Your connected AI assistant can write a client's content itself and save it straight into ContentForge. It writes to the same rules as your agency's own writing, and writing and saving this way is free.
What your assistant gets
Before it writes, your assistant asks for the writing guidelines for the client and the kind of content (get_writing_guidelines). It gets:
- the client's brand voice (the full one for pages, blog posts, landing pages and emails; the short one for posts, texts and ads);
- your agency's own writing rules for that kind of content (Settings → ContentForge writing and Blog writing rules) and your banned words;
- our writing rules in plain words: only use facts the client has given you, sentence shapes to avoid (each with how to fix it), link and button words that say where they go, and never say who or what wrote it (the content is the business's own);
- the exact parts and limits the save takes: for example caption lengths and hashtags per network, a press release's word count and links, a text's length, a landing page's fields, and an ad's headline and description lengths;
- the client's real pages to link and its pictures, where they matter;
- the facts the client has given you, and the facts it hasn't (your assistant leaves those out).
- the same answer-first rules Zeus follows for blog posts, pages and landing pages: answer the main question in the first 40–60 words, some headings as real questions, lists and tables where they help (they stay lists and tables in the post), sources linked only when you give their address, and the business's name, services and area written the same way every time. Posts and Business Profile updates get the short version: the point in the first sentence;
- for blog posts, the article types (standard, problem / troubleshooting, local guide, comparison, cost guide) and that every post ends with its FAQ section;
- the client's price ranges per service, when set on the client's Details: the only prices the copy may state.
Kinds of content: blog posts, social posts (per network), press releases (per kind of news, including product launches), campaign emails, text messages, website page copy, landing pages, Google Business Profile posts, Google Ads ads, email replies to a conversation, automation emails and texts, Facebook and Instagram message replies in an automation, store products, and collection items.
- "Get the writing guidelines for a Smith Roofing press release about the new Round Rock office, then write it and save it as a draft."
- "Write an Instagram post for Jones Bakery about the new sourdough, following its writing guidelines."
Some AI apps also show ready-made prompts (for example write_blog_post or write_press_release): pick one, give the client and a sentence about the topic, and it starts with the guidelines already in. Apps that don't show prompts work the same way through the guidelines tool.
Checks on every save
When your assistant saves content, ContentForge checks it the same way it checks Zeus's writing, and saves exactly the words it sent. Nothing is ever reworded on save. The result lists anything to look at as recommendations, each with the words it found and a suggestion. It can also check a draft before saving (check_copy). Nothing is saved by a check.
Your own words are never changed without asking you. When the words came from you (text you typed or pasted, a design, an artifact or a document you shared), your assistant is told to keep them exactly, show you any recommendations, and change them only if you say so. It only improves words it wrote itself. For example, a "Learn more" button or two questions in a row in your own copy stay as you wrote them.
The checks look for:
- numbers, prices and claims that aren't on the client's record (this is a check for invented details, not a full fact-check);
- sentence shapes that sound machine-written, and your banned words;
- link and button words like "Learn more" or "Click here";
- lengths and limits for the kind of content (for example a post over the network's limit, a press release under 250 words, a text that runs to 3 texts, an ad headline over 30 characters).
The same checks run when your assistant:
- drafts or sends an email reply in a Business Hub conversation (the reply is still saved as a draft, or shown to you before it's sent);
- makes or changes an automation: each email and text step it wrote is checked, and the warning names the step. For steps where Zeus writes each message, the backup email or text (the one sent when Zeus can't write one) is checked; Zeus's instructions aren't. A marketing text in an automation must say how to stop ("Reply STOP to opt out"), so that line is never flagged there;
- writes a "Reply to their Facebook or Instagram message" step in an automation: the reply it wrote (or, when Zeus writes each reply, the backup reply) is checked like any message, plus the network's limit: about 1,000 characters on Instagram (emoji count as more than one) and 2,000 on Facebook Messenger. When the step replies on whichever network they messaged on, the Instagram limit is used. It also flags formatting like bold or headings (messages show it as typed), links that aren't the client's own pages, and opt-out lines. Your assistant can't send a Facebook or Instagram message straight from a conversation; replies there go through an automation step or your team;
- adds or changes a collection item: only the item's text fields are checked (not pictures, links or numbers);
- imports a CSV into a collection: the same checks on every row's text fields;
- adds or changes a store product: its name, description and search engine title and description;
- imports products from a CSV: the same checks on every product.
For an import, the warnings are grouped by row, with the row number and the product or item name, both on the preview and after the import. A big import lists the first 50 rows that have something to look at and counts the rest. The warnings never stop an import.
Content you write or save yourself in ContentForge isn't affected.
Landing pages and Google Ads ads
- Landing pages: your assistant writes the words for each part of the landing page (save_landing_page_copy). ContentForge builds the page the same way it builds one for a Google Ads campaign and adds the call button, the reviews and credentials, and the towns served from the client's record. It's saved as a draft, hidden from the menu, for you to check and publish.
- Your assistant's words are used exactly as written. Nothing is shortened or cut. If a part is over its limit, the result says so and gives a fix instead.
- The page also gets the same checks a Google Ads campaign's landing page gets: one main headline, a match to what people search, a call or quote form at the top, and phone speed (measured once the page is live). Anything that doesn't pass comes back as a warning, and the page is saved anyway.
- When your assistant names a campaign and ad group, the page is also checked against that ad group's keywords. This is ContentForge's own word match, not a Google rating. Your assistant is then offered the change that sends the ad group's ads to the new page. That's a separate step: it shows you exactly what will change, and nothing changes in Google Ads until you say yes. Publish the page first. You can undo the change later in Changes applied.
- Google Ads ads: your assistant writes the headlines and descriptions, and adds the ad to an ad group (change_google_ads). Like every Google Ads change, it first shows you exactly what it will do and any problems with the lines, and only adds it once you say yes. The ad is added paused, so nothing is spent until you turn it on.
What's free and what costs money
- Writing and saving: free. Whatever your assistant writes and saves costs no credits.
- Distribution: paid by card at checkout. Sending a press release out to news sites is still done from the client's Press Releases tab.
- Sending: from the usage balance. Campaign emails and texts are paid from the business's usage balance when they go out.
- Google Ads spend is billed by Google once an ad or campaign is on.