Who’s Texting You?

The next time you get a political text from a “Sarah” or a “John,” there may be no human on the other end. TIME and The Wall Street Journal reported last week on the multimillion-dollar campaign to rescue Israel’s reputation in the United States, built on AI-powered texting and a network of pro-Israel websites designed to shape what chatbots say. Millions of Americans have gotten texts from senders with names like Emma and Matt who say they are with a group called Friends for Peace. No such group officially exists. The texts open with a question, something like how peace talks with Iran might affect global security, and they are built to start conversations rather than push a position. One recipient noticed that everything she wrote came back rephrased in the next text. “I said to him, ‘I don’t think you’re a real person,’ and he tried to persuade me to think that he was.”

The texting runs through a firm called Sparkfire, which had collected $6.5 million by mid-May as a subcontractor to Clock Tower X, the company Brad Parscale registered under FARA to run the broader operation for Israel’s foreign ministry at $1.5 million a month. By the Journal’s accounting, the contract has grown past $45 million. The same operation built pro-Israel websites structured so chatbots would cite them; the Journal names Allyvia.org as one that ChatGPT and Claude have picked up. Inside the Israeli government, the verdict is rough. An official told TIME: “We are pissed at Brad Parscale. We have paid him lots of money. But what did he do with it? Things have only gotten worse.”

AI texting and creative attempts to shape chatbot responses are becoming more common in the run-up to the midterms.

The Signals This Week

1. AI-Powered Texting Startups Face-Off

Five vendors are running AI text conversations for the midterms, and one says it has held 30,000 of them this year.

AI-powered campaign texting is becoming more common ahead of the midterms, NPR reports. Platforms including Akillion, Convos, Vector Political, Peerly, and Scale to Win hold personalized back-and-forth conversations, simulate a candidate answering policy questions, and log what each voter says to shape the next message. Vector Political says it has sent 2.5 million texts this year and held 20,000 to 30,000 conversations, with response rates of 5 to 10 percent on first texts and 10 to 20 percent once a thread gets going. Convos worked 10 campaigns last cycle and is aiming for more than 100 this one. Replies come back inside 30 seconds, in any language.

The Signal: Campaigns have collected voter reactions for years through automated polls, robo-messages, and volunteer texting. Now that the cost of holding the conversations has plummeted, it remains to be seen whether or not AI results in better outcomes.

Source: NPR via WWNO, KPBS

2. Claude Cited a Russian Propaganda Network Seven Times

A Berlin think tank audited 4,811 chatbot citations and found a narrow news diet with Russian propaganda inside it.

The Berlin think tank Agora Digitale Transformation ran 675 news questions through the major chatbots and reviewed 4,811 source references. Across the models, the news diet was narrow: 46 percent of the 544 cited domains appeared only once, and ChatGPT pulled 34 percent of its references from a single Axel Springer title tied to an OpenAI licensing deal. Claude cited the Russian “Pravda” disinformation network seven times, unmarked, alongside legitimate journalism. The contamination, the authors argue, is a democratic problem rather than a technical glitch.

The Signal: A licensing deal put one publisher behind a third of ChatGPT’s news citations, and a propaganda network got in without a label. Nobody is auditing these lists.

Source: Agora Digitale Transformation (Jul 8), Noah News (Jul 14)

3. Quorum Built the Grassroots Campaign Into a Button

Quorum’s new agent turns a pasted press release into a launch-ready advocacy campaign, filling more than fifty setup fields on its own.

Quorum launched the Campaign Builder Agent inside Quorum Grassroots. Given a URL or a pasted press release, it produces a complete, ready-to-launch “Write a Letter” advocacy campaign, filling what the company says is more than fifty fields of manual setup, in minutes. It went live for Quorum Grassroots customers on July 15.

The Signal: When the build takes minutes, the constraint on grassroots work is having something worth saying while the policy window is still open.

Source: Quorum (Jul 14), EIN Presswire (Jul 15)

4. The Bill Tracker Raised $70 Million

State Affairs pitches itself as a Bloomberg Terminal for policy, and Founders Fund just paid $70 million for the attempt.

State Affairs raised $70 million from Founders Fund, Khosla Ventures, Tru Arrow Partners, and Alumni Ventures, along with KKR chairman Richard Sarnoff and former Washington Post editor Marcus Brauchli, to expand an AI platform that tracks and analyzes legislation and regulation across all fifty states and the federal government. The pitch, per CEO Evan Burns, is closing the gap between when a policy move happens and when the organizations it affects find out.

The Signal: Monitoring is becoming a commodity available to any firm that pays for it. Competitive differentiation remains in what happens after the alert, offered to whoever decides fastest or pulls the right relationships.

Source: PR Newswire (Jul 14), Refresh Miami (Jul 15)

5. Loomer Called a Real Photo AI, and Grok Agreed

The senator’s hospital photo was genuine, and proving that took a forensics professor and a sports section.

After Mitch McConnell was found briefly unconscious and treated for pneumonia, his office released a photo of the senator with his wife Elaine Chao. Laura Loomer declared it AI-generated, and X’s Grok backed her, asserting the image had been recycled from 2023. The Washington Post obtained the original file, a digital-forensics professor found no manipulation markers, and the newspaper in McConnell’s hand matched that Sunday’s Washington Post sports section. Poynter rated the claim False. In the AAPC’s March member survey, deepfakes that mislead voters ranked as consultants’ top AI concern for the second straight year, with 63 percent calling them a top-three worry.

The Signal: The liar’s dividend pays off because a lie moves faster than the truth, especially with a chatbot spreading it.

Source: Poynter (Jul 14), AAPC 2026 Member Survey

6. TikTok Has Labeled 3 Billion AI Videos

Three billion labels are live, and research says the small ones do not change what viewers believe.

TikTok announced it has labeled more than 3 billion AI-generated videos through Content Credentials, creator disclosure tools, and invisible watermarking, and launched an in-app AI literacy hub built with the National Association for Media Literacy Education and deepfake researcher Henry Ajder. Research cited alongside the rollout found that small overlay labels do not measurably change belief or sharing behavior; only full-screen interstitial warnings do.

The Signal: AI labels are now mandatory on many platforms, but the small versions tend not to register with viewers. That calls into question the disclosure laws that treat the label as the remedy.

Source: TikTok Newsroom (Jul 10)

7. Puerto Rico Hired the AI Song

A Suno track about the island went viral, and the official tourism board adopted it as the campaign.

Bill Stiteler, a Minnesota musician posting as Saxboy Billy, wrote lyrics and generated a song about Puerto Rico with Suno. It spread across TikTok with lip-syncs from Charlie Puth, Sarah Hyland, Luke Combs, and Jennifer Love Hewitt. Discover Puerto Rico, the island’s official tourism board, flew him down and released an official music video on July 8. CMO Storm Tussey: “By bringing Bill to the Island, we wanted to celebrate that authenticity while showcasing that Puerto Rico is so much more than a single destination.”

The Signal: Time and again, the AI content that resonates with the broadest audiences is the silly and the satirical. The song is fun and a bit campy, which is what made audiences receptive to it.

Source: Yahoo Entertainment (Jul 9), Euronews (Jul 9)

Thanks for reading. If The Influence Model is helpful, forward it to one person who should be reading it.

Best,

Ben