This fall I invited nearly 200 people running for school board to answer a questionnaire to help inform voters. What I got back was a mixed bag. Some candidates gave colorful answers. Some capitalized entire words or used exclamation points for emphasis. Some criticized their opponents, accused their boards of dysfunction and warned against what they believe are political agendas taking over the school board. Their emotions and opinions jumped off the page. But those candidates seemed to be the minority. There were a lot of others whose answers were frankly… less interesting. They sounded like essays written by a student who has not yet found their voice but is trying to write what they think the teacher wants. They contained many of the same words and ideas — calling things “essential” or “crucial” or “important,” or deploying buzz words like “transparency,” “engagement” and “efficiency.” After we published all the candidates’ answers, one reader raised the possibility that some candidates could have used artificial intelligence tools to write their answers. This was a possibility I hadn’t considered, and I had not asked candidates to refrain from using generative AI. Such AI tools have only started becoming widely used within the past few years, and this is the first general election in which they’ve been able to play a significant role. We’re still trying to understand how it is changing everything from politics to education to journalism. AI has suddenly presented us with a whole host of practical and ethical questions about how it will become part of our jobs and daily lives, and for what purposes we’re OK with it. It has also prompted deeper questions about identity: What exactly is it that makes us human? When it comes to communication, what is AI — at this point, at least — unable to reproduce from human voices and writing? Studying the school board candidates’ questionnaire submissions more closely seemed to suggest there are many things that distinguish us as human: the ability to tell stories, to exude a feeling, to be specific, to paint an image, to recall history, to speak authoritatively, to voice a call to action — to take a real stand. After I got the reader’s tip, I read through all 100-plus candidates’ answers a second time, this time looking for language I thought may have been written with AI assistance. I ran any answers that I suspected could be AI-generated through GPTZero, a leading AI detection tool. You copy and paste any text into GPTZero, then it analyzes it and tells you the percentage likelihood that it was AI-generated or human-generated; it also identifies which specific sentences were most likely produced by AI or a human. Perhaps not surprisingly, it was easy to predict which answers would be flagged as likely AI-generated. If a candidate told a story, their answer was very likely to be deemed human. If they cited specific examples of history, events, statutes, schools or documents — if they cited anything specific at all, especially by name — it was almost certainly human. If somebody used adjectives, adverbs or nouns that were unique or suggested an opinion — words like “extravagant,” “drastic,” “fully,” “absolutely,” “dysfunction” — then it was almost certainly human. If somebody expressed an emotion — “This is an exciting time,” or “I am proud” — or used exclamation marks or all-caps for emphasis, it was definitely human. On the other hand, if I thought the writing sounded bland, or if it just didn’t seem to be saying much at all, it was very likely to be flagged as likely AI-generated. The GPTZero team says their tool looks for several things when deciding if a text is likely AI-generated, including how similar the text is to AI patterns of writing. Human writing tends to vary in style and tone throughout, while AI’s doesn’t, GPTZero says. GPTZero also compares the text to what it predicts ChatGPT would have written. In the end, I came up with 20 candidates for K-12 school boards who had at least one answer that GPTZero flagged as 95% to 100% likely AI-generated. For some of these candidates, virtually all of their answers were flagged as 100% likely. I compared all the answers that GPTZero flagged as AI-likely, and I was surprised to see how many of them were nearly identical not just in tone, but in word choice and the ideas they proposed. For example, for my question about what is causing student absenteeism, several AI-likely answers claimed absenteeism is caused by a “variety” or “combination” of factors such as socioeconomic challenges, issues at home or disengagement from school. Then the answers proposed similar solutions, including “re-engaging” students at school, providing “early intervention” and advocating for increased access to counselors. For my question about how candidates would approach their district’s budget, many people said they would prioritize funding for things that “directly benefit” students like “teacher support” and “classroom resources” while working collaboratively with stakeholders. Of course, just because an answer was flagged as likely AI-generated doesn’t mean it was. So I reached