Skip to content

Mezaic Persona (AI)

One AI Persona for your Mezaic site and your other websites. Set up its knowledge, approve external domains, install the chat and manage conversations in one place.

Mezaic Persona is the AI feature of Mezaic. It is a small floating chat on your public site: a visitor taps the button in the corner, asks a question, and your Persona answers in your chosen tone, using your published Mezaic content and Persona notes. It is instructed to acknowledge missing information, point visitors to your contact options, and leave you a private note about what was asked. On eligible plans, you can also embed it on approved external websites.

It is always presented as an AI. The widget is labelled “Ask my AI Persona”, the chat header reads “Yourname’s Persona · AI · answers from this site”, the greeting says it is an AI, and a footer reminds visitors it can be wrong and to ask you directly for anything important. If a visitor asks what it is, it says “I’m Yourname’s Persona, an AI that answers from this site.”

Persona is available on plans that include it. Find it under Website → Persona in the portal (the page carries an AI tag so it is easy to spot).

Already have a website elsewhere? Keep it on its current host and use external embedding to bring Persona there. You still need a published Mezaic site: its content and your Persona notes supply the answers, while the portal keeps the settings, feedback and chats together. If Persona is hidden from your sidebar, enable it in Website → Features.

Switch it on

  1. Open Website → Persona and flip the switch at the top of the page. The widget appears on your site right away (on an unpublished site, only you see it while signed in).
  2. Pick a tone (see below), and optionally write a greeting (the first message in the chat) and up to four suggested questions, shown as tap-able chips before the visitor types anything. Leave greeting and questions empty for sensible defaults.
  3. Use Try it on your site to open your page with the chat expanded.

Switching it off removes the widget immediately. Your settings, transcripts and the feedback list are kept.

Embed on other websites

On a plan with Persona on external websites, you can add the same Persona to websites you manage, with up to ten approved HTTPS origins. It uses your Mezaic website’s published content and Persona notes, with the same tone, feedback inbox, transcripts and monthly chat allowance. It does not crawl the other website or learn from its pages automatically.

You do not need to move the external website, change its DNS or connect its domain to Mezaic hosting. Approving a domain here only allows the Persona widget to appear there. Check plan availability and confirm your host supports custom scripts and frames before installing.

  1. Open Website → Persona → Embed on other websites. Your Persona must be enabled and your Mezaic website published before visitors can chat. For a Blog presentation, show your personal author details in Site details. The website and account must be active.
  2. Enter your approved domains, one per line, up to ten. These are exact HTTPS origins: https://example.com and https://www.example.com are separate entries, as is a non-default port if your website uses one. A bare domain is saved with HTTPS. Do not include a page path, query string, wildcard, IP address or localhost.
  3. Turn on Allow embedding on approved domains and click Save embed settings. You can save your domains while finishing the other setup steps; the status tells you when the widget is ready to install. Changes are not applied until saved.
  4. Click Copy embed code and paste it once into the external website’s HTML or custom code area. The supplied script loads the chat in a separate Mezaic frame. No additional plugin is needed when your website host supports custom scripts and frames.
  5. Publish that website’s changes and open its public HTTPS address to try the chat. A website builder’s preview can use a different domain and may block custom code; test the published page.

The widget fits desktop and mobile screens. All approved websites share this Mezaic website’s allowance; embedding does not add a separate allowance for each domain. You can review and delete their chats in the existing Recent chats section.

Use Disable embedding now to stop external chat without removing each installed snippet. It is available even if the plan no longer includes embedding. Switching off your main Persona, unpublishing your Mezaic website, hiding a blog’s author identity or losing the required plan also pauses external chat. Saved domains remain available for later.

