Quality-check freelance drafts before an editor reads them

By General Input

Every weekday morning, overnight writer submissions get checked for AI writing and copied text, so only the questionable ones reach your editor flagged.

Integrations

  • Pangram
  • Google Drive
  • Google Docs
  • Slack

Type

Agentic Task

Categories

  • Marketing
  • Operations

Every weekday at 7am, quality-check the drafts my freelance writers submitted overnight, before an editor reads them or I approve an invoice.

Start with Google Drive "List Files" to find the documents added to my "Submitted drafts" folder since the last run. Use Drive query syntax scoped to that folder ID with a created time filter set to the previous run. On Monday, look back across the whole weekend so anything submitted from Friday evening onward is still picked up. For each file, record the file ID, the file name, the created time, and the writer, taken from the file owner or last modifying user, falling back to the document name if the account does not identify the person.

For each document found, use Google Docs "Get Document" to retrieve the latest version, and flatten the returned content structure into plain text before checking it.

Run that text through both Pangram "Detect AI-Generated Text" and Pangram "Check Text for Plagiarism". AI detection is asynchronous, so submit the text and poll the task until it reaches a terminal state before reading the result. Pace requests so several drafts in one run stay inside Pangram's rate limit. From the detection result, capture the overall AI likelihood and the segment level windows, so I know which specific passages were flagged rather than just a single score. From the plagiarism result, capture every matched source URL with its similarity score. Note that the two Pangram checks run against different base URLs, which is expected.

Treat a draft as needing review when the AI likelihood is at or above 80 percent, or when the plagiarism check returns a genuine match against an online source. Make the AI threshold easy for me to change in one place. Ignore trivial overlaps such as common phrases, correctly attributed quotes, standard disclaimers and boilerplate, so a properly cited statistic does not flag the whole piece.

Move anything that crosses either bar into the "Needs editor review" folder using Google Drive "Move File". Everything else stays exactly where it is, so clean drafts flow through untouched and my existing process is not interrupted.

Never delete, rewrite or edit a writer's document. The only change permitted is moving a flagged file between folders. Do not alter document contents, file names, or sharing settings under any circumstances.

Finish with one Slack summary posted to my editorial channel using "Send a Message". List every draft checked, with its title as a link to the document, the writer, the AI verdict with its score, the plagiarism verdict, the matched source URLs with similarity scores where plagiarism was found, and short quoted excerpts of the specific passages Pangram flagged. Say for each draft whether it was moved or left in place. Group the flagged drafts first so the editor sees them without scrolling. The goal is that an editor can open a document already knowing where to look.

Word the summary as things to check with the writer, never as accusations. These scores are one signal in an editorial judgement, not proof of anything, and a human editor makes the final call. Prefer phrasing like "worth asking about how this section was drafted" over stating that a writer used AI or copied text. If no drafts were submitted since the last run, post a short note confirming the check ran and found nothing waiting.

Related prompts

Explore more prompts
Call overdue Xero customers with an AI collections agentWin back LiveChat visitors whose chats went unansweredChat quality review board for LiveChat support leadsA searchable RFP answer library your bid team drafts fromRun your weekly Lever pipeline review and decide in the appLinkedIn creator sourcing and vetting workbench for B2BWin back no-show and cancelled appointments every morningLive Loop returns analytics with product-level drill-downNewsletter pre-flight and approval board for Mailjet sendsTurn a prospect spreadsheet into personalized sequence enrollments