Check an AI draft in Google Docs before you send it.
Publish Check is a free add-on for Google Docs. Open it from Extensions and it reads the document: the phrases that sound machine-made, the claims to check, markdown left over from a chat, blanks the AI left for you, private details and the links to open. Select a prompt and it grades it out of 10, with the same ten checks as the Prompt Grader. It all runs inside Docs, and nothing is sent to thinkhard.si.
Coming to the Google Workspace Marketplace. Updates are posted on the changelog.
What it checks
- Phrases that sound machine-made. Filler openings such as "Certainly", AI-favorite words such as delve and seamless, long dashes and the other habits readers notice, with the exact phrases. Click one and the cursor jumps to it. The same rules as the Robot-Speak Checker.
- Claims to check. Every sentence with a number, a date, a quote, a named study or a source, riskiest first, because these are the details AI makes up most often. The same rules as the Fact Check List.
- Markdown left over from a chat. Stars, pound signs and [links](like this) that come along when an answer is pasted from ChatGPT or Claude.
- Blanks the AI left for you. Text such as [insert company name] that is easy to miss in a long document.
- Private details. Email addresses, phone numbers and long numbers like card or account numbers, so nothing goes out by mistake.
- Links to open. Every link in the document, because AI writes links that look right and lead nowhere.
One press highlights everything it found in the document, and another clears the highlights again.
Grade a prompt where you wrote it
Many people draft their prompts in a document first. Select one, choose Grade, and Publish Check scores it out of 10 on the ten checks, laid out for ChatGPT, Claude, Gemini or any chat model, with the plain question to answer for each check it misses. Open in the Grader takes it to the full Prompt Grader to rebuild it.
Private by design
The checks run in the add-on's sidebar, inside Google Docs. Publish Check sends nothing to thinkhard.si or anyone else and stores nothing. It can read only the document you open it in, and it changes that document only when you ask: a highlight you can clear, or the cursor moved to a phrase. The buttons that open thinkhard.si put your text after the # in the address, which browsers never send to a server. The privacy page says the same.
Questions
Does it tell me whether AI wrote a document?
No, and no tool can tell reliably. It shows the habits that make writing read as machine-made and the details that need checking, so you can fix them before your reader notices.
Does it work in Google Sheets or Slides?
Not yet. It is built for Docs, where most AI drafts end up.
Why does it ask to view and manage the document?
To read the text it checks, and to highlight the phrases when you press Highlight. It can reach only the document you opened it in, never your other files.
Where are the other thinkhard.si integrations?
On the integrations page: the WordPress plugin, the browser extension, the MCP server for Claude and ChatGPT, the command line for CI and the REST API.