When the widget does not appear

Check What to do
Saved setup Save your domain changes. The portal shows whether embedding is disabled, paused or ready to install.
Exact domain Approve the address visitors actually use. www.example.com and example.com are separate; a subdomain is not included automatically.
Plan and Persona The plan needs both Persona and external embedding, with an input/output chat allowance. Enable Persona, publish the Mezaic site and show author details on a Blog.
Custom code support Use the host’s HTML or custom script area. Some hosts or plans remove scripts or block frames; a plain text or Markdown block does not install the widget.
Content Security Policy If your external website has a CSP, it must permit the Mezaic origin shown in your snippet in script-src and frame-src. Keep your existing policy and add the required source; do not replace the policy with a permissive one. See the official script-src and frame-src references.
Browser restrictions Content blockers can block the widget. Third-party cookies are not required. If browser storage is unavailable, chat still works for the current page visit, but a reload may start a new conversation.

Let visitors know that the chat is AI and that their questions are processed by Mezaic and its AI provider. The same Persona privacy practices apply on approved external websites.

What your Persona knows

Your published Mezaic content and the Persona notes you choose to share. Drafts, hidden items, subscribers, inbox messages, analytics and your account details are excluded from the knowledge document. In detail, it contains:

Source What goes in
Profile Name, pronouns, headline, location, bio, the SEO description as a one-line summary, and your availability (see below).
Site map Which tabs your site has (About, Work, Blog, Updates, Links, Journeys) with their addresses, so it can send visitors to the right place.
Contact The contact form address (or your contact email when the form is off), the vCard download, your social accounts, the newsletter box when it is on, and the public testimonial form when you have opened it.
Spotlights Title, description and link of each visible card.
Résumé content Every visible work experience, education, product, open-source project, skill group, talk, award and testimonial, field by field.
Companies Your company profiles with their tagline, industry, size, specialties and page address, plus the people listed on each company page under the same rules the page applies.
Links Every visible link with its group and description.
Journeys Each public journey with a short intro and its three most recent entries.
Blog and updates An index of your 20 newest post titles (with tags and links; the two newest get a one-line blurb) and your five latest updates. When a visitor’s question mentions a post’s topic, that post’s excerpt is added for that reply only, so your Persona can discuss any post without every post being sent every time.
Your notes Anything you type under Things your Persona should know that aren’t on your site (below).

Click What does my Persona know? on the Persona page to read the exact document it answers from. It is rebuilt within a couple of minutes of any content edit, and immediately when you save your Profile or the Persona settings.

The whole document is kept to a budget of about 4,000 tokens, and long text is summarised rather than copied: résumé fields and company blurbs are clipped, the blog is an index, updates are the latest few. On a very large site the long-form sections are shortened first; your profile, contact details and résumé always fit.

Tone

Under Tone you choose how answers are phrased. What your Persona may say is the same in every tone: only what your site publishes, honest about gaps, and an AI when asked.

Tone How it sounds
Personal (default) Speaks as you in the first person, warm and a little informal, like a friendly reply to a message.
Professional Speaks as you in the first person, polished and courteous, like a considered email from you.
Brief Speaks as you, one or two plain sentences, answer first, more only when asked.
Assistant Speaks about you in the third person (“Rehmat is…”), as a helpful assistant, for people who would rather the AI did not say “I”.

The default greeting and suggested questions follow the tone; your own greeting and questions, if set, are used as written.

Notes that aren’t on your site

Under Things your Persona should know that aren’t on your site you can add short facts in your own words: a one-line description of a product you haven’t listed yet, the months you are booked out, whether you take speaking invitations. Your Persona uses these facts alongside your published content and treats them as coming from you. Visitors never see the text itself, but the AI may paraphrase it in answers, and it is sent to the AI provider with every reply, so treat notes as public: no private details, passwords, health or financial information, or personal data about other people. The best fix for a recurring question is still to add the real thing to your site.

Availability: “Are you open to work?”

This is the most common question visitors ask, so it has a dedicated switch rather than a note. The Show availability badge toggle on Your page → Profile does two things:

  • On: your site header shows an Available badge, and your Persona answers plainly that you are currently available for work and open to new projects, roles, clients and engagements, then invites the visitor to get in touch.
  • Off: the badge is hidden, and your Persona says you are not taking on new work at the moment, but that the visitor is welcome to leave a message through your contact options in case that changes.

