Check For AI (CheckforAi)

The tool was free and sold nothing. The home page called it "a free nonprofit project for individuals" and offered no plan, no price and no allowance figure (archived home page, 27 May 2023, read 10 November 2026), so the card carries no plan table and the free tier fields record a free account with no stated quota. An account was required, so the no-account flag stays off.

On retention, the Privacy Notice (Effective & Last Updated 1/1/2023) states in section 7: "We will only keep your personal information for as long as it is necessary for the purposes set out in this privacy notice, unless a longer retention period is required or permitted by law", and on termination: "we will deactivate or delete your account and information from our active databases. However, we may retain some information in our files to prevent fraud" (archived privacy notice, captured 23 March 2023, read 10 November 2026). No period is stated for uploaded text, so the retention rank field is empty. The Terms of Use of the same date add that "any Contributions you transmit may be treated as non-confidential and non-proprietary", and they name the operator: a "legally binding agreement made between you ... and Uphold.ai", governed by the laws of the State of California with venue in San Francisco (archived terms, captured 24 March 2023, read 10 November 2026). Support ran through support@uphold.ai and ben@uphold.ai, per the archived footer.

Check For AI published no accuracy figure of its own. The March 2023 home page, the version Weber-Wulff 2023 tested, attributed a figure to its component model: "Our analysis leverages the combination of Open AI's existing Roberta-base model for GPT detection, and our own proprietary models. Open AI's Roberta-base boasts a 95% (accuracy), and our custom models help to limit false readings. We are currently in our first iteration, and look forward to increasing accuracy through feedback" (archived home page, captured 14 March 2023, read 10 November 2026). The May 2023 page dropped the sentence. Because the 95% describes OpenAI's model and not the product, the claim gap field is empty and the figure is quoted here as the vendor wrote it.

The vendor's caveat was unusually plain for 2023. The Learn More page read: "no model is perfectly accurate and even text that is 100% human made could be flagged as high risk. This is also the case for other AI detection tools. Risk scores are simply conversions of model probability outputs. Therefore, no decision should be made solely based upon any risk scores", closing with "we cannot guarantee the accuracy of these predictions" (archived Learn More page, 11 December 2023, read 10 November 2026). No false positive rate accompanied it, and no statement about paraphrased or humanized text was made.