Combine with display conditions so the unlimited-turns version only loads for logged-in members of a specific WordPress role. The token-usage log lets you spot any single user racking up usage and tighten the limits per role if needed.

GPT-4o or Claude Sonnet 4.5 handle this well. Both produce calibrated, non-shouty negotiation language and follow the system prompt's defer-to-coach rules consistently. Cheaper models like Haiku or 4o-mini are fine for the first turn of intake but tend to over-promise on later turns ('they will definitely match'), which is exactly the failure mode you want to avoid in this category.

If you write your bands and scripts that way, yes. EU offers usually quote annual gross without bonus, mention 13th or 14th month, and weigh meal-voucher and pension allowances. UK offers separate cash and equity more sharply than US offers. Store your regional playbook as separate WordPress posts tagged by country, and the system prompt can switch context once the user mentions their location.

Yes. Display conditions check WordPress user role and logged-in status, so the chatbot widget only renders for users who hold your member role, whether that comes from a membership plugin, WooCommerce subscription, or a custom capability. Anonymous visitors see a CTA to subscribe instead. Paid users get unlimited conversations on the API key you control.