Detecting-AI

Screen writing for AI-generated content and plagiarism, and humanize text, through the Detecting-AI API. Agents check submitted or outsourced copy before it is published under your name.

3 actions API key available

Detecting-AI website ↗ Platform Documentation ↗ Capability data verified 2026-08-01

A piece of writing arrives through the submission intake

Agent calls Detect AI Content on the text, pinned to the detector version this process was tuned against

Agent calls Check Plagiarism on the same text against published sources

Agent stores both results with the submission ID, the detector version and the time of the run

Agent routes only the submissions that cross the review threshold, with both results and the source text attached

A reviewer reads the writing itself and decides what, if anything, the results mean

What This Integration Enables

This connector does three things. Detect AI Content analyses a text and reports how likely it is to have been generated by a model rather than written by a person, across multiple languages. Two detector versions are available, so an established process can stay pinned to the version it was calibrated against while newer work adopts the latest. Check Plagiarism compares a text against published sources and returns what it found. Humanize Text rewrites AI-generated draft copy so it reads more naturally while preserving meaning, with two rewriting models, Cognia and Lexi, that differ in style. Running the first two together on one submission gives a reviewer both signals in a single pass.

The important thing to be clear about is what these outputs are. A detection result is a probabilistic assessment of a text. It is not a determination about a person, and it does not become one by being high. Systems that treat a score as a verdict produce outcomes that nobody can defend afterward, particularly when the subject is a named individual with a stake in the result. FlowRunner is built for the opposite arrangement: the agent screens at volume, and the digital andon cord stops the line at the point where a person's standing is on the line. That is not a limitation bolted onto the workflow. Human-in-the-loop is what makes a screening process usable at all, because it is the only version of it a reviewer can stand behind.

Without FlowRunner

Uneven screening Some submissions get checked and some do not, depending on who has time

No record of the check A score exists in someone's browser history, with no note of which detector produced it

Everything or nothing Either every submission gets read closely, or none of them do

With FlowRunner

Consistent first pass Every submission is screened the same way, with the same detector version, in the same order

A logged result Each check is stored with its submission ID, detector version and run time

Reviewer time on the few The reviewer reads the submissions that surfaced, with the evidence already assembled

Use Case Scenarios

Screening a submission queue so reviewers see the right ones first

Submissions arrive through a form. The agent runs Detect AI Content and Check Plagiarism on each one, then writes every result to Google Sheets with the submission ID, the detector version and the timestamp, so the whole queue has a consistent record whether or not anything surfaced. Submissions crossing the threshold post into the reviewer's channel in Slack with both results and a link to the text. The reviewer opens the submission itself. The queue is ordered by evidence rather than by arrival time.

Checking outsourced copy against published sources before it goes out

An agency receives an article from a freelance contributor. Before it reaches a client draft, the agent runs Check Plagiarism against published sources. A clean result means the draft continues through the normal editorial flow into WordPress. A match means the flow stops and the editor sees the matched passages next to the sources they matched, so the conversation with the contributor starts from specifics rather than from an accusation.

Tracking screening results over a period rather than case by case

Individual scores tell a reviewer very little on their own. A scheduled flow appends every result to a sheet with its date and detector version, and the pattern across a term or a quarter is what a program owner actually reads. When a change in the mix appears, it prompts a conversation about the process, not an action against any single submission.

Human-in-Loop Highlight

There is exactly one gate on this connector and it is not negotiable in the design: a detection score about a named person never becomes an outcome on its own. Detect AI Content returns a likelihood, and likelihoods are wrong in both directions. A careful writer with a plain style can score high. A model-written piece with light editing can score low. Attaching a consequence to that number automatically means attaching consequences to a coin flip that has someone's name on it. So the agent's job stops at assembling the case. It posts to the reviewer: "Submission 4471 from a named contributor. Detect AI Content returned a high likelihood using detector version 2. Check Plagiarism returned no matched sources. Full text and both raw responses attached. This is evidence for your review, not a determination. How do you want to proceed?" The reviewer reads the writing, weighs the context they have and the agent does not, and decides. The agent records who decided and when. Every other benefit on this page depends on that step existing, because a screening process that nobody has to answer for is one nobody should trust.

Agent processes routinely

Detects I Content returns a likelihood

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

3 actions

Detection

Detect AI Content Analyzes a text and reports how likely it is to have been generated by an AI model rather than written by a person, working across multiple languages and returning the result in the response's data field. Two detector versions are available, so an existing flow can stay pinned to the version it was tuned against while newer work adopts the latest model. Used as the first screening pass, with the version recorded alongside the result.
Check Plagiarism Checks a text for material copied from existing published sources and returns the findings, supporting several languages including English, Spanish, French and German. Used alongside Detect AI Content to screen a submission for both copied and machine-generated writing in one flow, and to give a reviewer specific matched passages rather than a summary judgment.

Humanizing

Humanize Text Rewrites AI-generated text so it reads more naturally while preserving the original meaning, returning the rewritten copy in the humanized text field. Two rewriting models, Cognia and Lexi, differ in style, so both are worth testing against a house tone of voice. Used as a drafting step on your own AI-drafted copy, before it goes to a human editor.

Frequently Asked Questions

What can FlowRunner do with Detecting-AI?

FlowRunner agents can run Detect AI Content, Humanize Text, and Check Plagiarism in Detecting-AI.

Does connecting Detecting-AI to FlowRunner require OAuth?

No. Detecting-AI connects to FlowRunner with an API key, no OAuth flow required.

Can Detecting-AI trigger a FlowRunner workflow automatically?

Detecting-AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with Detecting-AI

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us