How AI search changes the way voters find you

Try something before you read the rest of this post. Open ChatGPT, or Claude, or Perplexity, and ask it about a candidate in your area — any race, any level.

Read the answer carefully. Now ask yourself: where did that come from?

For twenty years, “being findable” meant one thing: ranking on Google. Campaigns learned the game, and the game was stable — ten blue links, and your job was to be the first one. That era isn’t over. But it’s no longer the whole picture, and the campaigns that notice first will have an edge the others don’t.

The new front door

When an AI assistant answers a question about a candidate, it doesn’t send the voter to a website. It synthesizes an answer from whatever sources it can read and trust: the candidate’s own site, news coverage, ballot databases, old social profiles.

Here’s the part that should get your attention: if your own site is thin, slow, or unreadable to crawlers, the answer gets built from everyone else’s version of you. The out-of-date news article. The opponent’s framing. The ballot database with the typo in your bio.

The single most important move a campaign can make right now is making sure the canonical answer lives on a site you control — and that machines can actually read it.

What AI systems actually read

The good news? Optimizing for AI search mostly means doing the fundamentals unusually well. There’s no dark art here.

• Crawl access. AI crawlers like GPTBot, ClaudeBot, and PerplexityBot respect robots.txt. Plenty of sites block them by accident through overzealous defaults — a one-line mistake that makes you invisible. Check yours today; it takes two minutes.
• Structured data. Schema.org markup tells machines, unambiguously, who you are, what the organization is, and what each page covers. It’s the difference between being parsed and being guessed at.
• An llms.txt file. A newer convention: a plain-text summary of your site written for language models — who you are, what matters, where the key pages live. Cheap to add, and you’re early if you do.
• Clean, semantic HTML. Headings that describe content, pages that answer one question well, and text that doesn’t need JavaScript to appear.
• Consistency. Your name, race, and key facts should read identically across your site, so a machine assembling an answer never has to resolve a contradiction. When machines find conflicting facts, they hedge — and a hedged answer about you is a weak one.

Notice what’s not on that list: tricks. Every item is something a well-built site should have anyway.

Is your site AI-ready? Five checks anyone can run

You don’t need a developer to find out where you stand. Fifteen minutes:

1. Ask the assistants about yourself. Pose the same question to ChatGPT, Claude, and Perplexity: “Who is [your name], running for [office]?” Note what’s wrong, what’s missing, and which sources they cite.
2. Check your robots.txt. Visit yoursite.com/robots.txt. If you see Disallow: / under any user-agent that looks AI-related (GPTBot, ClaudeBot, PerplexityBot), your site is invisible to that assistant on purpose, probably by accident.
3. Look for structured data. View your homepage source and search for application/ld+json. Nothing there means machines are guessing at who you are.
4. Turn off JavaScript and reload. If your issues page goes blank, AI crawlers see the same blank.
5. Search your name plus the office. The results Google shows are roughly the source pool an AI answer draws from. If your own site isn’t the obvious top source, the answer gets built from whoever is.

Any check that fails is fixable in days, not months, and each one compounds with the rest.

Speed and trust still compound

AI systems favor sources that look authoritative: sites that load fast, use HTTPS, publish real information, and get referenced elsewhere. The same craftsmanship that wins over a human visitor is what earns a citation in an AI answer.

We find that reassuring, honestly. The work doesn’t change. The stakes did.

What to do this cycle

You don’t need an “AI strategy,” and you should be suspicious of anyone selling you one. You need a website built to modern standards — structured data, open crawl access, and content that answers voters’ actual questions in plain language. It’s part of the baseline in every campaign website we build, and it belongs on your pre-launch checklist either way.

Go back to that answer you got at the top of this post. Campaigns that do this work now will be the ones AI assistants quote accurately when a voter asks. In a close race, being described correctly is worth real votes — and unlike most things in a campaign, this one is entirely within your control.

← All insightsView as Markdown