Stop Panicking Over AI Text on Candidate Websites Because Human Copy Was Already Terrible

Stop Panicking Over AI Text on Candidate Websites Because Human Copy Was Already Terrible

Hand-wringing over AI-generated text on political campaign websites is the easiest, laziest take in modern political journalism.

Every election cycle, media analysts run a few thousand congressional landing pages through notoriously unreliable AI detectors, clutch their pearls when the software flags synthetic syntax, and publish breathless warnings about the death of authentic democracy. They frame the discovery as a crisis of voter deception. They imply that prior to large language models, local candidates were sitting at their kitchen tables late at night, pouring their raw, unfiltered souls onto Squarespace templates.

That premise is a complete fantasy.

Having spent the better part of a decade auditing digital infrastructure and messaging pipelines for down-ballot campaigns, I can tell you what actually existed before ChatGPT arrived. It was not bespoke, heartfelt prose. It was a graveyard of copy-pasted consultant boilerplate, sterile talking points lifted directly from national party committees, and lifeless platitudes recycled from 2014. The arrival of generative AI did not destroy authenticity in political web copy; it simply automated a preexisting factory of mediocrity.

The Myth of Pre-AI Political Authenticity

To understand why the panic over synthetic campaign text misses the mark, you have to look at how political websites actually get built. Most Americans imagine a candidate carefully drafting their "Issues" page to reflect their unique worldview. In reality, unless a candidate is running a top-tier Senate or presidential operation with a seven-figure communications budget, their website copy is an afterthought.

Historically, a cash-strapped state legislative or House candidate had two choices for their website text:

  1. The Consultant Template: Hire a mid-tier digital firm that charges five thousand dollars to spin up a WordPress site, swap out the district name, change the color palette from red to blue (or vice versa), and paste the exact same paragraphs about "fighting for hardworking families" and "bringing common-sense leadership to Washington" that they used for forty other clients.
  2. The Party Playbook: Download a messaging memo from the DCCC, NRCC, or state party apparatus and transcribe the approved bullet points verbatim onto a webpage so they do not accidentally step on a third rail during a primary.

Neither of these methods produced human authenticity. They produced defensive, focus-grouped mush designed to offend nobody and inspire even fewer. When media outlets run scraping tools across campaign domains today and announce that half of them contain AI-written text, they are measuring a change in production cost, not a drop in editorial quality. A junior staffer prompting a language model to "write three paragraphs about lowering prescription drug costs in a moderate tone" is producing the exact same linguistic output as a $300-an-hour political strategist did five years ago—only now it takes twelve seconds instead of three weeks.

Why AI Detection Studies Are Fundamentally Flawed

Before accepting the premise that candidate websites are "filled" with AI text, we need to address the elephant in the room: the detection tools themselves.

Journalistic analyses of AI prevalence almost universally rely on commercial classifiers like GPTZero, Originality.ai, or Copyleaks. These tools do not possess a secret watermark reader. They operate on statistical probability, primarily measuring two metrics: perplexity (how predictable the next word in a sequence is) and burstiness (the variation in sentence length and structure).

Here is the fatal flaw when you apply those metrics to political writing: politicians are trained to write with low perplexity and low burstiness.

Good political communication is repetitive, predictable, and rhythmically uniform. Speechwriters and press secretaries actively strip out eccentric vocabulary, complex subordinate clauses, and unexpected metaphors because those elements create ambiguity—and ambiguity gets candidates killed in 30-second attack ads. When a human press secretary writes a standard campaign bio or an economic policy plank, they naturally gravitate toward the highest-probability phrasing in the English language.

Consequently, AI detectors routinely flag human-written political copy as synthetic. Even OpenAI shut down its own AI classifier in 2023 due to abysmally low accuracy rates, and researchers at institutions like the University of Maryland have demonstrated that simple paraphrasing or rigid stylistic constraints easily break detection models. When an analysis claims that a specific percentage of candidate websites use AI, you are looking at a rough approximation polluted by massive false-positive rates inherent to the political genre itself.

What Voters Actually Do on Campaign Websites

The panic over AI website copy assumes a voter behavior model that simply does not exist in modern politics. Critics argue that voters visit candidate sites to evaluate character, inspect policy nuance, and form an emotional bond with the prose.

Look at the analytics of almost any congressional or local campaign website, and a wildly different picture emerges.