out to those candidates who had had multiple answers flagged to ask whether they did use AI tools. The seven candidates who responded all said they did use AI tools, to at least some degree, to help write their answers. Most said they had used ChatGPT. Some said AI has already become a part of their daily life and work. AI is not the same as plagiarism, they said, and it’s not speaking for them; rather, it’s just one more technological tool to make life easier. “I don’t have someone to write my speech like Trump or Harris and I don’t have the money to hire someone to write my answers,” wrote Britni Mushet, a businesswoman running for the Julian Union High school board, in an email. “My answers are spoken from the heart. They are re-created in words that AI has come up with for the general audience to better understand.” Candidates said they didn’t just copy and paste what the AI tool produced. They said they used AI tools to help come up with wording or content, then edited the answers themselves before submitting them. Some candidates said they used AI to generalize their answers — to deliberately make them sound vanilla. Some said they did so to ensure they sounded nonpartisan. They wanted to word their answers carefully, particularly for the questions I had asked about politically polarizing topics, including whether schools should ban books or notify parents about their student’s gender identity or presentation. “Because of the sensitivity surrounding the topics, I wanted to be sure my answers were concise and did not have any room for misinterpretation,” said Sharmane Estolano, a real estate broker running for Chula Vista Elementary school board. “Vanilla is better in that instance, just because it’s so sensitive.” Some candidates didn’t know enough about the topic at hand to formulate a response, so they turned to AI for help. That was the case for Peter Kellner, a web training author running for Borrego Springs school board, who said he used the AI tool Perplexity to research answers for my question about what has been fueling student absenteeism. He said he prefers Perplexity because it links to sources of its information. Kellner said he visits those links, reads and learns more about the topic and vets those sources before deciding to include the information. It’s not drastically different from using Google, but it saves much more time, Kellner said. “I use AI as part of responding to those questions, but I didn’t use it in a way where, if I were an educated reader, I’d be upset with me for it. I used it as a research tool,” he told me. However, Kellner said he realized after speaking with me that Perplexity did end up heavily influencing his answers. For example, for the question on gender identity, Kellner’s submitted answer said schools should “approach these matters on a case-by-case basis,” and “it’s important that any conversations around sensitive topics, such as a student’s gender identity, are handled with care and respect for both the student’s privacy and the family’s values.” But when I read his own answer back to him, Kellner became unsettled: He realized that Perplexity had inserted a variation on the phrase “family values,” which he would not have chosen to do on his own because he considers it politically coded. “Now that you read it back to me, (Perplexity) probably influenced me too much,” Kellner told me. “That’s what I guess we all try to avoid, is to get influenced in an undue fashion by something you read.” Kellner said he wished he had written as resolute an answer as his opponent, Martha Deichler, had. Deichler, a retired teacher, had said simply: “No, I do not think that parents should be notified unless the student agrees to this.” “That’s exactly how I feel, but I feel like I can’t defend that the same way Martha can,” Kellner said. He said he hasn’t personally encountered the issue of gender identity with students before. “I just felt like, who am I to go out and make a really strong opinion when I don’t have the background or I don’t have the experience?” Some candidates said they wrote their own answers first, then used ChatGPT to condense them to fit the word limit I had prescribed of 150 words per answer. Jaime Chamberlin, who’s running for re-election to the San Marcos school board, said 150 words was too short to get into the nuances she felt answering the gender identity and book banning questions would require. So she asked ChatGPT to make her answers concise enough to fit the word count, and to generalize them for a “broader audience,” she said. Alison Emery, a Solana Beach elementary teacher running for the Carlsbad school board, also said she used ChatGPT to make her answers more concise to fit the word count. But she doesn’t think AI should play a significant role in generating a candidate’s answers or statements. “When (AI) starts taking away that voice and it just becomes a statement without any personality behind it, I think we lose something,” Emery said. “This is all extremely personal, and I care about what I’m doing. I don’t think you can convey those feelings and emotions in a genuine way through AI.”