The more I read about AI disclosure, the more I wanted to explain how I use AI in my blog.
Research published in Organizational Behavior and Human Decision Processes shows that AI disclosure "paradoxically erodes trust," and the "analysis did not point to either AI familiarity or AI usage diminishing the negative effect." Similar results were found in advertising. Overall, disclosure of AI-generated ads reduces trust in the product and the company, similar to the effect of sponsorship disclosures.
Closer to home, in business communication, a survey of working professionals by Peter Cardon and Anthony Coman published in the International Journal of Business Communication found that AI disclosure led readers to “begin to question the authorship, confidence, caring, sincerity, and ability of senders.” As an illustration, Jacob Zucker at the University of Massachusetts Amherst posted on LinkedIn that he won’t tolerate when a conference presenter discloses use of AI-generated slides.
I write this with the hope that my AI use will be perceived as reasonable and positive. Or maybe it's more for me—my own humility and integrity. This disclosure approach follows Lance Cummings’ advice, which favors explaining how we use AI rather than readers relying on AI detectors, which often incorrectly identify our use.
To be clear, AI never writes for me. I often have to remind ChatGPT and Claude of my preference. My process is sometimes to ask for research assistance, always draft my own post, and then upload the draft for AI to review:
Finding Sources
Sometimes, I will ask AI to find news articles or posts that take or contradict a particular angle on a story. I may also ask for help finding company posts and statements. Of course, I always find and read the original source (when it exists!). (Once, ChatGPT paraphrased a direct quotation for some reason I don't understand.)
Proofreading and Suggestions
After I have a well-developed draft, I will upload it into ChatGPT with this prompt: “Please list typos and suggestions for this blog post.” From my history, ChatGPT knows the purpose and audience. I run my second draft through Claude.
I accept some suggestions and ignore others. Some suggestions are quite helpful, for example, when I insert my own opinion more than might be appropriate given the context and audience (I have to keep my self-righteousness in check). Other times, I get stylistic suggestions, mostly to use a more formal, academic tone, which I tend to ignore.
Other suggestions, for example, for fact checking, have been useful. I might write a draft blog post on one day and run it through AI the next. In that window, more may be revealed about a story that I had missed. Or AI may find contradictory information that I missed. For example, for a blog post about the cargo flight crash at the Miami airport, ChatGPT flagged discrepancies in reported speed (touchdown or runway exit). For that post, AI also suggested that I mention the plane’s history as a converted freighter, which may be important for attribution theory and crisis response—a fact I had otherwise decided to skip.
My posts take more time with AI assistance, but I believe they turn out better with the help.
Image source (caption reads, “Image created with Chat GPT [sic].”