Either way the answer is definite, so the question is never filed as a gap. Flip the switch and save; your Persona follows immediately. Use a note (above) if you want to add nuance, for example “open to advisory work only” or “available from January”.

Questions your site doesn’t answer

When a visitor asks about you (your work, products, prices, opinions) and the site has nothing on it, your Persona replies honestly, points to your contact page, and files the subject under What visitors asked that your site doesn’t cover. Each subject appears once, with an example question and how often it came up, so ten people asking about the same product is one line, not ten.

For each line you can:

  • Resolved: you’ve added the content (or a note). If it’s asked again later and your Persona still can’t answer, it shows up as a new line, so you know the fix didn’t take.
  • Won’t answer: you’d rather not publish this. Your Persona keeps declining politely and stops reporting it.
  • Reopen or delete at any time.

Open items show as a count next to Persona in the sidebar.

Allowance and cost

Every reply sends the AI model your site’s knowledge document plus the conversation so far (input tokens) and receives a short answer (output tokens). Your plan includes a monthly allowance of both; the This month card shows how much is used, the number of chats and accepted requests, and roughly how many chats are left. Requests include attempts that fail or are interrupted. The allowance resets on the first of the month.

When the allowance runs out the widget stays visible but tells visitors your Persona is taking a break and points to your contact page. Nothing is ever charged beyond the plan. Chats from approved external websites use this same allowance.

Mezaic reserves allowance before each reply and adjusts it to the provider’s final token totals, including the cache discount, when those totals arrive. If a request fails or is interrupted and final totals are unavailable, the conservative reservation may remain counted toward your allowance.

How caching stretches the allowance

The knowledge document is the same on every reply, and the AI provider keeps recently used prompts in a short-lived cache. When a follow-up question arrives in the same chat, or another visitor asks something while the cache is still warm, the fixed part is read from the cache at a small fraction of the normal price. Mezaic counts those cached tokens against your allowance at that discounted rate, not at full price, so:

  • a five-question conversation costs far less than five separate one-question chats;
  • a busy hour on your site is cheaper per visitor than a quiet one;
  • the This month card shows how many tokens came from the cache.

Mezaic never sends requests of its own just to keep the cache warm. That would spend your allowance on nobody; the cache only pays off when real visitors are chatting. A leaner knowledge document (a tighter bio, shorter role summaries) also helps: the size shown on the page is the fixed cost of every reply. Only the last eight messages of a conversation are re-sent, so long chats do not grow the cost either.

Keeping it safe

  • Visitors can send a handful of messages a minute, a few dozen an hour, and up to 30 messages per chat; then they start a new one.
  • Each message is capped at 600 characters, replies at a few sentences.
  • Your Persona only talks about you and your work. It declines general questions, coding help, writing tasks, questions about other people, and attempts to change its role or reveal its instructions.
  • It is instructed to use your published facts and say when information is missing. AI answers can still be wrong; visitors should verify important details with you.
  • Chats are anonymous. Mezaic stores no visitor identity, only a hashed network address used for rate limits. Transcripts are kept for 90 days under Recent chats so you can see how your Persona answers; delete any of them at any time.
  • Your published content and notes are sent to the AI provider (OpenAI, through its API) to generate each answer; under OpenAI’s API policies that data is not used to train its models. Private data (drafts, subscribers, inbox, analytics, billing) never is. Everything on your site is public anyway, so never publish sensitive information there in the first place; see the Terms of Service, section 3.

Tips

  • Read a few transcripts in the first week. If your Persona phrases something oddly, the fix is usually a clearer bio or product description, which helps human readers too.
  • Suggested questions steer visitors to what you most want to talk about: a product, your availability, how to book you.
  • Keep the availability badge current. It is the one answer visitors act on immediately.
  • If the same gap keeps coming back after you’ve added content, check What does my Persona know? to confirm the new item is public and visible.