The Lab Results: AEO

Why this matters

Voters are starting to ask AI who's on their ballot instead of searching for it. About 49% of all American adults use AI chatbots in some capacity (Pew Research), and 51% of Gen Z already use it daily or weekly (Gallup). When an AI answer leaves a candidate out, or gets them wrong, most voters never see a correction.

Down-ballot candidates are the most exposed: they have the thinnest footprint online, so AI has the least to draw from. We wanted to know whether a campaign can change that with content it controls.

What is aeo?

Answer engine optimization, the next step after SEO

AEO IS

Clear, well-sourced facts that AI can trust

AEO IS NOT

Gaming the model, keyword stuffing, or spam

The test

Eight Pennsylvania campaigns, two content formats, three AI engines

OUR HYPOTHESIS

Publishing structured, question-driven content will get candidates mentioned and cited more often in AI answers.

WHAT WE TESTED

Each campaign published one format first, then switched to the other, so we could separate the effect of the format from the effect of the candidate. Every district got its own bank of prompts built from questions its voters actually ask.

We read results weekly by topic and by engine, against a May baseline.

Format 1 - FAQ

Specific questions with direct answers, like "What is the candidate's position on affordable housing?"

Format 2 - Article

A long-form piece on one topic, like "The candidate's plan to make housing more affordable."

Measurement

Share of voice: how often AI mentions the candidate at all
‍Citation share: how often AI cites the pages we published

RACES

4 congressional, 4 state Senate

STATE

Pennsylvania

TIMING

May - Aug '26

ENGINES

ChatGPT, Gemini, Perplexity

What we found

Every candidate gained, and A few topics did the most work

Every candidate gained, but not equally

State candidates grew their share of voice nearly twice as much as federal candidates because they started with lower AI visibility and published directly on their official sites.

Race discovery pages were cited most

"Who is running for Congress in my district?" beat broad policy questions two to one.

Key learnings

Format isn't the unlock. Topic is. Articles and FAQs both won, but results varied widely within each format. The strongest pages answered specific, high-intent questions.
AEO works up and down the ballot, working hardest down-ballot. Every federal and state candidate gained. But local candidates struggle to earn media next to higher-profile races, so AI has little to draw on. Publishing their own content filled that gap, and state candidates grew their share of voice nearly twice as fast as federal.
AI engines are getting pickier, but are favoring owned content more. ChatGPT now cites about 72% fewer sources per answer, favoring owned, authoritative pages. Total AI citations fell 21% in August while citations to our content rose 15%.

What we recommend

Publish on the official campaign site
AI engines increasingly favor owned, authoritative sources. Your campaign domain is the one you control.

Start with the basics voters ask
Who's running, who the candidate is, and how they compare to the opponent earned the most citations.

Optimize for the question, not the template
Pick the topic first, then choose an article or FAQ based on what best answers it.

Track each engine separately
ChatGPT, Gemini, and Perplexity are moving in different directions. What works on one may not hold on another.

Results through August 31, 2026. Data from The Prompting Company.

AI ENGine Optimization for Candidates