The AI Texting Debate Reached NPR. Here's Where We Stand.

The AI Texting Debate Reached NPR. Here's Where We Stand.

NPR just brought the AI texting debate to a national audience. The July 2026 piece, Political text messages could get more effective and annoying. Blame AI, describes a wave of platforms selling autonomous AI agents that converse with voters over text. The playbook it reports is consistent: a human writes and sends the first message, and when the voter replies, the bot takes over.

The numbers in the piece explain the appeal. Vendors describe bots that respond within 30 seconds in any language. One firm reports millions of texts producing tens of thousands of AI conversations, with some voters chatting with an agent for hours. When 5 to 10 percent of recipients reply and a fifth of those go ten messages deep, no volunteer program can match that responsiveness at scale.

We build AI for political texting, and we think the technology is genuinely valuable. We also think the critics quoted in that piece are right.

The concerns are legitimate

Three objections run through the article, and none of them is hand-wringing.

Voters deserve to know who is talking. When software carries a conversation and nothing says so, the voter is having a relationship with a machine under the impression it is a person. Some vendors' own clients decline to advertise that they use AI, which says plenty about how they expect voters would react.

Language models get things wrong. A bot speaking in a campaign's name can state false information with total confidence, or be goaded into saying something no campaign would ever approve. In most industries a bad chatbot answer is a support ticket. In an election it is a news cycle, and possibly a legal problem.

Accountability gets fuzzy. When no human approves each message, who answers for the one that goes wrong? "The model did it" is not an answer a candidate can give a reporter, a regulator, or a voter.

These are not reasons to abandon AI in voter contact. They are requirements for how it has to be built.

The industry drew this line in 2023

None of this debate is actually new. In May 2023, the board of the American Association of Political Consultants, the profession's ethics body, unanimously condemned the use of deceptive generative AI content in political campaigns, ruling it contrary to the AAPC Code of Ethics. The statement went further than most people remember: it declared that issuing a warning or disclaimer would not be accepted as a way around the standard.

That last part matters. The AAPC's position is that a fine-print label does not launder deception. If the practice misleads, disclosing it does not make it honest. The line the industry drew is not about which tools you use. It is about whether a real person stands behind what the voter receives.

The AAPC's own membership shows what that looks like in practice. Its foundation's 2025 survey of political consultants found that 86 percent have used AI and a majority use it weekly. But look at what they use it for: drafting internal materials, brainstorming, and writing copy that a person then owns and sends. Only 2 percent reported using AI to personalize voter-facing experiences. The number one challenge they reported, at 62 percent, was the accuracy and reliability of AI output, and their top fears were deep fakes and misinformation reaching voters.

Read those two documents together and the profession's position is clear. AI as a drafting and efficiency tool with human-owned output is the mainstream. Autonomous, voter-facing AI is the frontier the industry itself does not trust yet. The survey also confirms the partisan adoption gap NPR reported, with 44 percent of Republican consultants using AI daily against 28 percent of Democrats, but the accuracy worry is bipartisan.

Our answer: AI drafts, humans send

This is the model we built into Political Comms from the start, and the NPR piece is a good occasion to restate it plainly.

Lincoln, our AI agent, drafts a reply to every inbound voter message within seconds. It builds that draft from the conversation itself and from campaign-approved facts only: poll hours, event details, links someone vetted. If the approved context does not contain an answer, the draft does not contain one either, which removes the raw material hallucination is made of.

Then the draft stops. It sits in a review queue until a staffer approves it, edits it, or rejects it. There is no auto-send mode to toggle on. A draft nobody reviews sends to nobody, and every message that does go out traces back to a person who chose to send it. We wrote up the full workflow in AI Drafts, Humans Send.

Some will frame that as slower than a bot that fires back in 30 seconds, and in a literal sense it is: the reply waits for a human decision. We think that pause is the entire point. The draft is ready in seconds, so staff never face a blank box, and a program that once needed a day to clear its reply queue clears it in an afternoon. What the pause buys is the thing the critics in the NPR piece are asking for: a person who read the message, approved the words, and is accountable for them. Speed you get from removing the human is speed you get by removing the accountability, and that trade is not worth it in political speech.

The law is heading the same way

The regulatory direction is not subtle. As the NPR piece notes, North Dakota and California already require senders to disclose in the first message when a recipient is talking to a virtual assistant. New Jersey is weighing disclosure requirements for generative AI used to provide election-related information. More states will follow, because political texting already sits under active scrutiny and autonomous bots invite more of it.

Disclosure rules are a floor, not the finish line. The AAPC's survey found that consultants judge disclosure as something that can make a questionable AI use less unethical, not something that makes it ethical. We agree, and the AAPC's 2023 statement says the same thing about disclaimers. The durable answer is structural: keep a human in the loop so that accountability exists whether or not a given state has written it into statute yet. A workflow where a person approves every message does not need a carve-out from bot-disclosure laws, because there is no unsupervised bot to disclose. Norms and statutes in this area are still developing, and we welcome clearer standards.

Speed and responsibility are not a trade-off

The lesson we take from the NPR piece is not that AI in political texting is dangerous, and not that it is inevitable. It is that the design choice is the whole game. Built one way, AI is an unsupervised agent conducting political conversations nobody reviews. Built the other way, it is the fastest drafting tool a texting program has ever had, with a person owning every word that reaches a voter.

We chose the second way on purpose, before the national press started asking about the first. Campaigns should not have to pick between answering every voter quickly and standing behind every message. With the right architecture, they never do.

See how the workflow runs end to end on the Lincoln page, or talk to us about putting it to work on your next campaign.