  • Time on page for "Issues" sections averages under 25 seconds. Voters do not read 800-word essays on agricultural subsidies or municipal zoning reform. They scan bolded headers to confirm the candidate holds the baseline partisan position they expect, and then they leave.
  • Over 80% of traffic lands on the homepage or a direct donation funnel. People visit candidate websites for three primary reasons: to verify the candidate exists after seeing a yard sign or TV ad, to find out where and when to vote, or to give money via ActBlue or WinRed.
  • Persuasion happens elsewhere. Nobody is converted from an undecided voter to a passionate supporter by a paragraph of text on a static bio page. Persuasion happens through short-form video, local news coverage, peer-to-peer text messaging, and face-to-face canvassing.

Because website copy is functionally a digital brochure—a hygiene factor rather than a persuasion engine—spending scarce campaign resources on artisanal, hand-crafted web prose is terrible resource allocation. If a first-time school board candidate or an underfunded challenger uses a language model to draft a clean, typo-free platform page in an afternoon so they can spend the rest of the week knocking on 400 doors, that candidate is making the correct strategic decision.

The Real Threat Is Not AI Copy—It Is Strategic Laziness

Does this mean campaigns should blindly copy-paste raw ChatGPT outputs onto their domains? Absolutely not. The problem with unedited AI text on campaign websites is not moral or ethical; it is tactical.

When campaigns rely on default model outputs without heavy human steering, three predictable failures occur:

1. The Sea of Sameness

If every candidate in a twelve-person primary feeds the same basic biographical facts into a default LLM prompt, their websites become indistinguishable. Voters comparing candidates in a local race will read identical cadences, identical transition phrases, and identical hollow promises. In a crowded field, sounding like everyone else is a death sentence.

2. Hallucinated Policy Commitments

Language models are notoriously bad at understanding the hyper-specific jurisdictional boundaries of different political offices. I have seen local city council candidates publish AI-generated platform pages promising to "reform the federal tax code" and "secure the southern border"—issues over which a municipal council member has zero authority. Worse, an unedited prompt can easily generate a subtle policy stance on healthcare or energy that contradicts the candidate's actual voting record, handing opposition researchers an effortless "flip-flop" narrative.

3. Tone Deafness to Local Context

A model trained on the broad internet does not know that the local factory closed six months ago, that the county has a specific water-rights dispute, or that a particular phrase carries baggage in a rural versus suburban precinct. Generic AI text reads like it was written by an outsider—because it was.

How Smart Campaigns Actually Use AI (And What You Should Look For)

If you want to know whether a campaign is sophisticated or sloppy, stop running their "About Me" page through an AI detector. Look at how they use artificial intelligence behind the scenes to solve real operational bottlenecks.

The sharpest campaigns treat language models as research assistants and formatting engines, not ghostwriters. Instead of asking a model to invent their message from scratch, they feed the model raw transcripts of the candidate speaking naturally at town halls, rotary club meetings, or podcast interviews. They instruct the system to extract the candidate's actual cadence, real anecdotes, and specific vocabulary, and organize those raw thoughts into structured web copy.

Imagine a scenario where a candidate gives an off-the-cuff, ten-minute answer at a local diner about why small businesses are struggling with commercial permitting. A competent digital director records that audio, transcribes it, and uses an AI workflow to clean up the filler words and structure it into a readable policy brief. The resulting text is technically "AI-assisted," and an automated detector might even flag it. Yet it is infinitely more authentic, specific, and true to the candidate's voice than anything a Washington consultant would have written from scratch.

The Uncomfortable Truth About Political Communication

There is an obvious downside to normalizing AI-assisted copy in politics: it lowers the barrier to entry for phantom candidates and bad-faith operators. When anyone can spin up a convincing-looking campaign website, complete with comprehensive policy positions and a polished biography, in twenty minutes, voters and local journalists have to work harder to verify whether a candidate is a serious contender or a digital mirage.

That is a real governance and media-literacy challenge. It requires local reporters to pick up the phone, attend live forums, and test whether the person behind the website actually knows what their own platform says.

What it does not require is pretending that the pre-AI era of campaign websites was a golden age of literary sincerity. Voters are not being robbed of authentic prose by algorithms. They are simply seeing the exact same sterile political boilerplate they have ignored for decades—just generated at a fraction of the price.

Stop measuring whether a candidate used AI to write their website, and start asking whether they actually understand the words on the page when the microphones turn on.

EC

Emily Collins

An enthusiastic storyteller, Emily Collins captures the human element behind every headline, giving voice to perspectives often overlooked by mainstream media.