Tips for Better Results with AI

The assistant writes from what it is given: your profile, your rules, the post and your request. Most weak answers come from a thin request, and most high bills from long chats. These habits fix both.

Get better drafts

• Give it facts, not just a topic. The AI does not know your product's price, your test results or what happened at last week's event. Paste them into Describe my post or the chat: a request can be 10,000 characters long. A draft built on your facts needs far less fixing.
• Say who it is for and how long. "For beginners, around 1,200 words, one H2 per step" gets you a post you can use. "Write about houseplants" gets you an average one.
• Fill in Target Audience once. It is sent with every request, so no writer has to repeat it. See How to Teach the AI Your Writing Style.
• Use the guided steps for a new post. Choosing the title and then the outline before the post is written costs two short requests and saves rewriting a long one that went the wrong way.
• Fix the outline, not the post. If the outline options are close but not right, say so in the box ("Use the second outline but add a section on costs") before the content is written.
• Point at the part you mean. With ⋯ (Choose parts of the post), "make it punchier" changes the excerpt you picked, not the whole post.
• Reply instead of regenerating. Regenerate throws the answer away and rolls the dice again. A reply such as "Good, but shorter and keep the statistic" keeps what worked.
• Read every change card. The count ("2 changed · 1 added · 0 removed") and the Before toggle show exactly what moves. A whole-post edit that should touch one paragraph but says "6 changed" deserves a closer look.
• Check facts, names and numbers yourself. The AI can state wrong things confidently. You publish under your name; give every statistic, quote and date a look before Publish.
• Finish with Fix grammar & spelling. On the Checklist tab it fixes mistakes and changes nothing else, so it is safe as the last pass.

Spend fewer tokens

Every request with a connected provider is billed to your account by the token, input and output. A token is a piece of a word, the unit providers count text in; the counts Foxiz shows come from your provider.

• Watch the counter. Once a chat has used tokens, the window header shows the total, such as "12.4K tokens". Point at it to see the input, the output and the last request. For scale: in our test, one full run of the guided steps, from Get ideas to a 45 block post with its excerpt, tagline, meta description and tags, used about 38,000 tokens with a Claude Sonnet model. To turn that into money, look up your model's price per token on your provider's pricing page.

  

• Start a New chat when you change subject. Each request carries a summary of the chat so far, so a long chat makes every request bigger.

• Use the small tasks for small jobs. An excerpt or tags from the Checklist costs a fraction of a whole-post edit. Whole-post edits (the Improve buttons, a content change in the chat) send and return the entire post.

• Rewrite a paragraph, not the post. Rewrite with AI on the block toolbar sends one paragraph.

• Pick the model by the job. Auto (newest) is a good default. A smaller, cheaper model pinned under AI Model still writes fine titles, excerpts and tags; the strongest model earns its price on long articles.

• Turn off what your team never uses. A task unticked under Editor Features > Tasks leaves the editor and the chat.

• Cap it twice. Set a monthly spending limit in your provider's console, and a Monthly Token Limit in Access & Limits if you want the site to stop first. Usage details shows which task, model and user spends the most.

• Switch to Prompt mode for heavy drafting when you already pay for a ChatGPT, Claude or Gemini plan: it builds the same prompts and costs the site nothing. Copy prompts with Get the prompt instead after a failure too. See How to Use Prompt Mode.

Work faster in the window

• The window remembers its place. Drag it beside your text, or collapse it with the arrow at its top left while you write.
• From the keyboard: Alt+Arrow keys move the window, Alt+Shift+Arrow keys resize it.
• In the block editor, open the assistant from the command palette (Ctrl+K or Cmd+K) with Open Foxiz AI Assistant.
• A chat lives in its browser tab. Keep the tab open while you work on a post; closing it ends the chat, but not what you added to the post.
• Administrators can turn on Prompt Inspector under Access & Limits. Each answer then gets a View prompt link showing exactly what was sent, with its token counts: the fastest way to see why an answer ignored a rule.

When something goes wrong

What you see	Why	What to do
No Foxiz AI box in the editor, and no AI Assistant in the Foxiz menu	The Foxiz license is not activated	Activate it: How to Activate the Theme
Nothing at all, on an activated site	WordPress is older than 6.2	Update WordPress
The menu item is there but the editor box is not	Enable Foxiz AI Assistant is off, the post type is not ticked under Post Types, or your role is below Minimum Role	An administrator changes it in Foxiz > AI Assistant
No Rewrite with AI on the block toolbar	Prompt mode, the Classic editor, or the Rewrite paragraph (block toolbar) task is off	Connect a provider, or tick the task
Prompt mode, though you added a key	The provider refused the key or did not answer, or WordPress is older than 7.0	Read the note at the top of the window; see How to Connect an AI Provider
"Request limit reached. Try again in … minutes."	You used up Requests per User (10 min)	Wait, or ask an administrator to raise it
"This month's AI token limit is used up."	The site reached Monthly Token Limit	Wait for the 1st, raise the limit, or copy the prompt into a web AI chat
"The answer was cut off or could not be read, so nothing can be added."	The answer was too long or broken	Retry, or Get the prompt instead
An answer cannot be added: it lost an image, changed a shortcode or came back incomplete	The assistant protects your post	Regenerate; see How to Edit a Post with AI
"Your request is over 10,000 characters. Shorten it to send."	The request is too long	Send the material in two requests, or trim it