Improve a post with AI
When a social media post is In review, you can ask the AI to improve it without writing it from scratch. It starts from the text you have on screen right now, so your edits are respected.
How to use it
- Open the publication (in In review status) and go to the Post tab.
- Press Improve with AI, below the text.
- Write what you want to change or add: "make it shorter", "add the figure from…", "give it a friendlier tone". If you leave it empty, the AI improves the text in general.
- To rely on other content, type "@" and choose a WorkOS page (up to 5).
- Press Improve the post.
- The proposal appears in the post box. Undo brings back your previous text. Nothing is saved until you press Save.
What the AI takes into account
- The current text of the post.
- The publication's research, with the pages linked in it and their attachments. It also works if there is no research. Linked pages are only used if they are not private and do not come from emails, meetings or contacts created from email and, in addition, if you can read them; the rest are skipped and you are told. When you save the research, those links are removed (only their text stays). Tasks and pages from databases WorkOS does not recognise are not used either. If you edit it directly in Notion, the filter is still applied when writing, and if it skips a page, the automatic writing says so in a comment on the publication.
- The pages you cite with "@" and your instruction.
- The network's best practices and your account's personalization, just like when it writes from scratch. It keeps the post's language unless you ask for another.
What you can cite with "@"
Wiki articles, notes, documents, projects, companies, contacts, meetings and publications, as long as you can see them. Emails and messages cannot be cited: if you choose one, the AI will only see its title and you will be warned.
Save your instruction as research
Tick Save as research information to add your instruction to the publication's research section. It is useful when you provide new data that should count in later improvements. It is unticked by default.
What happens to your data
When you press the button, the post text, the research, the cited pages and your instruction are sent to OpenAI. Cited pages are read with your own permissions. From the button, emails and messages are not sent, and OpenAI does not use the data to train models. The AI usage log records only the model, the tokens and the user, never the text.
From Kairos
You can also ask Kairos: "improve this week's LinkedIn post and make it shorter". Kairos proposes the text and saves nothing; if you approve it, it asks for confirmation to save it and shows you the exact text. Pages cannot be cited with "@" this way: if it needs data from other pages, Kairos reads them and summarizes them in the instruction, and that summary is also sent to the AI. The proposed text stays in your conversation with Kairos.
Where it is available
- Only on social media publications with a post text and in In review status. It does not appear on newsletters, Web articles or long YouTube videos.
- One improvement runs at a time, with a maximum of 10 every 10 minutes.
Tip: if you are not happy with the result, press Undo and try again with a more specific instruction.
Regenerate the post from the research
The Regenerate with AI button appears next to "Improve with AI" when the publication is in review and has research (without research, the button is disabled) and writes the whole post again from the research information and the pages it links to, without you having to record or type it again. It replaces the current text, including anything you have edited, and has no "Undo": that is why it asks for confirmation; if you care about the current text, copy it first. When it finishes, open the publication again to see the result. If the AI cannot write it, the publication goes back to "In review" with the previous text and a comment tells you. A publication that is already scheduled or being published cannot be regenerated: cancel the schedule first.
If the research links a note or wiki article that is not shared with the team, the AI only reads it if you cited it with "@" when you created the publication by voice, less than 7 days ago. In any other case (linked by hand, or regenerated after 7 days) it is skipped and the publication tells you in a comment: share that page so the AI can keep using it.
The Naturalness score and the Humanize button
Below the text of a social media post you will see a Naturalness score from 0 to 100 and the Humanize button. The score is calculated in WorkOS with simple rules, with no AI, without sending the text to any external provider and without storing it, and it measures five checks: rhythm (sentences should not all be the same length), concreteness (figures, proper names and links), filler (stock phrases), fingerprint (long dashes and too many emojis) and voice (you can tell a person is speaking). The score mixes the average of the five with the worst one, so one weak check weighs more.
If there are no warnings but the score is below 100, Where the score is lost appears under the chip: the weakest checks, with what lowers them and what to do. It is only a guide and not an AI detector: a 100 only means none of the flaws it looks for are present.
Humanize only fixes the mechanical things (invisible characters, long dashes, some filler phrases) and can be undone; nothing is saved until you press Save post.